diff --git "a/quant_strategy.json" "b/quant_strategy.json" new file mode 100644--- /dev/null +++ "b/quant_strategy.json" @@ -0,0 +1,4228 @@ +{ + "measurement": { + "model.layers.0": { + "accuracy": 0.9235353469848633, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.1": { + "accuracy": 0.9165844917297363, + "total_bits": 486917664, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.2": { + "accuracy": 0.8836932182312012, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.3": { + "accuracy": 0.9150410890579224, + "total_bits": 449761824, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.4": { + "accuracy": 0.9069223403930664, + "total_bits": 482790432, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.5": { + "accuracy": 0.9044737815856934, + "total_bits": 482790432, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.6": { + "accuracy": 0.9058353900909424, + "total_bits": 486917664, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.7": { + "accuracy": 0.9102394580841064, + "total_bits": 519946272, + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.8": { + "accuracy": 0.9279017448425293, + "total_bits": 449761824, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.9": { + "accuracy": 0.9575139284133911, + "total_bits": 360997440, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.10": { + "accuracy": 0.95622718334198, + "total_bits": 360997440, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.11": { + "accuracy": 0.949416995048523, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.12": { + "accuracy": 0.9479968547821045, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.13": { + "accuracy": 0.9440830945968628, + "total_bits": 360997440, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.14": { + "accuracy": 0.9474369287490845, + "total_bits": 394026048, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.15": { + "accuracy": 0.9413189888000488, + "total_bits": 360997440, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.16": { + "accuracy": 0.944807767868042, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.17": { + "accuracy": 0.9402191638946533, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.18": { + "accuracy": 0.9468686580657959, + "total_bits": 394026048, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.19": { + "accuracy": 0.9472098350524902, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.20": { + "accuracy": 0.9527584314346313, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.21": { + "accuracy": 0.9607383012771606, + "total_bits": 394026048, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.22": { + "accuracy": 0.9615957736968994, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.23": { + "accuracy": 0.9649553298950195, + "total_bits": 449761824, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.24": { + "accuracy": 0.963486909866333, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.25": { + "accuracy": 0.9628657102584839, + "total_bits": 365124672, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.26": { + "accuracy": 0.973232626914978, + "total_bits": 486917664, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.27": { + "accuracy": 0.9689710140228271, + "total_bits": 486917664, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.28": { + "accuracy": 0.969882607460022, + "total_bits": 519946272, + "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 + }, + "up_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.29": { + "accuracy": 0.9750123620033264, + "total_bits": 604586016, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.30": { + "accuracy": 0.9754680395126343, + "total_bits": 604586016, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.31": { + "accuracy": 0.9848775267601013, + "total_bits": 664451616, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.32": { + "accuracy": 0.9839056134223938, + "total_bits": 664451616, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.33": { + "accuracy": 0.981873095035553, + "total_bits": 664451616, + "q_proj": { + "group_size": { + "2": 64 + }, + "bits": [ + 2 + ], + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.34": { + "accuracy": 0.9870665073394775, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.35": { + "accuracy": 0.9864365458488464, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.36": { + "accuracy": 0.9846726059913635, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.37": { + "accuracy": 0.9842717051506042, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.38": { + "accuracy": 0.9814689755439758, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.39": { + "accuracy": 0.980954110622406, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.40": { + "accuracy": 0.9813097715377808, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.41": { + "accuracy": 0.9818326830863953, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.42": { + "accuracy": 0.981928288936615, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.43": { + "accuracy": 0.9821763634681702, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.44": { + "accuracy": 0.9825348258018494, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.45": { + "accuracy": 0.9833475351333618, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.46": { + "accuracy": 0.9827241897583008, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.47": { + "accuracy": 0.926713228225708, + "total_bits": 697480224, + "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 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + } +} \ No newline at end of file