{ "in_dir": "../models/dbrx-instruct", "out_dir": "dbrx", "cal_dataset": null, "bits": 7.0, "dataset_rows": 100, "measurement_rows": 16, "length": 4096, "measurement_length": 2048, "head_bits": 8, "shard_size": 8192.0, "compile_full": "dbrx", "rope_scale": null, "rope_alpha": null, "output_measurement": null, "progress": "finished", "cal_filename": "dbrx/cal_data.safetensors", "last_module_idx": 82, "measurement": { "model.layers.0.self_attn": [ { "accuracy": 0.9766400855426726, "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": 0.9823410942366249, "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": 0.9837163330889062, "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": 0.9860431830349722, "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": 0.9865386483602618, "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": 0.9872829027375892, "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": 0.9878091478083086, "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": 0.9887715210451892, "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": 0.9905900257688603, "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": 0.9907734892821234, "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": 0.994194732929923, "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": 0.9947365149610529, "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": 0.9947038777050023, "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": 0.9952606072428783, "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": 0.9969800165915665, "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": 0.9976415273206505, "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": 0.9971056761967606, "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": 0.998962870924611, "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": 0.9991672076223614, "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": 0.9808896456256901, "total_bits": 7093506816, "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": 0.982329665536159, "total_bits": 7357747968, "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": 0.9830274851992726, "total_bits": 8211542016, "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": 0.9840934623621012, "total_bits": 9215029248, "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": 0.9923123262185407, "total_bits": 10383803136, "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": 0.9928054694502958, "total_bits": 10670075904, "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": 0.9932551523057842, "total_bits": 11472446208, "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": 0.9951929920720622, "total_bits": 13112832768, "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": 0.9956647995736843, "total_bits": 13306195968, "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": 0.9961729654552121, "total_bits": 13501219584, "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": 0.9966416872155509, "total_bits": 13787492352, "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": 0.9980159806590038, "total_bits": 16618636032, "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": 0.9982824173363808, "total_bits": 16904908800, "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": 0.9986294933539619, "total_bits": 19248464640, "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": 0.998959807121489, "total_bits": 19903945728, "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": 0.999081271100103, "total_bits": 21690719232, "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": 0.9994237050260907, "total_bits": 25478175744, "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": 0.9353808621434789, "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": 0.9530706936983686, "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": 0.9578224097035433, "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": 0.9718280015023131, "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": 0.9724271290592457, "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": 0.973671465818035, "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": 0.9767954083845803, "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": 0.9782315597526337, "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": 0.9800023482622284, "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": 0.9808496266211334, "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": 0.9862882796871034, "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": 0.9877260965540221, "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": 0.9875655684522108, "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": 0.9890654219902659, "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": 0.9923218371297576, "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": 0.9943961227851871, "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": 0.9929382969184142, "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": 0.9965237161806343, "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": 0.9979652358329316, "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": 0.9535420675712981, "total_bits": 7093506816, "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": 0.9565316690132022, "total_bits": 7357747968, "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": 0.9611088649712896, "total_bits": 8211542016, "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": 0.9623671742156148, "total_bits": 9215029248, "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": 0.983641623666412, "total_bits": 10383803136, "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": 0.9851529320309821, "total_bits": 10670075904, "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": 0.9873771293848557, "total_bits": 11472446208, "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": 0.991547693768026, "total_bits": 13112832768, "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": 0.992489237509864, "total_bits": 13306195968, "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": 0.9922800194594616, "total_bits": 13501219584, "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": 0.9932131001198861, "total_bits": 13787492352, "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": 0.996095142278232, "total_bits": 16618636032, "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": 0.9966421230346896, "total_bits": 16904908800, "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": 0.997816523423688, "total_bits": 19248464640, "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": 0.9979712567904866, "total_bits": 19903945728, "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": 0.9984350846988443, "total_bits": 21690719232, "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": 0.9992696846602485, "total_bits": 25478175744, "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": 0.9445337501207465, "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": 0.9512198822768896, "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": 0.958160980604589, "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": 0.9695258823860633, "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": 0.9707244334644393, "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": 0.9717433834448457, "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": 0.9766249870146183, "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": 0.9776884273772961, "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": 0.979888150983147, "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": 0.9805750549574824, "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": 0.9851793757958436, "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": 0.9858246390021553, "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": 0.9866495981910511, "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": 0.9873208776238913, "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": 0.992079330571486, "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": 0.993712365290297, "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": 0.9927877290296907, "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": 0.9968914055682131, "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": 0.9981063970909944, "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": 0.9622737136914542, "total_bits": 7093506816, "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": 0.9688787696588981, "total_bits": 7357747968, "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": 0.9699361156672239, "total_bits": 8211542016, "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": 0.9700016144074892, "total_bits": 9215029248, "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": 0.9948667024237741, "total_bits": 10383803136, "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": 0.9944773110512056, "total_bits": 10670075904, "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": 0.9946974766636757, "total_bits": 11472446208, "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": 0.99825886362171, "total_bits": 13112832768, "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": 0.9982651387537388, "total_bits": 13306195968, "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": 0.9981861070570487, "total_bits": 13501219584, "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": 0.9983244025957232, "total_bits": 13787492352, "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": 0.9988975728491297, "total_bits": 16618636032, "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": 0.9989761213977257, "total_bits": 16904908800, "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": 0.9991888767154274, "total_bits": 19248464640, "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": 0.9992584494757466, "total_bits": 19903945728, "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": 0.9992987263448365, "total_bits": 21690719232, "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": 0.9993976275038318, "total_bits": 25478175744, "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": 0.9932222575267875, "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": 0.994418225441079, "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": 0.9954146107683253, "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": 0.9961045001124914, "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": 0.9963329385684215, "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": 0.9963818413180936, "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": 0.9968991973713089, "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": 0.9971583030082433, "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": 0.9974433064711966, "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": 0.9974907171287525, "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": 0.9978717460578895, "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": 0.99806027697387, "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": 0.9982795540617142, "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": 0.9984885529924404, "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": 0.9990552016219896, "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": 0.9992283995915817, "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": 0.9991029559437026, "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": 0.999724811672135, "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": 0.9997588932187922, "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": 0.9916348275912338, "total_bits": 7093506816, "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": 0.9919083684094643, "total_bits": 7357747968, "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": 0.9930307147044101, "total_bits": 8211542016, "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": 0.9933120928226823, "total_bits": 9215029248, "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": 0.9960588849835882, "total_bits": 10383803136, "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": 0.9963593836077244, "total_bits": 10670075904, "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": 0.9967695052282101, "total_bits": 11472446208, "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": 0.9980095344898904, "total_bits": 13112832768, "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": 0.9981741186424379, "total_bits": 13306195968, "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": 0.9980217093068763, "total_bits": 13501219584, "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": 0.9982648597598247, "total_bits": 13787492352, "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": 0.9989977905015143, "total_bits": 16618636032, "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": 0.999139151495336, "total_bits": 16904908800, "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": 0.9994651264205459, "total_bits": 19248464640, "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": 0.9994864075775504, "total_bits": 19903945728, "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": 0.9995617682821161, "total_bits": 21690719232, "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": 0.9998585801091203, "total_bits": 25478175744, "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": 0.9903139247264909, "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": 0.9904361690002444, "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": 0.9927916401310971, "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": 0.9935266668663213, "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": 0.993763469858095, "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": 0.9938958456225105, "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": 0.9940793644575598, "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": 0.9943397809616535, "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": 0.9950663617836606, "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": 0.9951150576939414, "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": 0.9958231663699016, "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": 0.9958683844704769, "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": 0.9965988816019442, "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": 0.997016919736971, "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": 0.9982315417085039, "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": 0.9985154799145867, "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": 0.998288536321764, "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": 0.9995359246914679, "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": 0.9995401424086211, "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": 0.9876921525176027, "total_bits": 7093506816, "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": 0.9880844993565819, "total_bits": 7357747968, "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": 0.989639420834321, "total_bits": 8211542016, "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": 0.9900299060329991, "total_bits": 9215029248, "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": 0.9941629056610461, "total_bits": 10383803136, "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": 0.9946175296545813, "total_bits": 10670075904, "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": 0.9951802400571882, "total_bits": 11472446208, "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": 0.9970527549464206, "total_bits": 13112832768, "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": 0.9972986742995043, "total_bits": 13306195968, "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": 0.9970658084248977, "total_bits": 13501219584, "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": 0.9974311009803052, "total_bits": 13787492352, "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": 0.9985126975638865, "total_bits": 16618636032, "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": 0.9987252065130681, "total_bits": 16904908800, "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": 0.9992070265158729, "total_bits": 19248464640, "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": 0.9992379084218747, "total_bits": 19903945728, "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": 0.9993409552078351, "total_bits": 21690719232, "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": 0.9997906420695041, "total_bits": 25478175744, "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": 0.9877233699356255, "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": 0.9886307849083096, "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": 0.9899964732804188, "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": 0.9931365098669439, "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": 0.9936337966692487, "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": 0.9937972436279157, "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": 0.9955950350145271, "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": 0.99574157562548, "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": 0.9960852864498616, "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": 0.9961950739226794, "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": 0.9968080113268163, "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": 0.9970106441913614, "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": 0.9971533521280722, "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": 0.9973735934778999, "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": 0.9983727950633405, "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": 0.9986682233881622, "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": 0.9987238980339546, "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": 0.9993787104207226, "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": 0.9996650030968196, "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": 0.9837771109305322, "total_bits": 7093506816, "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": 0.9842730217098602, "total_bits": 7357747968, "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": 0.986477317141467, "total_bits": 8211542016, "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": 0.9870271738863697, "total_bits": 9215029248, "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": 0.9921570391222638, "total_bits": 10383803136, "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": 0.9927814258752685, "total_bits": 10670075904, "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": 0.9935694930486774, "total_bits": 11472446208, "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": 0.9960364039258198, "total_bits": 13112832768, "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": 0.9963650826861052, "total_bits": 13306195968, "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": 0.9960532358565711, "total_bits": 13501219584, "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": 0.9965429698656264, "total_bits": 13787492352, "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": 0.9979993422518754, "total_bits": 16618636032, "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": 0.9982842649899938, "total_bits": 16904908800, "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": 0.9989341575993947, "total_bits": 19248464640, "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": 0.9989747232864733, "total_bits": 19903945728, "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": 0.99912027758393, "total_bits": 21690719232, "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": 0.9997205977302465, "total_bits": 25478175744, "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": 0.9862762849140716, "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": 0.9877981515111107, "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": 0.9899420709545282, "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": 0.9923973304378849, "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": 0.9928357815737591, "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": 0.9933861982457242, "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": 0.9947683851442937, "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": 0.9952821365288017, "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": 0.9955073520602462, "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": 0.9956253090089089, "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": 0.9966014107738278, "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": 0.9968245014162292, "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": 0.9970805191183708, "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": 0.9973312360173288, "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": 0.998383659684087, "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": 0.9986372391959824, "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": 0.9986435151995944, "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": 0.9993825157013673, "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": 0.9996325979583688, "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": 0.9795784696593488, "total_bits": 7093506816, "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": 0.9801731272110421, "total_bits": 7357747968, "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": 0.9827807048384688, "total_bits": 8211542016, "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": 0.9834482140692049, "total_bits": 9215029248, "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": 0.9899951423591885, "total_bits": 10383803136, "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": 0.9908015627205666, "total_bits": 10670075904, "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": 0.9917469920137113, "total_bits": 11472446208, "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": 0.9949440321162049, "total_bits": 13112832768, "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": 0.9953613137446442, "total_bits": 13306195968, "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": 0.9949524706646212, "total_bits": 13501219584, "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": 0.9955835517060845, "total_bits": 13787492352, "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": 0.997439240727353, "total_bits": 16618636032, "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": 0.9978057868092468, "total_bits": 16904908800, "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": 0.9986369550782623, "total_bits": 19248464640, "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": 0.9986874932671382, "total_bits": 19903945728, "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": 0.998859111545148, "total_bits": 21690719232, "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": 0.9996438620828006, "total_bits": 25478175744, "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": 0.9836413144673172, "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": 0.9845680530722204, "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": 0.9872024576424768, "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": 0.990543299918308, "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": 0.9917148824560603, "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": 0.9919845786705417, "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": 0.9946204755933171, "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": 0.9946688416193387, "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": 0.9952955704586777, "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": 0.9954903891775757, "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": 0.9958919823402539, "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": 0.9961659499860712, "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": 0.9966712935312995, "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": 0.9969054589623978, "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": 0.9981459698955357, "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": 0.9984032708713417, "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": 0.9985690198556864, "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": 0.9991573504735961, "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": 0.99962733712585, "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": 0.9759150837597094, "total_bits": 7093506816, "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": 0.9766093037140212, "total_bits": 7357747968, "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": 0.9796920091305908, "total_bits": 8211542016, "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": 0.9804738396288533, "total_bits": 9215029248, "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": 0.9882791756995415, "total_bits": 10383803136, "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": 0.9892344762277054, "total_bits": 10670075904, "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": 0.9903475213183188, "total_bits": 11472446208, "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": 0.9940768256660943, "total_bits": 13112832768, "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": 0.9945600951262015, "total_bits": 13306195968, "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": 0.9940778956209358, "total_bits": 13501219584, "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": 0.9948325812093619, "total_bits": 13787492352, "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": 0.9970042591203788, "total_bits": 16618636032, "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": 0.9974314317591243, "total_bits": 16904908800, "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": 0.9984060854614271, "total_bits": 19248464640, "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": 0.9984652550345728, "total_bits": 19903945728, "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": 0.99867240293411, "total_bits": 21690719232, "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": 0.9995784213824663, "total_bits": 25478175744, "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": 0.978427231458849, "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": 0.9803257097097996, "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": 0.9830038712703084, "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": 0.9870507750096486, "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": 0.9889480737650669, "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": 0.9889997668368252, "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": 0.9930589757982249, "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": 0.9931660660199428, "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": 0.9938091824970845, "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": 0.9940754748562253, "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": 0.9942760703974942, "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": 0.9946560318471471, "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": 0.9956069079619881, "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": 0.995955096036931, "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": 0.9975868087310932, "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": 0.997937203139851, "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": 0.9982098860277083, "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": 0.9989322433837898, "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": 0.9995185340724818, "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": 0.9713106311269497, "total_bits": 7093506816, "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": 0.9721706984809747, "total_bits": 7357747968, "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": 0.9759871775382444, "total_bits": 8211542016, "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": 0.9769594695811209, "total_bits": 9215029248, "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": 0.9859602680111206, "total_bits": 10383803136, "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": 0.9870870471319282, "total_bits": 10670075904, "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": 0.9884561927817566, "total_bits": 11472446208, "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": 0.9928908016299829, "total_bits": 13112832768, "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": 0.9934736312563089, "total_bits": 13306195968, "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": 0.9929049456504321, "total_bits": 13501219584, "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": 0.9937889279282995, "total_bits": 13787492352, "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": 0.996391960073532, "total_bits": 16618636032, "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": 0.9969057361595333, "total_bits": 16904908800, "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": 0.998066715837922, "total_bits": 19248464640, "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": 0.9981541703148794, "total_bits": 19903945728, "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": 0.998412156278439, "total_bits": 21690719232, "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": 0.9994918623664202, "total_bits": 25478175744, "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": 0.9755725602766401, "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": 0.9767142407547095, "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": 0.9793881371519283, "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": 0.983662656251047, "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": 0.9876262972646049, "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": 0.9879008108553918, "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": 0.9927816166432182, "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": 0.9929404869785925, "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": 0.9935941640716536, "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": 0.9939580804080164, "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": 0.9938899986997345, "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": 0.9944004386197776, "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": 0.9952717390352566, "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": 0.9956953972363599, "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": 0.9974295031279325, "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": 0.9977985528745003, "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": 0.998162889824016, "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": 0.9988371228537289, "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": 0.9994986603537415, "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": 0.9647371404265103, "total_bits": 7093506816, "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": 0.965771301551477, "total_bits": 7357747968, "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": 0.9707053999514564, "total_bits": 8211542016, "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": 0.9719695689548787, "total_bits": 9215029248, "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": 0.982819257034479, "total_bits": 10383803136, "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": 0.9841771395631919, "total_bits": 10670075904, "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": 0.9859675298763537, "total_bits": 11472446208, "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": 0.9912779334270837, "total_bits": 13112832768, "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": 0.9919915324150536, "total_bits": 13306195968, "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": 0.9913215106598249, "total_bits": 13501219584, "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": 0.9923915757834422, "total_bits": 13787492352, "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": 0.9955928140613985, "total_bits": 16618636032, "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": 0.9962190189184376, "total_bits": 16904908800, "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": 0.9976441569344484, "total_bits": 19248464640, "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": 0.997745967494618, "total_bits": 19903945728, "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": 0.9980911265991285, "total_bits": 21690719232, "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": 0.999383818609395, "total_bits": 25478175744, "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": 0.9728767771616971, "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": 0.9746289542925201, "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": 0.9777197357218125, "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": 0.9829467140554794, "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": 0.9864730900521145, "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": 0.9866332920072111, "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": 0.9923987662064304, "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": 0.9926264361221049, "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": 0.9931701553654612, "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": 0.9935121060545115, "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": 0.9932933418067956, "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": 0.9938498833277998, "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": 0.9947396695221725, "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": 0.9951807733710953, "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": 0.9971401985232277, "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": 0.9975827659547672, "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": 0.9980459790263252, "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": 0.9986934549064914, "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": 0.999463235719687, "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": 0.9622880494908282, "total_bits": 7093506816, "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": 0.9633838749539695, "total_bits": 7357747968, "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": 0.9687427275225913, "total_bits": 8211542016, "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": 0.9700826459172133, "total_bits": 9215029248, "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": 0.9815833052179139, "total_bits": 10383803136, "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": 0.9830396666044467, "total_bits": 10670075904, "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": 0.9849296054116597, "total_bits": 11472446208, "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": 0.9906698661415201, "total_bits": 13112832768, "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": 0.9914205396229303, "total_bits": 13306195968, "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": 0.9906940519025451, "total_bits": 13501219584, "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": 0.9918402945867887, "total_bits": 13787492352, "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": 0.9952767024453926, "total_bits": 16618636032, "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": 0.9959436360697606, "total_bits": 16904908800, "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": 0.9974797992035747, "total_bits": 19248464640, "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": 0.997580703782947, "total_bits": 19903945728, "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": 0.9979425950962315, "total_bits": 21690719232, "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": 0.9993407646964586, "total_bits": 25478175744, "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": 0.9662591008665531, "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": 0.967334614037291, "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": 0.9713694798809133, "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": 0.976913543172965, "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": 0.9822241535508319, "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": 0.9827315843370008, "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": 0.990239655145591, "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": 0.9905518950943492, "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": 0.9912115776372191, "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": 0.9915748101070916, "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": 0.991504794231763, "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": 0.9920557727792153, "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": 0.9933384782083234, "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": 0.9938742387304572, "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": 0.9963170961403337, "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": 0.9969648100576658, "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": 0.9974616961247337, "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": 0.9983599311166663, "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": 0.9993184164678065, "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": 0.9579026409375825, "total_bits": 7093506816, "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": 0.9591278630847994, "total_bits": 7357747968, "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": 0.9651336534144848, "total_bits": 8211542016, "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": 0.9666253312147761, "total_bits": 9215029248, "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": 0.9794147823888221, "total_bits": 10383803136, "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": 0.9810521867567379, "total_bits": 10670075904, "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": 0.9831755246574941, "total_bits": 11472446208, "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": 0.9895462667397958, "total_bits": 13112832768, "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": 0.9903879681798188, "total_bits": 13306195968, "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": 0.9895790215263045, "total_bits": 13501219584, "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": 0.9908658633204667, "total_bits": 13787492352, "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": 0.9947028417373076, "total_bits": 16618636032, "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": 0.9954556831232223, "total_bits": 16904908800, "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": 0.9971650215815507, "total_bits": 19248464640, "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": 0.9972840085300911, "total_bits": 19903945728, "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": 0.997697169144981, "total_bits": 21690719232, "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": 0.9992544101382679, "total_bits": 25478175744, "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": 0.9670119156668845, "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": 0.9683795294871456, "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": 0.9715489355455104, "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": 0.9773297442221328, "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": 0.9830017803040775, "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": 0.9832941702518024, "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": 0.9907655224792267, "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": 0.9908968351021605, "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": 0.9915343469351923, "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": 0.992037963551028, "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": 0.9918485176136815, "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": 0.9924626313640099, "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": 0.9933646278529379, "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": 0.9939347713891613, "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": 0.996417482803257, "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": 0.9969452234922516, "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": 0.9976037060504926, "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": 0.998323189336684, "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": 0.9993374689059381, "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": 0.9520755606262308, "total_bits": 7093506816, "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": 0.9534276432700848, "total_bits": 7357747968, "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": 0.960368190303837, "total_bits": 8211542016, "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": 0.9620943858141178, "total_bits": 9215029248, "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": 0.9765727469501527, "total_bits": 10383803136, "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": 0.9784272180281972, "total_bits": 10670075904, "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": 0.980875433618693, "total_bits": 11472446208, "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": 0.9881041235241451, "total_bits": 13112832768, "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": 0.9890638057167005, "total_bits": 13306195968, "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": 0.9881503599769387, "total_bits": 13501219584, "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": 0.9896057886561673, "total_bits": 13787492352, "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": 0.9939809909923688, "total_bits": 16618636032, "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": 0.9948282247764597, "total_bits": 16904908800, "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": 0.9967813585504988, "total_bits": 19248464640, "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": 0.996915835145812, "total_bits": 19903945728, "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": 0.9973903365347436, "total_bits": 21690719232, "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": 0.999148175565336, "total_bits": 25478175744, "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": 0.9628652474891982, "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": 0.9650391539754837, "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": 0.9686440067169697, "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": 0.9770047516856146, "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": 0.9812373236851081, "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": 0.9814315357392556, "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": 0.9896836777242195, "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": 0.989892428936927, "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": 0.9905313385993635, "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": 0.9910802079340149, "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": 0.9906140966218357, "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": 0.9912997010212979, "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": 0.9923547012976518, "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": 0.9930223695517174, "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": 0.9958610604333348, "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": 0.9964795975094219, "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": 0.9973052022733578, "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": 0.9981129311092541, "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": 0.9992730684668794, "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": 0.9480698473359409, "total_bits": 7093506816, "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": 0.9495609906551085, "total_bits": 7357747968, "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": 0.9570354347754466, "total_bits": 8211542016, "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": 0.9588695669448689, "total_bits": 9215029248, "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": 0.9747357221908475, "total_bits": 10383803136, "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": 0.9767396702235075, "total_bits": 10670075904, "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": 0.979377192175506, "total_bits": 11472446208, "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": 0.9871735534336614, "total_bits": 13112832768, "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": 0.9882067861770721, "total_bits": 13306195968, "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": 0.9872160963836665, "total_bits": 13501219584, "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": 0.9887929384130985, "total_bits": 13787492352, "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": 0.9935020225024537, "total_bits": 16618636032, "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": 0.9944235565704539, "total_bits": 16904908800, "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": 0.9965223139274473, "total_bits": 19248464640, "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": 0.9966699919804294, "total_bits": 19903945728, "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": 0.9971846005370791, "total_bits": 21690719232, "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": 0.9990847652401211, "total_bits": 25478175744, "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": 0.9589363466948271, "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": 0.9621087481806937, "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": 0.9659702890601597, "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": 0.9750562413644633, "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": 0.9807391243398582, "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": 0.9809664704003617, "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": 0.9889031649861288, "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": 0.9890711692431452, "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": 0.989703601544821, "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": 0.9903176834218597, "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": 0.9902879778834942, "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": 0.9909331760565309, "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": 0.9917128994497225, "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": 0.9923448392900786, "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": 0.9954302484279891, "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": 0.9961077670664772, "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": 0.9970689838212964, "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": 0.997864588234255, "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": 0.9991776506388927, "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": 0.9458863694514883, "total_bits": 7093506816, "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": 0.9473944529891014, "total_bits": 7357747968, "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": 0.9552750222286895, "total_bits": 8211542016, "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": 0.9572565496261967, "total_bits": 9215029248, "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": 0.9734562993980944, "total_bits": 10383803136, "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": 0.9755584703954426, "total_bits": 10670075904, "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": 0.9783624841488505, "total_bits": 11472446208, "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": 0.9864815833218592, "total_bits": 13112832768, "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": 0.9875634903161737, "total_bits": 13306195968, "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": 0.9865639517696476, "total_bits": 13501219584, "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": 0.9882072379052835, "total_bits": 13787492352, "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": 0.9931692618962475, "total_bits": 16618636032, "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": 0.9941298590458342, "total_bits": 16904908800, "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": 0.9963360788165837, "total_bits": 19248464640, "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": 0.9965018197299765, "total_bits": 19903945728, "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": 0.997052383051548, "total_bits": 21690719232, "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": 0.9990293732446031, "total_bits": 25478175744, "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": 0.9564743615980995, "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": 0.9591768436427963, "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": 0.9637337147601341, "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": 0.9726746574436363, "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": 0.9786609854598186, "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": 0.9789566580255172, "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": 0.9876936754996055, "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": 0.9879271346512005, "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": 0.9887803797855189, "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": 0.9894316670558366, "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": 0.9893972186224633, "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": 0.9900473693606296, "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": 0.9910394896655098, "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": 0.9916998886527788, "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": 0.9950968780505814, "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": 0.995840826335265, "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": 0.9968203649071878, "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": 0.9977046770478697, "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": 0.9991295792677096, "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": 0.9463486742032201, "total_bits": 7093506816, "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": 0.9478349685178775, "total_bits": 7357747968, "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": 0.9557660198035209, "total_bits": 8211542016, "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": 0.9577017562757981, "total_bits": 9215029248, "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": 0.9737372195201093, "total_bits": 10383803136, "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": 0.975849361894162, "total_bits": 10670075904, "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": 0.9786008519708718, "total_bits": 11472446208, "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": 0.986669208560335, "total_bits": 13112832768, "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": 0.9877265399738558, "total_bits": 13306195968, "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": 0.9866892056490638, "total_bits": 13501219584, "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": 0.9883356635630327, "total_bits": 13787492352, "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": 0.9932201881402809, "total_bits": 16618636032, "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": 0.9941871639304353, "total_bits": 16904908800, "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": 0.996365184282982, "total_bits": 19248464640, "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": 0.9965194959628494, "total_bits": 19903945728, "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": 0.9970584716617228, "total_bits": 21690719232, "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": 0.9990378211981519, "total_bits": 25478175744, "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": 0.9589185901966534, "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": 0.9613385818604576, "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": 0.9653979469776938, "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": 0.9733745127818302, "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": 0.979485945678071, "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": 0.9796488350992533, "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": 0.9879532293784187, "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": 0.9881334885048043, "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": 0.9890130316029841, "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": 0.9897128918061131, "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": 0.9896844181746832, "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": 0.9903876244115006, "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": 0.9913825487080765, "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": 0.9920906221714655, "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": 0.9953220699252071, "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": 0.9960011971109596, "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": 0.9968744791694917, "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": 0.9977458780508881, "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": 0.9991500600797161, "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": 0.9425105176081783, "total_bits": 7093506816, "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": 0.9440726014344316, "total_bits": 7357747968, "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": 0.9526822581691177, "total_bits": 8211542016, "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": 0.9548283569318683, "total_bits": 9215029248, "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": 0.9717958689373183, "total_bits": 10383803136, "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": 0.9740294501381485, "total_bits": 10670075904, "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": 0.9770925707722965, "total_bits": 11472446208, "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": 0.9856476538177383, "total_bits": 13112832768, "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": 0.9868040986628713, "total_bits": 13306195968, "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": 0.9857094062884387, "total_bits": 13501219584, "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": 0.9874607462828097, "total_bits": 13787492352, "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": 0.9927311233171311, "total_bits": 16618636032, "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": 0.9937548088890157, "total_bits": 16904908800, "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": 0.9961050185313645, "total_bits": 19248464640, "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": 0.9962745930533856, "total_bits": 19903945728, "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": 0.9968847066078237, "total_bits": 21690719232, "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": 0.9989718180915684, "total_bits": 25478175744, "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": 0.9509565522404093, "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": 0.9533285491266533, "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": 0.9593139136406152, "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": 0.9696394952112123, "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": 0.9759020537481105, "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": 0.9762413348071277, "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": 0.9860948357412493, "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": 0.986374348083413, "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": 0.9872690768373248, "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": 0.9879064198313771, "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": 0.9879772168535151, "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": 0.9887062420357803, "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": 0.9900557166333065, "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": 0.9908267801357923, "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": 0.9945787728371981, "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": 0.9953697945702037, "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": 0.9963778652753191, "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": 0.9974479718795536, "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": 0.9989797407378271, "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": 0.9400318888457198, "total_bits": 7093506816, "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": 0.9416533800724306, "total_bits": 7357747968, "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": 0.9507201740910348, "total_bits": 8211542016, "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": 0.9529903544309108, "total_bits": 9215029248, "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": 0.9706984548094241, "total_bits": 10383803136, "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": 0.973004770955365, "total_bits": 10670075904, "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": 0.9762120248357716, "total_bits": 11472446208, "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": 0.9850801810818283, "total_bits": 13112832768, "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": 0.9862664551474154, "total_bits": 13306195968, "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": 0.9851767938142937, "total_bits": 13501219584, "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": 0.9869820114323183, "total_bits": 13787492352, "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": 0.9924676579447758, "total_bits": 16618636032, "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": 0.9935203259443178, "total_bits": 16904908800, "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": 0.9959601595336082, "total_bits": 19248464640, "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": 0.9961456649532345, "total_bits": 19903945728, "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": 0.9967790941805824, "total_bits": 21690719232, "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": 0.9989341639478593, "total_bits": 25478175744, "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": 0.9514403735826674, "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": 0.9542605337735853, "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": 0.9596625476291305, "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": 0.9699621305948025, "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": 0.9759975389430398, "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": 0.9763375332235897, "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": 0.986249762170605, "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": 0.9865248845645079, "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": 0.9874980178122458, "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": 0.9881518118475613, "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": 0.9880900281775546, "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": 0.988882420363983, "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": 0.9901281333234357, "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": 0.9909298422403241, "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": 0.9946120397959787, "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": 0.9954492525981837, "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": 0.9964468048603617, "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": 0.9974590100862674, "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": 0.9990340427841722, "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": 0.9370567677052397, "total_bits": 7093506816, "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": 0.9387302122225887, "total_bits": 7357747968, "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": 0.9482297763032348, "total_bits": 8211542016, "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": 0.9506700960919261, "total_bits": 9215029248, "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": 0.9693728486860269, "total_bits": 10383803136, "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": 0.9717622430701005, "total_bits": 10670075904, "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": 0.975156325626334, "total_bits": 11472446208, "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": 0.9843565703761813, "total_bits": 13112832768, "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": 0.9856156848223978, "total_bits": 13306195968, "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": 0.9845129898994377, "total_bits": 13501219584, "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": 0.9864002350521716, "total_bits": 13787492352, "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": 0.9921294093548664, "total_bits": 16618636032, "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": 0.9932277613613558, "total_bits": 16904908800, "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": 0.9957591121660938, "total_bits": 19248464640, "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": 0.9959793082058528, "total_bits": 19903945728, "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": 0.9966512883547693, "total_bits": 21690719232, "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": 0.9988701582478825, "total_bits": 25478175744, "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": 0.9511657173402215, "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": 0.9534961662108177, "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": 0.9589379678823446, "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": 0.9689071731347787, "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": 0.9753305212545552, "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": 0.9756804061808476, "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": 0.9863152931150245, "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": 0.9866197773892629, "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": 0.9874388513302332, "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": 0.9881370912036417, "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": 0.9877804233079874, "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": 0.9885142695918483, "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": 0.9898797561855692, "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": 0.9907131967085757, "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": 0.9944719444703016, "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": 0.995309612150059, "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": 0.996399076998626, "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": 0.9974588896056, "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": 0.9989849898847751, "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": 0.9384572382428145, "total_bits": 7093506816, "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": 0.9401290903946287, "total_bits": 7357747968, "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": 0.9491998495436028, "total_bits": 8211542016, "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": 0.9514660770563703, "total_bits": 9215029248, "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": 0.9701499812991211, "total_bits": 10383803136, "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": 0.9724857893429304, "total_bits": 10670075904, "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": 0.9757046858338934, "total_bits": 11472446208, "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": 0.9848520896691633, "total_bits": 13112832768, "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": 0.9860517192190807, "total_bits": 13306195968, "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": 0.9849269546225274, "total_bits": 13501219584, "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": 0.9867612603365591, "total_bits": 13787492352, "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": 0.992347357169676, "total_bits": 16618636032, "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": 0.9934134625854265, "total_bits": 16904908800, "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": 0.9959019841442472, "total_bits": 19248464640, "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": 0.9960863532584259, "total_bits": 19903945728, "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": 0.9967150693195627, "total_bits": 21690719232, "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": 0.9989165441672269, "total_bits": 25478175744, "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": 0.9482694739769948, "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": 0.9496131257006997, "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": 0.9553110170619268, "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": 0.9661885762685224, "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": 0.9737150697527748, "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": 0.9740463293304569, "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": 0.9854240608401597, "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": 0.9857600656779189, "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": 0.9867852181686383, "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": 0.9875213467810107, "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": 0.9871468220494295, "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": 0.9880685699049776, "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": 0.989388348424415, "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": 0.9901496458332986, "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": 0.994214087087465, "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": 0.9950988811967698, "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": 0.9962551449057891, "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": 0.9973528621638635, "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": 0.998989792718356, "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": 0.9372060004817813, "total_bits": 7093506816, "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": 0.938912953593229, "total_bits": 7357747968, "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": 0.9480477309736767, "total_bits": 8211542016, "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": 0.9503187301911806, "total_bits": 9215029248, "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": 0.9694943577541333, "total_bits": 10383803136, "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": 0.9718641082413102, "total_bits": 10670075904, "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": 0.9750544040120745, "total_bits": 11472446208, "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": 0.984531556084556, "total_bits": 13112832768, "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": 0.9857445745973995, "total_bits": 13306195968, "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": 0.9845849857910683, "total_bits": 13501219584, "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": 0.9864643615621486, "total_bits": 13787492352, "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": 0.9921755757501447, "total_bits": 16618636032, "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": 0.993266823174628, "total_bits": 16904908800, "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": 0.9958119984397567, "total_bits": 19248464640, "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": 0.9959959947736934, "total_bits": 19903945728, "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": 0.9966200459809778, "total_bits": 21690719232, "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": 0.9988915617700274, "total_bits": 25478175744, "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": 0.9465032156164709, "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": 0.9488164854088896, "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": 0.955533953285531, "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": 0.9669129210769346, "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": 0.973069724969958, "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": 0.9735215523917424, "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": 0.9849066261545216, "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": 0.9853463888560471, "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": 0.9863190258559036, "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": 0.9870119590900446, "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": 0.9867925401777029, "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": 0.987672898501746, "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": 0.9893368715910535, "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": 0.9901153608592913, "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": 0.9941915656979147, "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": 0.9950577143065044, "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": 0.9960841064917316, "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": 0.9972904863534495, "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": 0.9989307516723209, "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": 0.9378711936113081, "total_bits": 7093506816, "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": 0.9395295831521875, "total_bits": 7357747968, "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": 0.9484304504371003, "total_bits": 8211542016, "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": 0.9506907753254238, "total_bits": 9215029248, "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": 0.9696346455694813, "total_bits": 10383803136, "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": 0.9720051312623056, "total_bits": 10670075904, "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": 0.9751454286375328, "total_bits": 11472446208, "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": 0.9845615675868956, "total_bits": 13112832768, "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": 0.9857759608053848, "total_bits": 13306195968, "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": 0.9846461839777859, "total_bits": 13501219584, "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": 0.9865122547520226, "total_bits": 13787492352, "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": 0.9922016450498057, "total_bits": 16618636032, "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": 0.9932826750691196, "total_bits": 16904908800, "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": 0.9958080983794245, "total_bits": 19248464640, "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": 0.9960073521238213, "total_bits": 19903945728, "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": 0.9966209657232914, "total_bits": 21690719232, "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": 0.9988657103125373, "total_bits": 25478175744, "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": 0.9530019240551874, "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": 0.9552443758829644, "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": 0.9611216754603543, "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": 0.9701092793655238, "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": 0.9769281433699163, "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": 0.9771951025744018, "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": 0.9868648717866132, "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": 0.9871973703174215, "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": 0.9880814185415051, "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": 0.988565627902158, "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": 0.9887457352229639, "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": 0.9895028199657405, "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": 0.990877338366485, "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": 0.9915771791833992, "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": 0.9949516207531193, "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": 0.9957280006860137, "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": 0.9965282861008554, "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": 0.9975502159550639, "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": 0.99903013857731, "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": 0.9366084202927978, "total_bits": 7093506816, "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": 0.9383203847039687, "total_bits": 7357747968, "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": 0.9473598416110403, "total_bits": 8211542016, "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": 0.9497066012730724, "total_bits": 9215029248, "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": 0.9692729145876671, "total_bits": 10383803136, "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": 0.9716457539777222, "total_bits": 10670075904, "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": 0.9749011363049871, "total_bits": 11472446208, "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": 0.9843678459966261, "total_bits": 13112832768, "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": 0.9856168104727802, "total_bits": 13306195968, "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": 0.9844917219965473, "total_bits": 13501219584, "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": 0.9863769271852154, "total_bits": 13787492352, "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": 0.9921307383521804, "total_bits": 16618636032, "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": 0.993223868788367, "total_bits": 16904908800, "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": 0.9957721745105166, "total_bits": 19248464640, "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": 0.9959787784793758, "total_bits": 19903945728, "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": 0.9966130682426554, "total_bits": 21690719232, "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": 0.9988665646707983, "total_bits": 25478175744, "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": 0.9548567339385811, "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": 0.9569427824922299, "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": 0.9625597371180591, "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": 0.9719832419466815, "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": 0.9774011267153057, "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": 0.97775442403202, "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": 0.9875322015171772, "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": 0.9878295546789703, "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": 0.9886641848440233, "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": 0.9892715722518531, "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": 0.9889782419869382, "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": 0.9896744767596063, "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": 0.9910925951124611, "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": 0.9918352075068182, "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": 0.9951524884611564, "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": 0.9958746439507721, "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": 0.9967574715356979, "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": 0.99775279193847, "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": 0.9990591806877331, "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": 0.9369259215891361, "total_bits": 7093506816, "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": 0.9385450906855495, "total_bits": 7357747968, "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": 0.9472335564266694, "total_bits": 8211542016, "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": 0.9494960345327854, "total_bits": 9215029248, "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": 0.9691578643886667, "total_bits": 10383803136, "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": 0.9715479653524725, "total_bits": 10670075904, "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": 0.9747234429477861, "total_bits": 11472446208, "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": 0.9843469184186113, "total_bits": 13112832768, "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": 0.9856120857017997, "total_bits": 13306195968, "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": 0.9844289576928866, "total_bits": 13501219584, "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": 0.9863259176184472, "total_bits": 13787492352, "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": 0.9920946960705087, "total_bits": 16618636032, "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": 0.9931972748109776, "total_bits": 16904908800, "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": 0.9957676596241072, "total_bits": 19248464640, "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": 0.9959567458013465, "total_bits": 19903945728, "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": 0.9965715852480284, "total_bits": 21690719232, "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": 0.9988540697285269, "total_bits": 25478175744, "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": 0.9548582684640822, "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": 0.9565701296454981, "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": 0.9629428112193158, "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": 0.9708177789947704, "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": 0.9776193742199164, "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": 0.9781002023404366, "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": 0.987225138974425, "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": 0.9874348269138289, "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": 0.9883001415598157, "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": 0.9888649955636969, "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": 0.9889461082455359, "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": 0.9897825131437888, "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": 0.9911382081673333, "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": 0.9918138500254013, "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": 0.9951922148772466, "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": 0.9959060148654604, "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": 0.9966864685564744, "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": 0.9976655360599245, "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": 0.9990914711180314, "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": 0.9375230130788527, "total_bits": 7093506816, "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": 0.939171228753893, "total_bits": 7357747968, "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": 0.947240618008532, "total_bits": 8211542016, "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": 0.9492741811432337, "total_bits": 9215029248, "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": 0.9693429786221761, "total_bits": 10383803136, "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": 0.9717366369931322, "total_bits": 10670075904, "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": 0.9746006054215526, "total_bits": 11472446208, "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": 0.9845221205859592, "total_bits": 13112832768, "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": 0.9857478218741322, "total_bits": 13306195968, "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": 0.984534799709524, "total_bits": 13501219584, "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": 0.98642235653671, "total_bits": 13787492352, "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": 0.9921690502424577, "total_bits": 16618636032, "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": 0.9932545837823694, "total_bits": 16904908800, "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": 0.9958351385522339, "total_bits": 19248464640, "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": 0.9959946390192368, "total_bits": 19903945728, "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": 0.9965286218242622, "total_bits": 21690719232, "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": 0.9988915876234196, "total_bits": 25478175744, "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": 0.9608023464679718, "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": 0.9625698247140175, "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": 0.9679579775486338, "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": 0.9756014713233239, "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": 0.9804704419680332, "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": 0.9808318996311802, "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": 0.9887384000038238, "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": 0.9891263801898611, "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": 0.9898940670598102, "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": 0.9903213227118709, "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": 0.9903437921679333, "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": 0.9909237675417802, "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": 0.9922630582862583, "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": 0.9928950228189167, "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": 0.9957757713334439, "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": 0.9963793242595306, "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": 0.9970697625111299, "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": 0.9979665540097478, "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": 0.9991652549544693, "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": 0.9385566335955733, "total_bits": 7093506816, "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": 0.9401561313945996, "total_bits": 7357747968, "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": 0.94805034201004, "total_bits": 8211542016, "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": 0.9500378540864116, "total_bits": 9215029248, "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": 0.969747908756529, "total_bits": 10383803136, "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": 0.972116078495195, "total_bits": 10670075904, "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": 0.9749446417156019, "total_bits": 11472446208, "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": 0.984749008386739, "total_bits": 13112832768, "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": 0.9859535040480918, "total_bits": 13306195968, "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": 0.9847451408001545, "total_bits": 13501219584, "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": 0.9866016302502861, "total_bits": 13787492352, "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": 0.992277188545191, "total_bits": 16618636032, "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": 0.9933475996857803, "total_bits": 16904908800, "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": 0.9958981226541495, "total_bits": 19248464640, "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": 0.9960496856978065, "total_bits": 19903945728, "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": 0.9965711207999113, "total_bits": 21690719232, "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": 0.9989079529309253, "total_bits": 25478175744, "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": 0.9671125258564165, "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": 0.968471129375853, "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": 0.9735907077495205, "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": 0.978959363640139, "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": 0.9837888504966701, "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": 0.9840746687627152, "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": 0.9906555611738249, "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": 0.990989057224636, "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": 0.9915791471660333, "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": 0.9918704661418145, "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": 0.9919781574219662, "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": 0.9926443170192406, "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": 0.9937848168376245, "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": 0.9942629867919573, "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": 0.9965403506618091, "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": 0.9970881533073751, "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": 0.9975856055297252, "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": 0.9983735425324228, "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": 0.9992957190659485, "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": 0.9376472752345235, "total_bits": 7093506816, "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": 0.9392712015266481, "total_bits": 7357747968, "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": 0.9472068504087234, "total_bits": 8211542016, "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": 0.9491956465338406, "total_bits": 9215029248, "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": 0.9691516893768781, "total_bits": 10383803136, "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": 0.9715934414906722, "total_bits": 10670075904, "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": 0.9743982905423955, "total_bits": 11472446208, "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": 0.984447703427194, "total_bits": 13112832768, "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": 0.9856797877797171, "total_bits": 13306195968, "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": 0.9844322299682781, "total_bits": 13501219584, "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": 0.9863375432691291, "total_bits": 13787492352, "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": 0.9921178184449673, "total_bits": 16618636032, "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": 0.993215556207456, "total_bits": 16904908800, "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": 0.9958149783595122, "total_bits": 19248464640, "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": 0.9959683113572139, "total_bits": 19903945728, "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": 0.9964848713900306, "total_bits": 21690719232, "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": 0.9988936387235299, "total_bits": 25478175744, "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": 0.9670298771050415, "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": 0.9685107511223147, "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": 0.9728267422730201, "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": 0.9781485284727655, "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": 0.9830276077417167, "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": 0.9835234156466628, "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": 0.9904098345800058, "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": 0.9907603674372168, "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": 0.9913776417900073, "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": 0.9917635622091199, "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": 0.9917231496227416, "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": 0.9923060287133252, "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": 0.9934322770660449, "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": 0.9939944856849156, "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": 0.9964432495057975, "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": 0.9969714526393402, "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": 0.997536855455684, "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": 0.998345204579987, "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": 0.9993082964624351, "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": 0.9388852943911364, "total_bits": 7093506816, "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": 0.9403983541813336, "total_bits": 7357747968, "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": 0.9481458253178158, "total_bits": 8211542016, "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": 0.9501014760646381, "total_bits": 9215029248, "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": 0.9696136894880941, "total_bits": 10383803136, "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": 0.9720061943915329, "total_bits": 10670075904, "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": 0.974794240717433, "total_bits": 11472446208, "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": 0.9846598741500393, "total_bits": 13112832768, "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": 0.9858846570071029, "total_bits": 13306195968, "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": 0.9846520946655226, "total_bits": 13501219584, "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": 0.9865261647409123, "total_bits": 13787492352, "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": 0.992229122080301, "total_bits": 16618636032, "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": 0.993308155650371, "total_bits": 16904908800, "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": 0.9958723491045499, "total_bits": 19248464640, "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": 0.9960260328738705, "total_bits": 19903945728, "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": 0.9965392205815174, "total_bits": 21690719232, "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": 0.9989062283987694, "total_bits": 25478175744, "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": 0.964397800105967, "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": 0.9659834233553786, "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": 0.9706079974574479, "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": 0.9771782926430828, "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": 0.9820928501063272, "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": 0.9824663467803284, "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": 0.989762548038638, "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": 0.9899230425393111, "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": 0.9905982974573577, "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": 0.9910095438459202, "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": 0.9912525629624724, "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": 0.9917060563978004, "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": 0.9928183867718632, "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": 0.9934060162202897, "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": 0.9960734312251014, "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": 0.9966318576817254, "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": 0.9972977154630873, "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": 0.9980929992529318, "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": 0.9992099245228976, "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": 0.9349395794499862, "total_bits": 7093506816, "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": 0.9366305614576528, "total_bits": 7357747968, "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": 0.9446628082935747, "total_bits": 8211542016, "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": 0.9467117573673788, "total_bits": 9215029248, "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": 0.9676527708866879, "total_bits": 10383803136, "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": 0.9702140919275974, "total_bits": 10670075904, "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": 0.9730929415182847, "total_bits": 11472446208, "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": 0.983663550185922, "total_bits": 13112832768, "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": 0.9849710636772215, "total_bits": 13306195968, "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": 0.9836797880646038, "total_bits": 13501219584, "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": 0.9856693586334586, "total_bits": 13787492352, "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": 0.9917388675538333, "total_bits": 16618636032, "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": 0.9928841326312211, "total_bits": 16904908800, "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": 0.9956095274047632, "total_bits": 19248464640, "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": 0.9957786945405563, "total_bits": 19903945728, "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": 0.9963059395698732, "total_bits": 21690719232, "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": 0.9988400889499309, "total_bits": 25478175744, "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": 0.967921022619856, "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": 0.9692596270653763, "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": 0.9738333411024589, "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": 0.978688223581565, "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": 0.9841783135933312, "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": 0.984515060180504, "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": 0.9911351704136714, "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": 0.9914268867700899, "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": 0.9920336014689192, "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": 0.9924528666183745, "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": 0.99235046762777, "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": 0.992998140802803, "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": 0.9940354054867241, "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": 0.9945129403893492, "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": 0.9967095896136016, "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": 0.9972231890025892, "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": 0.9976848851227643, "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": 0.9984689472782377, "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": 0.9993141220643578, "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": 0.9336769051457706, "total_bits": 7093506816, "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": 0.9353972004824563, "total_bits": 7357747968, "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": 0.9432997101623761, "total_bits": 8211542016, "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": 0.94528968700845, "total_bits": 9215029248, "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": 0.9669171534968835, "total_bits": 10383803136, "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": 0.9695615584128782, "total_bits": 10670075904, "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": 0.9723677256781804, "total_bits": 11472446208, "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": 0.9833275124822792, "total_bits": 13112832768, "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": 0.9846617990467501, "total_bits": 13306195968, "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": 0.983305258039189, "total_bits": 13501219584, "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": 0.9853614752523994, "total_bits": 13787492352, "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": 0.9915481793547147, "total_bits": 16618636032, "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": 0.99273338565897, "total_bits": 16904908800, "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": 0.9955213357584158, "total_bits": 19248464640, "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": 0.9956790830474347, "total_bits": 19903945728, "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": 0.9961808240837663, "total_bits": 21690719232, "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": 0.9988233165656167, "total_bits": 25478175744, "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": 0.9618221714504456, "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": 0.9633709922628968, "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": 0.9698955758327716, "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": 0.9756063700800663, "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": 0.9807933820411563, "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": 0.9812672397513923, "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": 0.9887506757911882, "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": 0.9892659209864704, "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": 0.9900641592562591, "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": 0.9905098605851986, "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": 0.9905094889139658, "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": 0.9910741942003369, "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": 0.9925312958657742, "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": 0.9932239183322772, "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": 0.9959201573663833, "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": 0.9965537094364041, "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": 0.9970652074859429, "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": 0.9981256752477126, "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": 0.9991684000415262, "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": 0.934720535909659, "total_bits": 7093506816, "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": 0.9364299742799056, "total_bits": 7357747968, "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": 0.9440396441832969, "total_bits": 8211542016, "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": 0.9459329243553312, "total_bits": 9215029248, "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": 0.967443375161996, "total_bits": 10383803136, "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": 0.9700456399862704, "total_bits": 10670075904, "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": 0.9727011570510896, "total_bits": 11472446208, "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": 0.9835894386235037, "total_bits": 13112832768, "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": 0.9848904452335677, "total_bits": 13306195968, "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": 0.9835749071728634, "total_bits": 13501219584, "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": 0.9855844726001746, "total_bits": 13787492352, "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": 0.9916883857651172, "total_bits": 16618636032, "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": 0.9928477660777342, "total_bits": 16904908800, "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": 0.9955899390195938, "total_bits": 19248464640, "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": 0.9957503643021673, "total_bits": 19903945728, "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": 0.996225166973952, "total_bits": 21690719232, "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": 0.9988436796181639, "total_bits": 25478175744, "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": 0.9633966088294983, "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": 0.9650172508860889, "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": 0.9704605461539406, "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": 0.9770399028141248, "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": 0.9816746195092013, "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": 0.9819282754569462, "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": 0.9894079811822035, "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": 0.9897765618454861, "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": 0.9905075129415644, "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": 0.9909116639755666, "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": 0.9909384504303729, "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": 0.9915575593170759, "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": 0.9928791122633571, "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": 0.993420009280702, "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": 0.996033225514877, "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": 0.9966355461051295, "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": 0.9972087651442149, "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": 0.9981801014448712, "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": 0.9992085380761541, "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": 0.9371845063410307, "total_bits": 7093506816, "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": 0.9388348404318094, "total_bits": 7357747968, "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": 0.9462737075396275, "total_bits": 8211542016, "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": 0.9481479970640257, "total_bits": 9215029248, "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": 0.9686614044598842, "total_bits": 10383803136, "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": 0.97116728381891, "total_bits": 10670075904, "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": 0.9737666950708157, "total_bits": 11472446208, "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": 0.984166847125284, "total_bits": 13112832768, "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": 0.9854066702047068, "total_bits": 13306195968, "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": 0.9841715689797542, "total_bits": 13501219584, "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": 0.9861142614268159, "total_bits": 13787492352, "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": 0.9919879364560505, "total_bits": 16618636032, "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": 0.9931059072627441, "total_bits": 16904908800, "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": 0.9957345570866508, "total_bits": 19248464640, "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": 0.9959000042627418, "total_bits": 19903945728, "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": 0.9963682702509686, "total_bits": 21690719232, "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": 0.9988720117528972, "total_bits": 25478175744, "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": 0.9642174077269278, "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": 0.9664878527584829, "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": 0.9708938571183305, "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": 0.9767862245146381, "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": 0.9826090739745843, "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": 0.9830705117140162, "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": 0.9899787599171855, "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": 0.9903487474657595, "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": 0.9908398360161013, "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": 0.9912973390892148, "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": 0.9914799390458747, "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": 0.9920613956392595, "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": 0.9931442929235729, "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": 0.9937453510876941, "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": 0.9962436785480302, "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": 0.9968172543571869, "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": 0.9973523308106355, "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": 0.9981934990470396, "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": 0.9992273937769871, "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": 0.9335743370220849, "total_bits": 7093506816, "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": 0.9353400628621641, "total_bits": 7357747968, "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": 0.9431152795686534, "total_bits": 8211542016, "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": 0.9450619967752382, "total_bits": 9215029248, "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": 0.9668507627946765, "total_bits": 10383803136, "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": 0.9694845510744735, "total_bits": 10670075904, "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": 0.9721938040793726, "total_bits": 11472446208, "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": 0.9832622873792914, "total_bits": 13112832768, "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": 0.98457225644961, "total_bits": 13306195968, "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": 0.9832594715990126, "total_bits": 13501219584, "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": 0.985310208616092, "total_bits": 13787492352, "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": 0.9915296060253719, "total_bits": 16618636032, "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": 0.9927074104351433, "total_bits": 16904908800, "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": 0.9954965014476329, "total_bits": 19248464640, "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": 0.9956672488478944, "total_bits": 19903945728, "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": 0.9961534142065304, "total_bits": 21690719232, "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": 0.9988137408049385, "total_bits": 25478175744, "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": 0.9592955493809361, "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": 0.9612370340251609, "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": 0.9681219235062599, "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": 0.97530325100218, "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": 0.9795902697859626, "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": 0.9799840645864606, "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": 0.9882138671696579, "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": 0.9886856201889092, "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": 0.9894024654977808, "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": 0.9899039946643537, "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": 0.9899765872151444, "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": 0.9906147575299991, "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": 0.992293253957637, "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": 0.9927760691283957, "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": 0.9957351840444302, "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": 0.9963520235798665, "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": 0.9969329161716527, "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": 0.9980204012980195, "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": 0.9991576626167731, "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": 0.9279329890483304, "total_bits": 7093506816, "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": 0.9298515155127174, "total_bits": 7357747968, "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": 0.9380792925428403, "total_bits": 8211542016, "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": 0.9401301885709951, "total_bits": 9215029248, "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": 0.9640697184950113, "total_bits": 10383803136, "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": 0.9669499593835912, "total_bits": 10670075904, "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": 0.9698138549727829, "total_bits": 11472446208, "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": 0.981861217682691, "total_bits": 13112832768, "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": 0.9833002298286087, "total_bits": 13306195968, "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": 0.9818708580967627, "total_bits": 13501219584, "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": 0.98408945184201, "total_bits": 13787492352, "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": 0.9908260307764929, "total_bits": 16618636032, "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": 0.9921027398300603, "total_bits": 16904908800, "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": 0.9951212795598334, "total_bits": 19248464640, "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": 0.9953100520651788, "total_bits": 19903945728, "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": 0.995825959963871, "total_bits": 21690719232, "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": 0.9987184098204843, "total_bits": 25478175744, "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": 0.9589326166008648, "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": 0.961164979362174, "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": 0.9674062877893448, "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": 0.9744730272183293, "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": 0.9794953397818302, "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": 0.9799771960824728, "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": 0.9878879264242163, "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": 0.9883760750097664, "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": 0.9890886734631893, "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": 0.9895180665005586, "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": 0.9898504169755861, "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": 0.9906267664689374, "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": 0.9921233433445817, "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": 0.9925848073571136, "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": 0.9956398408639392, "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": 0.9962741808451124, "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": 0.9968663791292592, "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": 0.9978914621200314, "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": 0.9990942562489133, "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": 0.9248444432098615, "total_bits": 7093506816, "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": 0.9268629907777435, "total_bits": 7357747968, "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": 0.9353764751239827, "total_bits": 8211542016, "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": 0.9375389083043525, "total_bits": 9215029248, "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": 0.9624190630489274, "total_bits": 10383803136, "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": 0.9654509875629294, "total_bits": 10670075904, "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": 0.9684480996802449, "total_bits": 11472446208, "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": 0.9809880433604121, "total_bits": 13112832768, "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": 0.9825130597639241, "total_bits": 13306195968, "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": 0.9810091515414809, "total_bits": 13501219584, "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": 0.9833542110251361, "total_bits": 13787492352, "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": 0.9903770618777918, "total_bits": 16618636032, "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": 0.9917327183985004, "total_bits": 16904908800, "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": 0.9948760467875553, "total_bits": 19248464640, "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": 0.9950765735028606, "total_bits": 19903945728, "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": 0.9956298793641556, "total_bits": 21690719232, "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": 0.9986509155592342, "total_bits": 25478175744, "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": 0.9536738031984944, "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": 0.955292024110493, "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": 0.9627347824800956, "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": 0.9718480795425805, "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": 0.9767233560744085, "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": 0.977070155112367, "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": 0.9867083858943692, "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": 0.9871328800034366, "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": 0.9879941109714931, "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": 0.9885417234005505, "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": 0.9885172067749265, "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": 0.9892267779398122, "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": 0.990814798928209, "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": 0.9915278294417811, "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": 0.994980683219374, "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": 0.9957362292702064, "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": 0.996540244564561, "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": 0.9976913269645976, "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": 0.9990150652838087, "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": 0.9140995070338249, "total_bits": 7093506816, "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": 0.9164593758943834, "total_bits": 7357747968, "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": 0.9258942096249053, "total_bits": 8211542016, "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": 0.9284487000029338, "total_bits": 9215029248, "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": 0.957816799220286, "total_bits": 10383803136, "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": 0.9611973374297744, "total_bits": 10670075904, "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": 0.9644560011005715, "total_bits": 11472446208, "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": 0.9783021086140683, "total_bits": 13112832768, "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": 0.9801033150992895, "total_bits": 13306195968, "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": 0.9786317246524912, "total_bits": 13501219584, "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": 0.9812986695844876, "total_bits": 13787492352, "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": 0.9891460785936368, "total_bits": 16618636032, "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": 0.9906857912428677, "total_bits": 16904908800, "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": 0.9941184829650936, "total_bits": 19248464640, "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": 0.9944531969897645, "total_bits": 19903945728, "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": 0.995097328884233, "total_bits": 21690719232, "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": 0.9984129667067655, "total_bits": 25478175744, "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": 0.9397781444223303, "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": 0.9417588881363994, "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": 0.9504411177415597, "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": 0.9604533158830906, "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": 0.9696531604583326, "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": 0.97026406071688, "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": 0.9829324281921512, "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": 0.9831970975195107, "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": 0.9846106124061503, "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": 0.98545937803819, "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": 0.9853440216301304, "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": 0.9863208723009417, "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": 0.988246536495066, "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": 0.9892112023785318, "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": 0.9935226564442641, "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": 0.9945328718198365, "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": 0.9955355283853254, "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": 0.9970262414886077, "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": 0.998735527792872, "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": 0.9126170779529371, "total_bits": 7093506816, "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": 0.9149987783871198, "total_bits": 7357747968, "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": 0.924614464373965, "total_bits": 8211542016, "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": 0.9270763365845931, "total_bits": 9215029248, "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": 0.9567279229430776, "total_bits": 10383803136, "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": 0.9601882986332241, "total_bits": 10670075904, "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": 0.9635214600712061, "total_bits": 11472446208, "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": 0.9780645592040137, "total_bits": 13112832768, "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": 0.979784318981202, "total_bits": 13306195968, "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": 0.9781379653631073, "total_bits": 13501219584, "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": 0.9808394239333115, "total_bits": 13787492352, "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": 0.9889222847199753, "total_bits": 16618636032, "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": 0.9904750008136034, "total_bits": 16904908800, "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": 0.994077654003999, "total_bits": 19248464640, "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": 0.994330335200127, "total_bits": 19903945728, "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": 0.99494942503148, "total_bits": 21690719232, "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": 0.9984354302409644, "total_bits": 25478175744, "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": 0.9463129037697064, "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": 0.9498804776292098, "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": 0.9569748080286541, "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": 0.9663097977050041, "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": 0.9729950146651581, "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": 0.9735247900611476, "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": 0.984484249431836, "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": 0.9850177908022153, "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": 0.9861582681340607, "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": 0.9867848371106543, "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": 0.9865669293427154, "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": 0.987521539981428, "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": 0.9893518158871877, "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": 0.9902304255060459, "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": 0.9941214787479686, "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": 0.9950726233104146, "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": 0.9959537429538997, "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": 0.997333654110614, "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": 0.9988892913679592, "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": 0.8992744680298002, "total_bits": 7093506816, "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": 0.903073870822003, "total_bits": 7357747968, "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": 0.9127782818518186, "total_bits": 8211542016, "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": 0.9156479398278814, "total_bits": 9215029248, "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": 0.9516523123571747, "total_bits": 10383803136, "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": 0.9558969053782915, "total_bits": 10670075904, "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": 0.9597928319126368, "total_bits": 11472446208, "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": 0.9753303652335155, "total_bits": 13112832768, "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": 0.9774649993663556, "total_bits": 13306195968, "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": 0.9757277108728886, "total_bits": 13501219584, "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": 0.9787911559504113, "total_bits": 13787492352, "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": 0.9876263859975887, "total_bits": 16618636032, "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": 0.9894083060177141, "total_bits": 16904908800, "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": 0.9933384570146078, "total_bits": 19248464640, "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": 0.9936932929252323, "total_bits": 19903945728, "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": 0.9944640513950664, "total_bits": 21690719232, "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": 0.9982105189027559, "total_bits": 25478175744, "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": 0.9102760196516388, "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": 0.9143611735811359, "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": 0.9312123566081649, "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": 0.9486354446332705, "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": 0.9541842212998554, "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": 0.9559579216139882, "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": 0.9737800505796546, "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": 0.9753927331613866, "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": 0.9777112084587938, "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": 0.9785143001691291, "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": 0.9779113707083621, "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": 0.9797714109483519, "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": 0.983448976846902, "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": 0.9850262387312556, "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": 0.9907598198440514, "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": 0.992413669260905, "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": 0.9931422688164994, "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": 0.9963097820965279, "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": 0.9981809005962292, "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": 0.8896537525089163, "total_bits": 7093506816, "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": 0.8929342023636165, "total_bits": 7357747968, "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": 0.903871973682391, "total_bits": 8211542016, "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": 0.9069213110365366, "total_bits": 9215029248, "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": 0.944825464369435, "total_bits": 10383803136, "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": 0.9495417357079292, "total_bits": 10670075904, "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": 0.9536783553071713, "total_bits": 11472446208, "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": 0.9718937591502541, "total_bits": 13112832768, "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": 0.9741466969839836, "total_bits": 13306195968, "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": 0.9720148018708354, "total_bits": 13501219584, "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": 0.9756134158294452, "total_bits": 13787492352, "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": 0.9857806719741539, "total_bits": 16618636032, "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": 0.9878497294052259, "total_bits": 16904908800, "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": 0.9923315662890673, "total_bits": 19248464640, "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": 0.99270670922277, "total_bits": 19903945728, "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": 0.9935213166877235, "total_bits": 21690719232, "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": 0.9979357063512955, "total_bits": 25478175744, "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": 0.9437755358062292, "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": 0.9483897503661483, "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": 0.9632578739210179, "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": 0.9692306608745926, "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": 0.9719048845336625, "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": 0.9732126791618372, "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": 0.9842365113037982, "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": 0.9853218824259544, "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": 0.9865379475155159, "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": 0.986784349685829, "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": 0.9866106892004609, "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": 0.9878134335832376, "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": 0.9914753367438128, "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": 0.9923009164141197, "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": 0.9952605395259237, "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": 0.996128760180191, "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": 0.9959153781865576, "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": 0.9984292346175367, "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": 0.9988893991471023, "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": 0.96013921763944, "total_bits": 7093506816, "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": 0.9625488336345083, "total_bits": 7357747968, "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": 0.966935324041467, "total_bits": 8211542016, "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": 0.9683279569603895, "total_bits": 9215029248, "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": 0.9815945075824857, "total_bits": 10383803136, "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": 0.9832863951297967, "total_bits": 10670075904, "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": 0.9850843286043719, "total_bits": 11472446208, "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": 0.990117408139141, "total_bits": 13112832768, "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": 0.9910194701270053, "total_bits": 13306195968, "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": 0.9906085013951126, "total_bits": 13501219584, "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": 0.9919182431246889, "total_bits": 13787492352, "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": 0.9951412564903301, "total_bits": 16618636032, "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": 0.99590376187361, "total_bits": 16904908800, "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": 0.9971714653074741, "total_bits": 19248464640, "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": 0.9974402289126852, "total_bits": 19903945728, "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": 0.9978378148087742, "total_bits": 21690719232, "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": 0.9990494702360593, "total_bits": 25478175744, "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 }, "strategy": { "model.layers.0.self_attn": { "accuracy": 0.9991672076223614, "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": 0.9982824173363808, "total_bits": 16904908800, "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 } }, "model.layers.1.self_attn": { "accuracy": 0.9979652358329316, "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": 0.9984350846988443, "total_bits": 21690719232, "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 } }, "model.layers.2.self_attn": { "accuracy": 0.9981063970909944, "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": 0.99825886362171, "total_bits": 13112832768, "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 } }, "model.layers.3.self_attn": { "accuracy": 0.999724811672135, "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 } }, "model.layers.3.block_sparse_moe": { "accuracy": 0.9967695052282101, "total_bits": 11472446208, "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 } }, "model.layers.4.self_attn": { "accuracy": 0.9995401424086211, "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": 0.9974311009803052, "total_bits": 13787492352, "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 } }, "model.layers.5.self_attn": { "accuracy": 0.9996650030968196, "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": 0.9982842649899938, "total_bits": 16904908800, "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 } }, "model.layers.6.self_attn": { "accuracy": 0.9996325979583688, "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": 0.9978057868092468, "total_bits": 16904908800, "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 } }, "model.layers.7.self_attn": { "accuracy": 0.99962733712585, "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": 0.9974314317591243, "total_bits": 16904908800, "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 } }, "model.layers.8.self_attn": { "accuracy": 0.9995185340724818, "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": 0.9969057361595333, "total_bits": 16904908800, "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 } }, "model.layers.9.self_attn": { "accuracy": 0.9994986603537415, "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": 0.9976441569344484, "total_bits": 19248464640, "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 } }, "model.layers.10.self_attn": { "accuracy": 0.999463235719687, "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": 0.9974797992035747, "total_bits": 19248464640, "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 } }, "model.layers.11.self_attn": { "accuracy": 0.9993184164678065, "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": 0.9971650215815507, "total_bits": 19248464640, "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 } }, "model.layers.12.self_attn": { "accuracy": 0.9993374689059381, "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": 0.9967813585504988, "total_bits": 19248464640, "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 } }, "model.layers.13.self_attn": { "accuracy": 0.9992730684668794, "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": 0.9971846005370791, "total_bits": 21690719232, "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 } }, "model.layers.14.self_attn": { "accuracy": 0.9991776506388927, "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": 0.997052383051548, "total_bits": 21690719232, "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 } }, "model.layers.15.self_attn": { "accuracy": 0.9991295792677096, "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": 0.9970584716617228, "total_bits": 21690719232, "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 } }, "model.layers.16.self_attn": { "accuracy": 0.9991500600797161, "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": 0.9968847066078237, "total_bits": 21690719232, "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 } }, "model.layers.17.self_attn": { "accuracy": 0.9989797407378271, "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": 0.9967790941805824, "total_bits": 21690719232, "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 } }, "model.layers.18.self_attn": { "accuracy": 0.9990340427841722, "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": 0.9988701582478825, "total_bits": 25478175744, "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": 0.9989849898847751, "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": 0.9989165441672269, "total_bits": 25478175744, "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": 0.998989792718356, "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": 0.9988915617700274, "total_bits": 25478175744, "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": 0.9989307516723209, "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": 0.9988657103125373, "total_bits": 25478175744, "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": 0.99903013857731, "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": 0.9988665646707983, "total_bits": 25478175744, "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": 0.9990591806877331, "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": 0.9988540697285269, "total_bits": 25478175744, "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": 0.9990914711180314, "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": 0.9988915876234196, "total_bits": 25478175744, "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": 0.9991652549544693, "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": 0.9989079529309253, "total_bits": 25478175744, "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": 0.9992957190659485, "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": 0.9988936387235299, "total_bits": 25478175744, "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": 0.9993082964624351, "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": 0.9989062283987694, "total_bits": 25478175744, "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": 0.9992099245228976, "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": 0.9988400889499309, "total_bits": 25478175744, "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": 0.9993141220643578, "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": 0.9988233165656167, "total_bits": 25478175744, "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": 0.9991684000415262, "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": 0.9988436796181639, "total_bits": 25478175744, "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": 0.9992085380761541, "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": 0.9988720117528972, "total_bits": 25478175744, "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": 0.9992273937769871, "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": 0.9988137408049385, "total_bits": 25478175744, "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": 0.9991576626167731, "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": 0.9987184098204843, "total_bits": 25478175744, "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": 0.9990942562489133, "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": 0.9986509155592342, "total_bits": 25478175744, "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": 0.9990150652838087, "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": 0.9984129667067655, "total_bits": 25478175744, "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": 0.998735527792872, "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": 0.9984354302409644, "total_bits": 25478175744, "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": 0.9988892913679592, "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": 0.9982105189027559, "total_bits": 25478175744, "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": 0.9981809005962292, "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": 0.9979357063512955, "total_bits": 25478175744, "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": 0.9988893991471023, "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": 0.9971714653074741, "total_bits": 19248464640, "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 } } }, "q_last_module_idx": 82 }