{ "measurement": { "model.layers.0.self_attn": [ { "accuracy": 0.9604678715726263, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9657230566402799, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9705055469745084, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9773159262871272, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9787535633597719, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.980010548410447, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9801279525027463, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9814336977310871, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9816533602578076, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.982049373860814, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9885926864502069, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914344004355371, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9889673247051082, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920597183763197, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927265490895432, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953046060338813, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929503300104683, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9973470130445141, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979490270580802, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.0.block_sparse_moe": [ { "accuracy": 0.9716189280152321, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9751367906206533, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9764581213362122, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9769711665024883, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942966188108059, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952248084836787, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957338326892472, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976378331745142, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979613301203292, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978858479421193, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981540889589509, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987138034745218, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987564638523556, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989940639689417, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991867736174945, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992819593267326, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992713859782701, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.1.self_attn": [ { "accuracy": 0.9860200671062461, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9879969402191866, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9892334457621664, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930943124345504, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.993483101207714, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9938606557338253, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956090290767239, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959206207024277, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9960213983125102, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9962125073009367, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996875159921872, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970705707842411, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971728471833845, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974058643060025, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984071974138619, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986688012983282, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988263255256339, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992570764979495, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999629968071914, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.1.block_sparse_moe": [ { "accuracy": 0.9610651466799409, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9645701881105962, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9681949491955733, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9681979793270952, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994622902742489, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967106307679052, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967325632092788, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984842733522926, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983200508378152, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984384583872988, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984559684127913, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987227654997505, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991901297172788, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991180705470278, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992085966085525, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992109895331516, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994416544520247, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.2.self_attn": [ { "accuracy": 0.998651913777729, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987687155953608, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989262398387502, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992856187853153, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993220838476095, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993700629688407, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995165385832814, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995718064141998, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996330828871578, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996459251177792, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996748601513804, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997019492875552, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997123367755097, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997389455644512, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998248989807172, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998683417755896, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998610657639802, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999289374253223, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999582988602924, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.2.block_sparse_moe": [ { "accuracy": 0.9979133659446808, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9979915965691601, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9982108271705281, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982603771526268, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990763307892179, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991438215273727, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992191153501871, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995291351591923, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995660546318457, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995293291193728, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999594250381434, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997596481430504, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997764832239083, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998690759564756, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9998780585167544, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998922311305681, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999626243273891, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.3.self_attn": [ { "accuracy": 0.997984172451604, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981423380270013, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984285517375132, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988690608236204, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990418364851441, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999079827504159, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993148810830653, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993733723439571, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994281972528705, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999453127822275, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995317906125963, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995729544990402, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995841786072369, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996317345878771, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997612200554853, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998113880306176, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998152482439764, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999896758877807, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999467114891893, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.3.block_sparse_moe": [ { "accuracy": 0.9947472171563851, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9950363243268313, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9951767348561829, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9952458721021876, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973030627496835, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964397866156345, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965144441414037, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978765365973997, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99660999843189, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987347210806451, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981931471204581, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998704133353098, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985553939109914, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987554171116484, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9994733191246938, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994889280443251, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995258002217222, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.4.self_attn": [ { "accuracy": 0.9959456188800303, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974410141646666, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974309124885813, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979359736470016, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980251166148504, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983449301952007, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990751416036417, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999150771736535, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991863468067573, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992102080907085, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992518353423006, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993063318512501, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999396969159869, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994673193983832, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996642412795489, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997296551813825, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997494497174079, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998584914160586, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999292734273216, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.4.block_sparse_moe": [ { "accuracy": 0.9967083106503675, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9968185609598693, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9971443980834201, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9972163047518974, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984856568442314, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998598636563034, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987028746920834, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.999222594135637, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.999273825761568, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992429250575553, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993342473263558, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996096116083208, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996652102762662, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997811124002941, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9997949240802738, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998131419507492, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999278476578567, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.5.self_attn": [ { "accuracy": 0.9966417127533963, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969604835036749, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974119154197213, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980525475448782, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983252416438374, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983550902995232, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988790081084804, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998928896549746, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990553335795228, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990915563233246, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991383934270983, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992133827824881, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993072216166183, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993761779807852, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996111471708374, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996824880960814, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997066683365694, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998347077581504, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999134470367848, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.5.block_sparse_moe": [ { "accuracy": 0.996166872044437, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.996296133870553, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9966930859158502, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996782133570186, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982032733356678, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983520268793463, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984781505190767, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990971916061091, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991696222844583, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990978990650834, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999211534508504, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995429230529178, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999606895824208, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997546809907746, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9997641563914887, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997861986944612, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999327645014809, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.6.self_attn": [ { "accuracy": 0.9966165939051854, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968962400882063, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973239372852013, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978692371361447, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983084896260774, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983188592167082, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989407767057664, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989691918568784, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990437361227936, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990751950198932, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991335200397052, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991955172797469, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993126601942717, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993892107097628, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996151958674953, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996809794607352, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997068093461597, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998263281633722, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999914756200987, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.6.block_sparse_moe": [ { "accuracy": 0.9950301658337641, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9952060359799745, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9957517970255331, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9958713636243421, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976603323961363, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978289609885236, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997998975450173, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9988286860814122, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9988417052633227, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987544041206619, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988967870288578, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992676211404614, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993758034190825, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994646806888724, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995950706736614, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999620060126242, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997444948098028, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.7.self_attn": [ { "accuracy": 0.9961331902488478, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965422626731819, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970068706224036, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977304411647645, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981576806779853, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981958675688427, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987423858286715, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998785852166628, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988884310434131, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989346932822005, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990789310173377, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991359047738737, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992068539569644, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992756394203752, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995587850632881, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996259709418824, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996640298218393, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997969876192081, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998982558249018, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.7.block_sparse_moe": [ { "accuracy": 0.9938900313634229, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9940987808237735, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9948019088049861, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994958788487374, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971453861002565, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973733360866869, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997601928550897, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985702913377041, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9986836189667596, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985659595940409, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987452838655659, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992690340242054, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993733793395988, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996058378877167, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9996264391811565, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996682774370512, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998960448317934, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.8.self_attn": [ { "accuracy": 0.9945238184575972, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9950983128510416, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957348073489571, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969851379282773, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997417096122119, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974755792339381, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983714226951921, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998460725917922, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9985645178301064, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986302645472613, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986937305105752, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998790092919463, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988897350298143, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989762216604217, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993841977312082, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994782174211363, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995699829581243, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997180798787955, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998759646320985, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.8.block_sparse_moe": [ { "accuracy": 0.9931356913263076, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9933657516284209, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9941447816887184, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943197734751984, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968386379585258, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970906801570795, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973567113292551, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983177657888614, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984930658614949, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983835084498615, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985903818742372, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991492469715396, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999263087247719, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994799856840951, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995093581520976, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995517203941219, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997351641738389, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.9.self_attn": [ { "accuracy": 0.9945329436787257, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947987137803513, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995390102402062, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966217198678734, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971437411333778, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971454330646482, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981995610961396, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982146018576857, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984620173253413, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9985340551509964, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998579728490624, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986676872573107, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988148495934805, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988948709000588, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993183133222996, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994483351857573, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995152005683753, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997061126372213, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998600539555283, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.9.block_sparse_moe": [ { "accuracy": 0.9922447077145702, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9925072627436173, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9934222724925923, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936237271483007, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9963893709018042, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966784716265178, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969788131941306, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981824341658992, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983257938323444, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981842744041627, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984160290455053, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990799738811084, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992099212678639, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999506352916605, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995282766007891, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995841458409191, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999865847510687, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.10.self_attn": [ { "accuracy": 0.9937942865990886, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994166461799882, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948115459349203, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961575811336699, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968601987746201, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969523789630713, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979658187183208, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9980500145190346, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982145906205436, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982978073071296, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984169633703699, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998531070629772, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986382036800742, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987669804862602, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992489715958456, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993807600464941, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994535720217851, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996543732689277, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998461416551819, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.10.block_sparse_moe": [ { "accuracy": 0.9917685196146762, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9920450206845999, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.993054480291903, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932778665403786, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9961287274245957, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964473978738839, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996776813310326, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980223769518105, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981973098443919, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980774245800843, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983107187221513, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990249154200826, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991610043645395, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994794261231283, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9994976820966449, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995588732402355, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998559856103384, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.11.self_attn": [ { "accuracy": 0.9930735166723791, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936613659129331, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943089815787971, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957768015661522, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966670068127936, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967551950340796, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978627280491453, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979669166741109, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981118389098954, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981996287640772, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9983383804486182, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984419318957647, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985337103104317, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986563182779049, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991861950226226, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993154630922761, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994382995723036, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996200124334887, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998366829868724, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.11.block_sparse_moe": [ { "accuracy": 0.9905802125944511, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9908892939259347, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9920527745822543, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923119305780059, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9955626085008445, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959158052837378, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962904749095047, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99778161429506, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979599685297886, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977725868137848, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980494008203479, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988710124529653, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990310485836266, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994003159242167, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9994202399179094, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999488895273412, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99983981563051, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.12.self_attn": [ { "accuracy": 0.9910701899054019, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918534795165455, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927212578783694, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946410112622145, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957123278689227, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957653053693081, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972752855085817, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9973444345799324, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9975357768616002, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9976823880508738, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9978437561443761, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979851232216644, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981431195659465, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982991210662907, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989686389086082, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991389698851364, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999281768913773, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995140006266363, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997960757617349, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.12.block_sparse_moe": [ { "accuracy": 0.9891133168222088, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9894642160323105, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9908627763537592, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911803973005399, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994865857616165, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952726115006953, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957319178226355, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974303930495424, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976382735552952, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974228993094968, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997743230827741, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986938334554188, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988792257316067, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999306236335542, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9993297882641567, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994147370299814, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998148162366737, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.13.self_attn": [ { "accuracy": 0.9905862934201172, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912386606447399, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920446855065069, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942712528680108, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953165997530481, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995377159934785, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971407358044464, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9971984471807158, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9974322112791828, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9975576291755331, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9976416402986568, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978147646118152, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979744736705661, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998136778594926, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988683723832333, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990574352008192, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992510315251389, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994796265703993, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997932125506371, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.13.block_sparse_moe": [ { "accuracy": 0.9878638142131662, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9882745234491793, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9898179239642463, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901577050209438, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943389103159701, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9947900400546036, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952902737882381, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971644443239233, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973941937848729, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971576706922957, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975129377072383, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985576778601267, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987639722867722, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992316498690726, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992588510412086, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993508065021351, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997941728636904, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.14.self_attn": [ { "accuracy": 0.9889371264235753, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9892763468240829, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9904682142660022, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929661692708338, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943249166550997, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944809407583977, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966794644776535, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9967916894238442, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9970235233317668, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9971815807940928, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9971765437482023, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973637253095052, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976101567061912, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977993494892297, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986687957548144, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988987537584954, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991245881017101, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994046403138016, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997583389662108, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.14.block_sparse_moe": [ { "accuracy": 0.9865074089581245, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9869653015936676, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9886966968739503, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890841200065456, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936722535394916, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9941753862229618, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9947404524948644, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968212737554782, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9970780993738261, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968177529149934, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972158635010648, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983839250635356, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986157265038377, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991376481729707, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991695921849705, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992742750481213, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997690688676237, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.15.self_attn": [ { "accuracy": 0.9878840282265293, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9882607105139055, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893990615627876, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919839164634284, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9938726558251992, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9939377204711107, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964060776033684, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9964861185412461, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9967625421275826, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9969809060107524, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9969612795761541, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971908494021351, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974032330486041, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976231404151278, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985515281288443, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988049123264653, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990648372729897, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993436717746877, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997387731357441, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.15.block_sparse_moe": [ { "accuracy": 0.9852244614280368, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9857342300544444, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9876404004661661, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9880595046928838, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9930974560005492, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9936446872735886, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9942600565080187, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965335931628942, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968124486285409, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965332125215546, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969656551869488, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982410555484852, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984927181779456, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990613960330129, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990963718982561, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992083903236658, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997492915143468, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.16.self_attn": [ { "accuracy": 0.986430655233562, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9869568624876832, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9881813355574482, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9908890081324467, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931634706445038, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9932708335109055, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959758004467738, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996066026095497, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996374053010521, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965505940521038, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996564900122681, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968183951109255, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970563212961057, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973260947552166, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983961371901003, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986498271096743, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989676389806462, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999259277934662, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997104863838391, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.16.block_sparse_moe": [ { "accuracy": 0.9837781063918221, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9843497034652453, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9864895911397118, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9869576792575812, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.992427188437432, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9930297064741975, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9937110840048837, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9961932678094232, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965016815296718, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9961957377013996, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996671232688976, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980722369908012, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983436732896065, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989693835285787, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990099352343302, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991327066918051, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997240065184949, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.17.self_attn": [ { "accuracy": 0.9844646832758659, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9849868932934968, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9865049307822789, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900724188865799, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9921472237788533, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9922438488881055, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956163021515271, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995710155534509, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959335466846824, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9961882449784561, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9960641162469983, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963272406572574, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966564121545831, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969219531079656, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981620476801732, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984537476802392, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998853459994088, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991810907223752, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999685404063999, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.17.block_sparse_moe": [ { "accuracy": 0.9825105251940457, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9831104778047455, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9853937920967215, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9858946396332038, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918324704478053, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9924780161512133, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9932109040796364, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9958901665246996, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9962225895571081, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9959033604123091, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964116581372524, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979242502564662, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982173990727844, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988894053661314, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9989357259688213, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990698788621414, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997032720712014, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.18.self_attn": [ { "accuracy": 0.9828036132415658, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9833275445393825, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9850646385824994, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9886820009958587, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991231842726273, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9913170986906871, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948623460889059, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949991586250498, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954284277229913, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995745127812322, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956535067482802, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959781154205924, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963316604446032, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965895853667078, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979598992810536, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982944330274078, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986819496782693, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990861396440059, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996395151216635, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.18.block_sparse_moe": [ { "accuracy": 0.9812489978380894, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9818904757205593, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9843247119631422, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.984866554407697, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912287615190604, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9919250775863858, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9927071541008589, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9955690165270904, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9959279354048991, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9955992970591117, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961438940617403, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977696251405991, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980833796772027, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988014601863382, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9988574652417906, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999001235217101, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996774997301767, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.19.self_attn": [ { "accuracy": 0.9806188491516208, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9811822591643584, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9829903420453009, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9865948173560595, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899504865813804, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900370405888871, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994086166745738, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9942225768770042, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946282585641664, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995013602629402, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949172499320028, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953647021228742, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995787318673377, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961481112864261, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976866751247527, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980698025442267, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984638095576022, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989359417630016, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995807464645048, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.19.block_sparse_moe": [ { "accuracy": 0.9798193899424452, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9805070630911934, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9831431766873912, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9837284286163355, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9905176047774914, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9912746853293165, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9921148215959731, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9952172547652337, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9956032431218773, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9952359528427845, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958282224221253, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975830332780453, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979258630498263, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987041889567321, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987604270143876, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989137196343539, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996535784908032, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.20.self_attn": [ { "accuracy": 0.9798203191082728, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9808933359913921, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.98310127241635, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9871158007728426, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899453759977692, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9901323043496201, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9940506695328575, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941997768872074, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946098422852198, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948880906633445, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949464147763425, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953217978809813, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957962280359903, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961231816805115, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976971792057157, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980492579664937, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998454423725458, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989012331015578, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995737388717795, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.20.block_sparse_moe": [ { "accuracy": 0.9781068696787483, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9788552025627149, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9817545439459776, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9824011885119897, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9897076640474168, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9905290250960541, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9914461989936075, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947920163701239, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9952076139712804, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9948279509860042, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954730781334403, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973780378503235, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977489298662955, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985856543981323, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9986576295555815, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988246310876046, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996182570359547, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.21.self_attn": [ { "accuracy": 0.976924734974378, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9779487223315396, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9800821081490109, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.984678185495891, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9881886860943938, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884285602326456, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930532222956812, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932893143329573, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937334116793385, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9940942231516697, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941462551023027, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9945307791012487, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950375767473719, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954650433317415, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972574922660562, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977114338329748, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981801756535118, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987255934821931, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995057898846235, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.21.block_sparse_moe": [ { "accuracy": 0.9759732377960494, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9767935893645412, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9798948083955207, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9805961399780292, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9887366610530176, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9896303104530824, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.990611355212566, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943006511189436, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947563895073376, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943437918187364, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950413090225897, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971316052282131, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975344394456203, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984562673152572, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985312126503375, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987118855903023, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995845347615326, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.22.self_attn": [ { "accuracy": 0.9749391555884167, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9759144969284534, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9784093636431193, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9839448232301756, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9871605932222385, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9874788287683929, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926539791355792, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9928273394409763, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932884030583266, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937290960040531, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935892641446308, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9940774710603842, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946975250311784, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951053159924126, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970724476121464, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975562722017792, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980666177641404, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986794627283847, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994811464382923, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.22.block_sparse_moe": [ { "accuracy": 0.9752010911899177, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9760685031743426, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9792932536649076, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.980008681010651, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9884363692743998, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9893483204374972, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9903373037170815, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9941718544516909, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9946256940449146, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9941910071249463, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949115083405846, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970534731584945, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974694587290287, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984177256885328, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984903499533079, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986710688957062, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995758875616287, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.23.self_attn": [ { "accuracy": 0.9794642702235203, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9801405072212219, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9822646328002999, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.986362889899235, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9892276886752561, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9894269752855364, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935018287138327, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936754126346817, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943279604623584, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946085705642441, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947371195732174, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951293768826872, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955917007510403, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959439121823954, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975418483862948, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979494480955365, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982933608484209, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988794759847224, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995309356829813, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.23.block_sparse_moe": [ { "accuracy": 0.975117964768096, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9759559259798966, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9790387325184909, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9797335856251026, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9882436381947053, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9891819981367964, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9901517942106646, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9940867553906221, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9945471258343834, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9940871385563361, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.994824590823172, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970018943208024, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974270670944335, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983922244711337, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984615009866262, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986366573292272, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995633060660973, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.24.self_attn": [ { "accuracy": 0.97873045636439, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9803160557915506, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.981878205006452, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9859712270431613, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989489937985414, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9898260429158414, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935534476841751, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939467139836204, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944702785241565, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948155413332739, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949274398876649, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952543781776177, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956245833248096, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959913116487625, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975758846052677, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997988667559663, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983836706164048, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988867029590581, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995533878176693, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.24.block_sparse_moe": [ { "accuracy": 0.9740971721904842, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9749379843277367, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9780485297210122, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9787312297542629, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9876935172237848, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9886891496808905, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9896459233407912, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.993827546677111, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943119365918008, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937988942194926, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9945740439978085, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968484987896916, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973013762531704, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983191059436649, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998382098850255, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985539691498209, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995494806774475, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.25.self_attn": [ { "accuracy": 0.9810727159247586, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9817998023017457, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9837681244764673, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9873997200476495, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9903452100330278, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9904887702953267, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941082583299201, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994283892704468, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994744514776884, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994992020172312, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951380696760392, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955663901361588, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959868257987852, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99633618076577, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977515089558437, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981467223076738, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984308433241064, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989633641058677, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999555408573207, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.25.block_sparse_moe": [ { "accuracy": 0.973278108788164, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.974142904050256, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9772885995181767, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9779811472092804, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9873078588984514, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9883219728638467, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9892984758572358, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936236810478333, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9941207448902883, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936087852049815, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944033448118716, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967576928289705, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972151489622009, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982675012556443, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983358162602312, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985129192937165, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995326006150878, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.26.self_attn": [ { "accuracy": 0.978304250459922, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9787903076135799, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9817026787878651, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9857519556229052, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.988931242180498, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9891534884901423, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.993001533647705, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934343469427213, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941957801286327, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945683097466826, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947381973095042, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951245203295624, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955439009752712, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958774647307828, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997465079552249, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979253317302975, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982626062180651, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998843338270059, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995099867948355, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.26.block_sparse_moe": [ { "accuracy": 0.9717314954062826, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9726369286838331, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9759652696942028, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.976701185401333, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9865735003136491, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9876372551447467, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9886740673774559, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932503358362929, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.993779846932739, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932482292091376, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9940779232665112, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965738348393259, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970557580113804, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981737076917565, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982432303153664, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984349798300842, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995107068036879, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.27.self_attn": [ { "accuracy": 0.9798493443645144, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9806313611272919, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9827687190845609, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9861097910294407, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897638666198442, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9898240247643307, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937091395503989, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938137672694498, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936681353045922, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9942309879697859, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994782211038431, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951623949638911, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955482449088442, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959317195523334, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975638717020813, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979887531494329, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982931967640883, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998886351498138, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99952243091416, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.27.block_sparse_moe": [ { "accuracy": 0.9706536060885379, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9716056264740857, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9750233960190886, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9757918974776801, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9860751330362338, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9871779356740022, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9882717609993721, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9929912638311323, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9935450872749483, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9929978841131455, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.993860313766881, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964506388886979, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969458878898111, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981021025757256, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981812821566373, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983828700527451, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994861883659676, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.28.self_attn": [ { "accuracy": 0.9764010871417428, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9775153799472671, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9804078019469192, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9846572501486853, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879894449718689, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9883481468025007, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924496008867496, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9927142346768003, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934834081091379, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938407125882804, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941298290474438, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9945493789855391, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950644432959196, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954658378601858, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972614463369706, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977117471985126, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980740076249563, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987168279629642, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994594229976524, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.28.block_sparse_moe": [ { "accuracy": 0.9691095457069183, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9701086973869487, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9738671172802386, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9747444703978928, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9853156142328915, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9865024931924907, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9877517514764086, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9925564550549576, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9931827435318968, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926322895886475, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9935356588102877, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962640366713075, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967862000983012, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979945944904006, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980890869138468, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998318577430358, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999457976025664, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.29.self_attn": [ { "accuracy": 0.9750184048163263, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9761813461878582, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9792807003189075, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9841969390644839, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9873854016493026, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9876974420867076, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9923659580220517, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9924986446276307, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9930798000116882, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934166106640508, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936595350552938, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941690966462422, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994792222094379, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952067365306184, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971197588234454, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975832247322327, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979526761024701, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986639093319689, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999437710588896, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.29.block_sparse_moe": [ { "accuracy": 0.9694105410262158, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9703556649190815, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.974039958005673, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9749041369773055, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.985426432109977, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.986569060453851, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9877897502707416, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926261791567269, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932268623899865, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926736545994094, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9935730699400761, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962859394420919, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968020399468706, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980086355935782, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980998010815758, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983233846601491, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994594784312587, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.30.self_attn": [ { "accuracy": 0.9807080853436339, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9812987485508385, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9837313034130555, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9873730802516404, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900184964614087, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9902094686403871, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937644825856152, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939654433707658, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943768270436282, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99463111570252, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994938692813249, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995346651731157, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957676378912047, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961451452568566, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976390341762453, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998042388016505, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982987457494202, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989393263332251, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995257503270343, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.30.block_sparse_moe": [ { "accuracy": 0.9683395468286777, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9693498410480587, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9730781335383654, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9739540018827507, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9848676993462601, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.986071658173674, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9873011837909488, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923628985783771, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9929815334218898, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923918928313804, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9933314572921709, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961434900883193, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966852236906752, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979368214691547, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998027004852989, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982492786305222, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994411893222652, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.31.self_attn": [ { "accuracy": 0.978880679823066, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.979760362962751, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9826643172357428, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9864246005094365, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9891416661833462, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9895230286117447, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929712313276372, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934341562176613, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939157544684253, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9942382128394552, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946489219896888, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950861994079069, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955224654220632, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958989688709966, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974541710015681, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979404280280793, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980975514228799, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988824175027395, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999484602847492, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.31.block_sparse_moe": [ { "accuracy": 0.9675117940887025, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9685092049797899, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9721841901344689, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9730313734004372, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9844580401029241, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9856892260967901, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9868751117763551, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921674740402714, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9927992282533332, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921845156503352, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9931479099108592, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9960404903088745, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965947507758086, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978863311085948, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979743745276018, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998189499313747, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994306357801083, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.32.self_attn": [ { "accuracy": 0.9797911260669169, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9806900393511904, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9831804595397491, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9871793711116832, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9898241763738426, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900060785003006, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936789843670436, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939025831898969, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943685372188491, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946306515117421, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949027764684472, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952737835088843, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995704368982268, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960813681276417, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976304536685348, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980215685209259, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983344300539772, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988725667016728, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995303210376542, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.32.block_sparse_moe": [ { "accuracy": 0.966809544810339, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9678392374986097, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9714882264992124, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9723438575471702, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9840824523156411, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9853423513766182, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.986518852731311, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991984509593366, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926306263200546, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991995108534435, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9929804549013314, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959474683763754, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965123235689182, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978410754977775, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979273589376948, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981383255584851, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994203909928298, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.33.self_attn": [ { "accuracy": 0.9791210848151853, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9807662278119671, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9828984272225123, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9863934005169492, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897221459664012, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899065060001847, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.993545269667122, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937334640294706, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941542385342089, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944922346744294, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949142459142757, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952695866875154, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956943524976898, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960247649272022, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975910395824987, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979886492028048, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982717429152935, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988456677069495, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995035322637321, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.33.block_sparse_moe": [ { "accuracy": 0.9660906080940836, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9671223228307146, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9708264442650896, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.971687937057332, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9837220801334632, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9850239571870157, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9862020411528647, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918084011511191, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9924749615112025, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918070610024428, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9928224543599706, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958455275352064, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964333040835825, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977906521127903, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978739379684588, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980820734911647, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994086432209435, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.34.self_attn": [ { "accuracy": 0.9781104020382229, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9801524446688985, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9825544571503997, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9869678779505193, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989436553472555, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897576490123021, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9933524327842813, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993679861979265, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9942674423984595, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945650804714349, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994826928528614, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952031237220293, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956459522516908, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959817939609485, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975807017058527, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979650407698691, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998305932085618, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988692879989302, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995103362537185, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.34.block_sparse_moe": [ { "accuracy": 0.9666787087917328, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9677016268435278, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9712387904230702, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9720601587997455, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.983976003419804, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9852501392168435, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9863745021927905, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991946916169438, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9925967171101978, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991944875347575, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9929348877876213, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959186404992483, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964887907262892, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978294222260287, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979106124711076, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981076057638826, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994154455780517, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.35.self_attn": [ { "accuracy": 0.9851258067218097, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9852306933485364, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884262252353916, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9905652569753951, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924296333797669, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9925022141174659, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954059486105842, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956924723044626, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959552537551836, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9961124428082258, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9960689428507498, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963098128669357, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970385521760603, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997292182696248, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983404227042276, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986070477017691, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987643006024882, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992870607912712, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996474441508517, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.35.block_sparse_moe": [ { "accuracy": 0.9656802578584144, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9667221695968979, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9703145502625328, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9711425880362329, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.983481922284945, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9847999920480346, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9859239643831786, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9916945371384683, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923688288483965, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9916905077958578, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.992714228696729, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957896012837362, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963813756568063, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977607372351677, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978439866589677, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980427594058902, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993988972641282, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.36.self_attn": [ { "accuracy": 0.9823985607118199, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9830564632031479, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9849083500080987, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882251381579983, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907501411477202, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909193569206094, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943677633755693, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946568598725686, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950571569957232, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995305728044753, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953354599297439, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956431341798682, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961858928252599, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965076455111174, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978863690354812, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982490478892272, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998512081685476, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990302789558069, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995747820847962, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.36.block_sparse_moe": [ { "accuracy": 0.9645752383297995, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9657698246209245, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9692896560422684, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9701048815132756, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.983217219331939, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9845241037638564, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9856472739361619, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915438486802343, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921905955691871, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915635819105726, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9925974291798315, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995728902408461, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963258234293837, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977287039418068, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978131469657743, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980124806568614, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993892035548159, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.37.self_attn": [ { "accuracy": 0.9788082045453944, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9800269348350795, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9820342643774653, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9865340397840268, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893786564322287, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9896111680488837, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9932442336882416, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935410665348172, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994006344607394, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943415219527915, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946943080959547, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950516127344025, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954135410340601, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957599908015445, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974237962772972, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978596099015129, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981971446571773, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987882650120331, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994826659630666, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.37.block_sparse_moe": [ { "accuracy": 0.9647864422907955, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9658875198740708, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9693837961378066, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9701946991446772, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9831443979453883, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9844998329583752, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9856088839863476, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915241430954713, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9922179201813904, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915247799485529, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9925733604958575, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957080744329447, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963095576355332, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977171957970744, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978023640149715, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979969501789463, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993852876252373, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.38.self_attn": [ { "accuracy": 0.9841159040500459, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9851759376592542, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9873327172214263, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9895646276295578, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9918393320894163, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.992162714173135, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950682556526246, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953824735589718, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9957254322637853, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959502394531706, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959447200924746, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961931302922925, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969142147793287, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971975260659268, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982492896777234, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985883320240598, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986803282211584, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991959369140923, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996194080633455, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.38.block_sparse_moe": [ { "accuracy": 0.9656202734300965, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.966685011492748, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9700933540435998, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9708825140995415, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9833644692737021, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9846989243340335, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9857500434461001, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991624994791652, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923028454772735, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991626157376327, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9926601138554121, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957574394188429, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963528283651134, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977373000798061, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978279765805622, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980118199233211, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993901527992229, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.39.self_attn": [ { "accuracy": 0.9820882893706623, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.982786864336384, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9854818535478491, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9887486340870199, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907131194578189, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910280806698689, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943860770864902, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946460197557156, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951065440147527, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953282547036284, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953481214562137, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956903318923555, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963880425294567, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996681023940542, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980059605961862, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983132474865813, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985382013378272, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990463920464543, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995620909704532, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.39.block_sparse_moe": [ { "accuracy": 0.9653439011032644, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9663861091983945, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9697754101729706, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9705658843250651, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9831967931240797, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9845571650290176, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9856046329154388, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915506259391182, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9922371611342227, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915451054755402, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9925887252228629, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957172040165844, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99631706035794, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977230806834996, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978073575215316, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997988372119021, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993868514353189, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.40.self_attn": [ { "accuracy": 0.9844795256166866, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9852507360720713, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9870388461767059, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897521013443015, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919007607294541, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920967645256927, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947674662659043, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950697283677169, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954688493039852, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9957257397105231, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958842714611245, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961330965528951, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967028174708646, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969640234362727, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981548301691777, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998469288742758, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986089834618677, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991315339192577, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995823260105681, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.40.block_sparse_moe": [ { "accuracy": 0.964916215915429, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9659773138792891, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9693431095465234, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9701304035261273, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9829630779200479, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9843420113779997, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9853879859867064, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914288476452624, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921265008712286, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914210693861701, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9924827484953168, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956527184695005, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962632672885727, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976853688651892, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977718187054914, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979498604998777, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993735184635673, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.41.self_attn": [ { "accuracy": 0.9815396792500427, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9825508131792671, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9850833174821577, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9883272323610359, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907069414559948, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909178915580636, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994236603599826, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945888375667365, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949167857044622, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951638581206728, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952625135989174, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955385604407638, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961996496973657, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965741075207725, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979132339664686, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998256951625655, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984787619170291, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990101411638438, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995507965474022, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.41.block_sparse_moe": [ { "accuracy": 0.9650151646254879, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9660712038411906, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9694168648535484, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.970195124072856, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9829972567232815, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9843667495603624, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9854020090577634, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99145073884804, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.992147873670451, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99144289551891, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9925012284595716, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956663017847428, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962728868704289, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976990557410509, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977809891509088, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979578324924469, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993815713954207, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.42.self_attn": [ { "accuracy": 0.9819283818728045, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9828531190654949, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9849050523909298, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9880160375890371, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9906639136177929, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909086961936402, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942173458709332, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994550936365206, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949316475417858, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951771081939927, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952711936723637, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955610548342136, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961732094984894, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964929392720622, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978529804226893, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982434466517972, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984442515874674, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990006429970412, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995533731876334, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.42.block_sparse_moe": [ { "accuracy": 0.9644776538602615, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9655334301488965, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9688697893074468, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9696550986386443, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9827032840663665, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9841015079597893, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9851330955953974, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912953173349562, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9920102051263185, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912951739357883, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9923692908824274, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955910701843861, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962092110593068, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976571312627608, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977437180197356, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997919621613906, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993711628623331, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.43.self_attn": [ { "accuracy": 0.986739686396169, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9874564749433806, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897193579542402, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914827310774279, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929596364841258, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9934243292449728, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957715877037692, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9960271148991428, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9963975397404283, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965432413951739, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996634699367477, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996850751732525, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974431489030585, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976709808443526, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985760618001223, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988204581620123, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988917024977702, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993449973821713, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996625222709937, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.43.block_sparse_moe": [ { "accuracy": 0.9636772891604587, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9647782995904747, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9681171638106829, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9689014852537137, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9822970581682104, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9837398915796688, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9847631761559138, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910937431758564, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918249338424128, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991085294143934, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9921941549125078, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954852092349412, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961223488575534, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976010568942385, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976895168288856, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978629159250934, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993578874686806, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.44.self_attn": [ { "accuracy": 0.9827552912756801, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9838953444636181, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9857886268414164, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882540726347974, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9908857420331946, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907543821211316, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99424341475395, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944408899418226, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948685955697376, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951494143736598, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952266125596667, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956737682468405, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963762715066734, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966439610518711, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979937714647109, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983250363791165, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984700180724949, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990280499393855, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995313734413541, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.44.block_sparse_moe": [ { "accuracy": 0.9641923352487778, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9652819001164875, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9685685939498638, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9693337827430744, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9825518644954029, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9839775035844037, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9849698554332319, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912280295485336, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9919428713129539, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912268992905554, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9923110494978333, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955564646393453, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961800292796014, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976392821224994, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977254448458552, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978929011236974, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993698172574544, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.45.self_attn": [ { "accuracy": 0.9828679181243244, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.98333700824725, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9854810528555199, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884918807319513, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910235912539065, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9913700826858219, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944621283610008, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947589301553211, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951814592472816, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954227202841522, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954681128361508, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958391309054079, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963610381819308, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966680330322369, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979517453680992, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983118236138436, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985221379963485, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990624531974869, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995707296582527, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.45.block_sparse_moe": [ { "accuracy": 0.962963559419701, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.964090654038285, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9674288196685283, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9682130759189788, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9819302438317161, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.983406023661557, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.984421545884719, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9909073982485815, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991654768046972, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9909077732039517, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9920316662663888, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953929845635828, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996042082159731, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975516110242001, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976411549019089, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978120021742621, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993446491594353, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.46.self_attn": [ { "accuracy": 0.983164702983279, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9840481555775592, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9865624837969479, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9894610459818259, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912159529661662, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991527303318052, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947296685514677, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950321773557287, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954497053189889, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956701336256963, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956804936702707, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959428017517846, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965945022229693, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968810796149468, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998053157383478, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984129148682481, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985615848686153, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991283928485293, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999586598774553, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.46.block_sparse_moe": [ { "accuracy": 0.9627685673338802, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9638988844266063, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9672101137571429, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9679881649974146, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9818056575267723, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9832928283630233, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9842975320245483, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9908417414775804, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915953802533055, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.990844548067176, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9919795492345369, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953639759608593, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996013681778047, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975350320670068, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976260345429182, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977952250414283, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993395559308977, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.47.self_attn": [ { "accuracy": 0.9840167531939713, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.984729243736518, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9875817332710874, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899783750358773, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9917381855070984, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9921100106776545, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950876219051057, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954538203800392, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958054202755815, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959802522544602, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959210570975158, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996224659311242, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968462405162618, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971343193779161, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982201913861852, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985337710853568, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986971616438639, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992474658724141, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996141668234486, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.47.block_sparse_moe": [ { "accuracy": 0.9608605292282606, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9620586202332848, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9655043775901982, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9663151906018979, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.980892125370079, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9824401087274677, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9834705727223895, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9903872592227632, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911694890133252, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9903917636134123, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.991575924532586, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9951355673348237, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958163757171286, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974174442161855, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975109877689791, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976837209021476, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993131751451936, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.48.self_attn": [ { "accuracy": 0.9811732383739007, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9821883740001603, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9844349326173726, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882947292767073, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9901286633195061, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9903724000201022, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9940140066472324, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943003285627224, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947873504317709, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950468478307716, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950758102962649, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953821217151064, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996052194349958, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963832427012292, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978014677456629, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981598330869976, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984257978472957, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989909439709185, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995406324252192, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.48.block_sparse_moe": [ { "accuracy": 0.9590452082063022, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9603191755319896, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9638006138174158, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9646304016442675, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.980011476693969, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9816336561190454, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9826854350486476, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989914282053513, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907323209029671, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899421463671484, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9911833903017012, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949083953581163, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956195921833185, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972825867405749, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973935376068479, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975711130169466, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992703689816775, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.49.self_attn": [ { "accuracy": 0.9798006301647738, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9809172454437143, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9837043997586558, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9868143912110674, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893262299807056, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897223800714863, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935594709148925, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993931242754977, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945370297829964, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947901310913853, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948096690625933, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952286008843466, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959791318856572, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964064176096336, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977953196809578, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981925700987248, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983202140275879, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989747194922529, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995260482415018, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.49.block_sparse_moe": [ { "accuracy": 0.9552563896500751, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9567378762325174, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9604732544210396, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9613686114745705, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9783544302672932, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9800833418083034, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.981225414838838, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989061647388888, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899469741963243, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890983920838488, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9904450545283524, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944768248027877, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952525589370022, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970522813780821, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997172589879483, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973699426780896, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992053077592956, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.50.self_attn": [ { "accuracy": 0.9743706391830194, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9759944725389543, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9797063261564625, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9839410607359911, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9871105814940835, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9871919371285721, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924986082815418, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.992857730849401, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9933665620016032, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993707528729972, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936106475560289, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941221808426475, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950540375797764, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954419283098296, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972303953193324, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976921245137131, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980432716823232, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987457213972352, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994466588924344, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.50.block_sparse_moe": [ { "accuracy": 0.9538631692136589, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9548629004704325, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.958871432334969, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9597565723877204, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9781527160421798, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9798798371890658, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9810611009107608, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9889251604235094, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9898313859379605, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890183827006503, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9903569440847557, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944206987090997, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9951977993508703, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969868746794466, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971397599744561, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973501355117677, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991408995671296, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.51.self_attn": [ { "accuracy": 0.9692109957533447, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9705343643497479, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9748597227919259, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9801596774670639, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9847100997637761, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9848966852908856, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912720883853341, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914643012330329, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9920448064068822, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9923971922283894, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9922385753475522, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927493112504875, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937473589824023, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994261751539613, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965594447046322, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970812426794222, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976943507484209, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984506098859894, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993436126709352, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.51.block_sparse_moe": [ { "accuracy": 0.9548502283072785, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9563235234665244, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9600217315907541, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9609386072935242, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9780145702197364, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.979829109175817, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9810163533609164, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9888357066941497, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9897414204468461, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9889329509368461, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9903036438320812, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.994390783819223, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995176652385118, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969764230527768, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971393245014999, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973497557956236, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991878444561735, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.52.self_attn": [ { "accuracy": 0.968308788185057, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.970416539201611, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9741089738121158, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9791465731533734, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9848337765587004, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9850534006070933, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991178925565787, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914545385951274, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9919819312758351, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9923631075377527, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9924515989412995, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929067755461132, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936391525460702, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941748213513117, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965402768728765, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970260471240372, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976707361209648, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984029967913797, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993321518689443, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.52.block_sparse_moe": [ { "accuracy": 0.9530289929948355, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9545582530922011, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.958393585623095, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.959402001021724, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9769136674800202, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9788428015124641, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9802184386767054, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9881717564145985, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9891324616282394, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9883398646488786, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.989797854747035, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9940719125596317, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949204094806, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967673307921934, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969778200795286, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972327996020842, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991041701041946, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.53.self_attn": [ { "accuracy": 0.9646889231118717, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9669753940481889, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9722834242213714, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9784947423460452, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.98270906311901, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9831097413736739, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899667564926571, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9903679936143913, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9909866775238985, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.991402566089834, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914521724230757, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920412362308094, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931929831049944, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936921707428011, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962249142853054, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968418844571141, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973587644850149, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982898956299514, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992527705096443, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.53.block_sparse_moe": [ { "accuracy": 0.9498641586029216, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9511243368646032, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9554662163320341, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9565906670728797, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9755590866859022, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9775363990154705, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9790838437252923, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9875101763264913, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9885524113564507, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9876527436273662, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.989176789421196, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9937248585721183, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.994610231933429, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966092389444575, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9967918321122661, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970853483292127, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990598067800563, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.54.self_attn": [ { "accuracy": 0.9643398335105494, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9661937820676126, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9717872193279234, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9775128347897216, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9827873425460175, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.983049281774775, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900370865913206, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9904281928596136, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9911117951728796, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9915613495303612, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9913770603261104, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919180090686208, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930480602197349, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936474298960284, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961613303074908, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967794270732587, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973955505870675, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982702149029233, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992648379304324, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.54.block_sparse_moe": [ { "accuracy": 0.9517187368320792, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9532429294562653, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9570195979781841, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9580523151120073, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9763263041937822, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9782530770784146, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9796618363378864, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9877042156203013, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9888061055864551, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9879439011167147, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9894755496398399, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9938681569734686, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9947272821844212, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966338866414797, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968711095065564, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971472031351081, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990590097885089, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.55.self_attn": [ { "accuracy": 0.9676557597949317, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9689426476037816, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9748844130356845, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9794637824555761, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9850183778785562, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.985214282562466, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914367114879975, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9917525682402285, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9922337794774457, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9924648344075602, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9926084430251074, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928554722941235, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935455944291071, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944519471671236, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965411543870639, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971632962237651, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976885238490802, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9983573869348651, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992653593133007, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.55.block_sparse_moe": [ { "accuracy": 0.9596437212864035, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9609394465622149, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9639407208091334, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9649150388217286, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9809254742668647, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9825626305353484, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9838071117961877, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9898470462986121, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9908758094534278, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9903472984993928, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9915552364025069, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950000491287363, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957048502799711, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971330496911449, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974014385738165, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976832569503274, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989980552682506, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.norm.norm": null, "lm_head.linear": null }, "last_module_idx": 114 }