diff --git "a/measurement.json" "b/measurement.json" new file mode 100644--- /dev/null +++ "b/measurement.json" @@ -0,0 +1,62439 @@ +{ + "last_module_idx": 66, + "measurement": { + "lm_head.linear": null, + "model.layers.0.mlp": [ + { + "accuracy": 0.9285262258429277, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9303554359235262, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9416553723184686, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9520253068522403, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9647851680454455, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9671571599809747, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9750262640024486, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9777331462031916, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.980225312082391, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9820944321782965, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9841188443334479, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9907446409526625, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9919810130407936, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940162171658716, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.995440677200493, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997689467511679, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981327100019706, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.0.self_attn": [ + { + "accuracy": 0.8811917304992676, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.8879667708748266, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.8904559361307245, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9395275304191991, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.944050061075311, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9441651921523244, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9672367039479708, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.967439121321628, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9692750479045668, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9713402515963504, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9713598176052696, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9726411323798331, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9730142229481747, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9747795446922904, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9854298877088647, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9873481965378711, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913642124125832, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929743166032591, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976946807613498, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.1.mlp": [ + { + "accuracy": 0.9371567023427863, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9379960173054745, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9535687973624781, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9603903764172604, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.967907494620273, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9703595261824759, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9780855570968828, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9822292814129278, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9841477714086834, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9835293465539029, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9854994133899087, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9914999008178711, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9927130412114294, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9952140383814511, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956918794073557, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975697794242909, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985823336204416, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.1.self_attn": [ + { + "accuracy": 0.951452845021298, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9531696972094084, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9555412875978571, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9753530339190835, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9766138980263158, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9767048939278251, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.98717584343333, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9872850799246838, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9880189025088361, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9887843798649939, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9880676802835966, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.988645115965291, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9890749430970142, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9898261749430707, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940573428022234, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948712156007165, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965885501158865, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972981643912039, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.999049944116881, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.10.mlp": [ + { + "accuracy": 0.9780638955141369, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9784857458189914, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9817496083284679, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9828705379837438, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9886911370252308, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9895615444371575, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9910889032639956, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940614594440711, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9946032762527466, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9942084017552828, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.994900370114728, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970292219598043, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997437278691091, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983688911521121, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984688770614172, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987958773578468, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994959193526914, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.10.self_attn": [ + { + "accuracy": 0.98500247691807, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9857645489667591, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9874600760246578, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9897774842224623, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9924737175828532, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9927272796630859, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951215245221791, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.995371005252788, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958089237150393, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960917563814866, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962942078709602, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965983707653848, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9968981495813319, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971884936094284, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982342372991537, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985638602979874, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9986451294665274, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.99915353846001, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.999610710840084, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.11.mlp": [ + { + "accuracy": 0.9826417894739854, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.98295627299108, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.985677863422193, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9865858915605044, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9910624725253958, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917535358353665, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992987479034223, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9953164493567065, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957522228360176, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9954305263726335, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959808701747342, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976595763705278, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979821667075157, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987215577183586, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987976993189046, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990628150928962, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996129165795681, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.11.self_attn": [ + { + "accuracy": 0.9862002918594762, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9868625512248591, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.988885655214912, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9918929691377439, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9931162815344962, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933045843714162, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958274140954018, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960384674762425, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963571688062266, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965533069089839, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996596408909873, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9968049879136839, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971595014396467, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974151679167622, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984542259848431, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9986812911535564, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988909263752008, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992861300706863, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996722159220984, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.12.mlp": [ + { + "accuracy": 0.9838262639547649, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.98414115372457, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9866523915215543, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9875136049170243, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916628105075735, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923087632969806, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9934690626044023, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956200483598208, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960236439579412, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957374067682969, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996251649370319, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978151060640812, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981179129528371, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988003745675087, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988756410189366, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991240915107099, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996311764085763, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.12.self_attn": [ + { + "accuracy": 0.9870909170100564, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9879833114774603, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9900415069178531, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9926875547358864, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9937280114544066, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9938767430813689, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959745269856954, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996267563810474, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996640314397059, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9968013361488518, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969157262852317, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970756973090925, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974580830649326, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976796831347441, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985995920080888, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988210495365294, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9989427298699555, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993626806198767, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996922908007706, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.13.mlp": [ + { + "accuracy": 0.9853001456511649, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9855667525216153, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9879760248096365, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9888007429085279, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9924077148500242, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929931618665394, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9941054245359019, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960162051414189, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963912265865427, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961238225039682, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965920750247804, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980184912289444, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982913805073813, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9989181808931263, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9989816957202396, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992184151748293, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996720551709203, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.13.self_attn": [ + { + "accuracy": 0.9915127205221277, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9920944195044669, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936992008435098, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947825003611414, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955862517419615, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958348348736763, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969830546332034, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997231837362051, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997550230277212, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976075889640733, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997878009159314, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998084661403769, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982811661535188, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984502706088518, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990232064339676, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992290068123686, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991838337461415, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996071079696872, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997572418008196, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.14.mlp": [ + { + "accuracy": 0.9852072518122824, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9854693585320523, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.987940796896031, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.98878833337834, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923546447565681, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929411623038744, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940875835324589, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959901618330103, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.99637358165101, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996099799087173, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965718623838926, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980077780783176, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982836787637911, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998915230169108, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9989794915247905, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992205019256002, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996766198034349, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.14.self_attn": [ + { + "accuracy": 0.9926262741026125, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9930055251247004, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948758882911581, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956808839189378, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.99613669240161, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963565296248386, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972067918432387, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975057814858461, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977521692451677, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977823658601234, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981415332540086, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983310603389615, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985305002253306, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.99871602428979, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991542634211088, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993447890799296, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992467435567003, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.999654718405126, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997162582891944, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.15.mlp": [ + { + "accuracy": 0.9845446097223383, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9848173166576185, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.987484128067368, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9884068479663447, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9919934508047605, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9926133014653858, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9938508915273767, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958008472856722, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962081709190419, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959209702516857, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964177882985065, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979191142085352, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982091535470987, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988697580993176, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9989368177362179, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991974857096609, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996684776913178, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.15.self_attn": [ + { + "accuracy": 0.9899762947308389, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9905106738993996, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9920184094654886, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9939926164714914, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950170258158132, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9952227159550315, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996776961378361, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970194752279081, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972465357890254, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973534537773383, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975890305481458, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977185847727876, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979902007862141, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981558797390837, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988652578506031, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990520652775702, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991266894105234, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994787720375156, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997327327483186, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.16.mlp": [ + { + "accuracy": 0.9834930394825182, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9837719233412492, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9866283284990411, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9876297908393961, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9914351191959883, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9920940399169922, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9934272538674506, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.995502833473055, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959412656332317, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.995635478512237, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961657324119618, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977744058950951, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980847796327189, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987940205947349, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988649204177292, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991477616131306, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996544300137382, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.16.self_attn": [ + { + "accuracy": 0.991690465494206, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9921250272738306, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936335471115614, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9944298283049935, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957101886209688, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959898787109476, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.99693317927028, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972465873548859, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973559442319369, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974096964456534, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997937330093823, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980758526607564, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983744073267046, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984419693876254, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990007256794917, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992290289190254, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991055417217707, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996486606419479, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997332669353407, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.17.mlp": [ + { + "accuracy": 0.9827461525013572, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9830379674309179, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9860270901730186, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9870707761300238, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9910418179474378, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917357022825041, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9931285655812213, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9953027159759873, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957613792074355, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9954370311216304, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959940941710221, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976737132590068, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979994673477975, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987404342544707, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998813560330554, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991066636223542, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.999638811489077, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.17.self_attn": [ + { + "accuracy": 0.9914587822399641, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.991827883218464, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9934396496728847, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950712609447931, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956273377725953, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957236016267225, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972106520282594, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997388838937408, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975145733670184, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997694016875405, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979022617794966, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980142255754847, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998292672987047, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984340160889061, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990566145432623, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992032650564062, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992788123260987, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9995944180623874, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997785846016517, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.18.mlp": [ + { + "accuracy": 0.9806739797717646, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.98100346326828, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.984310098384556, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9854816323832462, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9899631307313317, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.990730524847382, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922904262417241, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.994736901239345, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9952489910157103, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948964275811848, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955114526183981, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974036367708131, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977621451804513, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985972725247082, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9986792804771348, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990035823889469, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996032707001034, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.18.self_attn": [ + { + "accuracy": 0.9927167586590114, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9930201423795599, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9946472052680818, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.995370543316791, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958825974087966, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962955266237259, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970799125730991, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975092646322752, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975808345173535, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976786906389814, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980690808672654, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981668936578851, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984556266939953, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985657963705691, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990214261373407, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992630854249, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991126863383933, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996798159309516, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997463991228295, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.19.mlp": [ + { + "accuracy": 0.9787639144219851, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9791297708687029, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9826965535941877, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9839639710752588, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9889609813690186, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9898044608141247, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9914930611848831, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9942044400070843, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947720648426759, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9943829308214941, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950592623729455, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971417769005424, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975360918201899, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984549134969711, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985445076109547, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988999264804941, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9995650617209705, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.19.self_attn": [ + { + "accuracy": 0.9915982108367117, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9921546601935437, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940296733065656, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9946645172803026, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955388680100441, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958962770669084, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967010272176642, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971248227122583, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997384635437476, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974575874052549, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978850601535094, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998059023759867, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983946210459659, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984832033514977, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990135965574729, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992278156017786, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990945703497058, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996360436964192, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997433781084654, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.2.mlp": [ + { + "accuracy": 0.9384229245938753, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9395429774334556, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9501308987015172, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9540043692839772, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9690210097714475, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9713947333787617, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9764502173975894, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9835170774083388, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.985120205502761, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.984137259031597, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9860186639584994, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9918428384944012, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992980422942262, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955259698786234, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958369429958495, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970055779344157, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9986975151849421, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.2.self_attn": [ + { + "accuracy": 0.957600198293987, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9590180610355578, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9628056695586756, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9764186567381808, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9789879902413017, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9791487267142848, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.988479985218299, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9886459253336254, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9893560840895301, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9900148550146505, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.989323294476459, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.989904017040604, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9906980787452898, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.991377843837989, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9949396570262156, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956562789647203, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969730963440318, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977572121118244, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991642145538017, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.20.mlp": [ + { + "accuracy": 0.9760888469846625, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9764982634469083, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9804707624410328, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9818772093245858, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9875709783089789, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9885131732413643, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.990405850504574, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9934882392224512, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9941167651038421, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936847169148294, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.994441516305271, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967942590776243, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972332306206226, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982720386040839, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983729433856512, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987600111451588, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9995220182953697, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.20.self_attn": [ + { + "accuracy": 0.9887156894332484, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9893835626150432, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9909252994938901, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933112616601744, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9943235842020888, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9944609807510125, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964954315831787, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9966742208129481, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969349099617255, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970784706896857, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971997877092738, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973307953853356, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976653915486837, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978438076612196, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987136899052482, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9989029552395406, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990629946911022, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994274000882318, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997283097258524, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.21.mlp": [ + { + "accuracy": 0.9728701177396273, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9733603816283377, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9777776630301225, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9793421202584317, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9858982170882978, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9869754502647802, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9890792809034649, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9926095761750874, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933245354577115, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992833369656613, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936964166791815, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963623367642102, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9968634386988062, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980399765466389, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981539959186002, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985890695335049, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994635239831711, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.21.self_attn": [ + { + "accuracy": 0.9897683011858087, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9905118079561936, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923392534255981, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940062620137867, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9949047573302922, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950335029708711, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996623627057201, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967883103772214, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969754989602064, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970846644749767, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974378539543403, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976111268134493, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979528649465034, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980942981415674, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988621045884333, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990359247711144, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990944948635603, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994784681812713, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9997237399465552, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.22.mlp": [ + { + "accuracy": 0.9698815283022428, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9704339566983675, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9752125112633956, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9769130615811599, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.98430587743458, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9855075447182906, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.987793325593597, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917657500819156, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9925679258610073, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9920140457780737, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929803609848022, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959372266342765, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965031170531323, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978106647337738, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979394237069708, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984207854263092, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994010049359578, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.22.self_attn": [ + { + "accuracy": 0.9852104390922346, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9859252430890736, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9877135792845174, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9909257253533915, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992703338986949, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9928628237623918, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956690920026678, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958364845890748, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960949111141657, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962601344052114, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963764116952294, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996525490754529, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9968928145734888, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971087110277853, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982919747892179, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998540815946303, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988275923227009, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992521340027452, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996686430255833, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.23.mlp": [ + { + "accuracy": 0.9652667516156247, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9659114385906019, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9712924863162794, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9731990255807575, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9819107745823107, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9832761883735657, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9858679598883578, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9905197839987906, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9914326259964391, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9908026848968706, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.991902580386714, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.995333223358581, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959713589203986, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974899231211135, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976355684430975, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981681865296865, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993205925841865, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.23.self_attn": [ + { + "accuracy": 0.9822858713175121, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9831445154390837, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9849233062643754, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9887664412197313, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9912359604710027, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913531204587535, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947927253026712, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9949606817803884, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9952897531421561, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955381833409008, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956230744719505, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958990498592979, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962544433380428, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965264314883634, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979797389946485, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982544119028669, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9986260541175541, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991000582905192, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.999612861018824, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.24.mlp": [ + { + "accuracy": 0.9607264901462355, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9614852823709187, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9673323192094502, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9694072353212457, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9795278075494265, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9810825037328821, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9839033766796714, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.989259419472594, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9902891014751635, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9895855784416199, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9908313257129568, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947108112667736, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9954370722958916, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971515936286826, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973173949279284, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979055880715972, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.999227140637997, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.24.self_attn": [ + { + "accuracy": 0.9803436273022702, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9811860495492032, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9830883898233113, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9875845822848772, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9903922355488727, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9905010438279102, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9944590051707468, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9945537730267173, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948640861793568, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951012997250808, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951506617822146, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9954310039940634, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957919818790335, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960863115756136, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977318650405658, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980304550967718, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985326522667157, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998983967559118, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9995928472643228, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.25.mlp": [ + { + "accuracy": 0.9558308657846952, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9567071199417114, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9630715752902784, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9653166407033017, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9769585509049264, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9787028024071142, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9817812631004735, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9879019456474405, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9890546367356652, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9882613663610659, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9896660705930308, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940397845287072, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948551588152584, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967876847245192, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969800365598578, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976246784392157, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991353988451394, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.25.self_attn": [ + { + "accuracy": 0.9835022358517898, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9842289545034107, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.985955181874727, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9891960714992724, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917702557224977, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9919131092335048, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950845496434915, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951766274477306, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955056552824221, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.995772239408995, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958609428844953, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960949734637612, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964222892334587, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9966800040320346, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980660416185856, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983130115035334, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9986744410309353, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990955556889898, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9996252300609884, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.26.mlp": [ + { + "accuracy": 0.9495238849991247, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9505400845879003, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9576337808056882, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9601174467488339, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.973600309146078, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9756317452380532, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.979041123076489, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9860847451184925, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9874227611642135, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.986505306080768, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.988127495113172, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9931196076305289, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940771789927232, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962830723900544, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965062337486368, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997251627084456, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9989972078290424, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.26.self_attn": [ + { + "accuracy": 0.9726053131254095, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9736760446899816, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9765300938957616, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9822047635128623, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9863629827373906, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9865300357341766, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9921103700211173, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922380510129427, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992813306419473, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933510642302664, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9932252843129007, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936593582755641, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9942849886260534, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9946728125214577, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969124821455855, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973369676031565, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980059395495214, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998607107585198, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994519348501375, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.27.mlp": [ + { + "accuracy": 0.9398523318140131, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9410988029680754, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9492736647003576, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9521197017870451, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9685502805207905, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9709579536789342, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9748952467190591, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9834578711735574, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9850178015859503, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9839507529610082, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9858620903993908, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9918313489148491, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929522315138265, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955904166165152, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958528484168806, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967070045439821, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998816384786838, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.27.self_attn": [ + { + "accuracy": 0.9742694104972639, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9751778574366319, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9777410218590185, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9827587275128615, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9872362841116754, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9874108269026405, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9927543691898647, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929538047627399, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933693424651497, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9937053805119113, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935720590384383, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9939749668303289, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9945784869946932, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9949658909126332, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970572775131777, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974686813197637, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981119199410865, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998705620240224, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994774842634797, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.28.mlp": [ + { + "accuracy": 0.9289437093232807, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9304871119950947, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9397805050799721, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9430050285238969, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9628361620401081, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.965670551124372, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.970156264932532, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9804498286623704, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9822754342305032, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9810116291046143, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9832781725808194, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9903386156809958, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916684541263079, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947845578977936, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950988026041734, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.99606226266999, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9986108966956013, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.28.self_attn": [ + { + "accuracy": 0.9702131936424657, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.971275950732984, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9740250298851415, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9803612624344072, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9848950897392473, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9850304456133592, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9915356377237722, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916713927921496, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9921480793701974, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9924506850932774, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923675021058634, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9927654062446795, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933555228145499, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9938259579633412, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964079578456125, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969163850734108, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997741121406618, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984062536570587, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993824422065365, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.29.mlp": [ + { + "accuracy": 0.9125575015419408, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9145241599333913, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9253960785112882, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9291442318966514, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9542517442452281, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9577574478952509, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9630059599876404, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9759357210836912, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9781572238395089, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9766264400984112, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9794122231634039, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9881048751504797, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9897411026452717, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935762462647337, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9939658386926902, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951042713303315, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982916928435627, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.29.self_attn": [ + { + "accuracy": 0.9673727656665602, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9683243977396112, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9709329354135614, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.976757537377508, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9836090301212511, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9837269155602706, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9909698939637134, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9910819279520136, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916317345280397, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922285793643248, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.99177385866642, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922328152154621, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9928619994928962, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933675480516333, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961585802467245, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9966757321044019, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976857943754447, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983022491398611, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993657132022475, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.3.mlp": [ + { + "accuracy": 0.9443089899263883, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9454103959234137, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9522324769120467, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9550921289544356, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9716685853506389, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9738717408556687, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9772098487929294, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9847907643569144, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9861881701569808, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9854916302781356, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9872373899346903, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992568585433458, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935975019869051, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958560368732402, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962093630119374, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969712615405258, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987502796085257, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.3.self_attn": [ + { + "accuracy": 0.9614194600205672, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9634105343567698, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9669317320773476, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9772041910573056, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9814003831461856, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9815970612199683, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9894407964066455, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9895913247999392, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9902092092915585, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9907470627834922, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.990613621316458, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9910342834497753, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917353313220175, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922594718242946, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955640639129438, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961037859320641, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972167268003288, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978979857344377, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992306058932292, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.30.mlp": [ + { + "accuracy": 0.8836468520917391, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.886378150237234, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9004689894224468, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9052901895422685, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9390157335682919, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9437329016233745, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9504735407076383, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9678573357431512, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9708012907128585, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9687819731862921, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9724985110132318, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9840699857787082, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9862779504374454, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.991371858276819, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9919060970607557, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9934163666085193, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976937059116991, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.30.self_attn": [ + { + "accuracy": 0.9623192956573084, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9635958295119436, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9672465387143587, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9740532260192069, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9808562366585982, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9810579293652585, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9894268034320128, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9896573526294608, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9903241984154049, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9908222148292943, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9903350119528017, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9908606010048013, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917319754236623, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923343227097863, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955151708502519, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996122699819113, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972130730748177, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980214328358048, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992357649791398, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.31.mlp": [ + { + "accuracy": 0.8020908330616198, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.8071841942636591, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.8284267124376798, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.8356714625107615, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.8966757937481529, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.904714396125392, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9150376194401791, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9451305050598948, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9502325497175518, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9469982197410182, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9533064804579082, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9729334210094652, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9766251225220529, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9851571321487427, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9862106489507776, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9885709215151636, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958886014003503, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.31.self_attn": [ + { + "accuracy": 0.9566826663519207, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9584607199618691, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9643716749392057, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9720102297632318, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9780330203081432, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9785660426867636, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.987405115836545, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.988019336211054, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9887370812265497, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9893162854407963, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9891512009658312, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9897342747763583, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9910114650663576, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916506903736215, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951071005902792, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957718574687054, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9968163553429278, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977732587017512, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9991376368623031, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.4.mlp": [ + { + "accuracy": 0.9592334094800448, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9599072023441917, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9656191029046711, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9677070567482396, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9791517414544758, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9807854875137931, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9834634975383156, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9889578991814664, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9899803243185344, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9893246206798052, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9906035849922582, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9945216582793939, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9952811524271965, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969851707941607, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971972994114223, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978050448392567, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9990981947630644, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.4.self_attn": [ + { + "accuracy": 0.9644391285745721, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.965633655849256, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9692778901049965, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9785860494563454, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9824158737533971, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9826152842295798, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9902989401629096, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9904911886704596, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9911410604652605, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916440036736036, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9911410173303202, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916303267604426, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9924010885389227, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929460138082504, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958884084695264, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964591664703268, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974861788122278, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981860142005118, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993056849037346, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.5.mlp": [ + { + "accuracy": 0.9662242061213443, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.966813724291952, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.97141800114983, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9730477144843653, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9827299886628201, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9841003512081347, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9861850644412794, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9908859094506816, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917345517560056, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9911547256143469, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922273841343427, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9954550678008481, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960966329825552, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975041821599007, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976722909823844, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981429331788891, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992471238794295, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.5.self_attn": [ + { + "accuracy": 0.9737078343567095, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9744287399869216, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.976561471035606, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9818744816278157, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9870125086683976, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.987088993191719, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9926799880830866, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9928101889396969, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9931071949632544, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9937417856172511, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9932459757516259, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9937547747241823, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9941933947174173, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9945682836206335, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996890024723191, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973050002989016, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981055155788597, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985733104771689, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994472437783292, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.6.mlp": [ + { + "accuracy": 0.9656124114990234, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9662259754381681, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9714343234112388, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9732917892305475, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9822657484757272, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9836635652341341, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9860718548297882, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.990612121789079, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9915039876573964, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9908918510926398, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9919984834758859, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9953250226221586, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959855067886805, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974452319898104, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976045269715158, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981516301631927, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9992484119475672, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.6.self_attn": [ + { + "accuracy": 0.9695235396686354, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.97146307167254, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9748778829449102, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9828046403433147, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9855607669604453, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9857381927339655, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916512942627856, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9918205408673537, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923372127507862, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992747995414232, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9927642925789482, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9930673884718042, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9937891344490805, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9941759062440771, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.996694558153027, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970792892732119, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978154548688939, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984203206473276, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993902537971735, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.7.mlp": [ + { + "accuracy": 0.9682550555781314, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.968813397382435, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9736718974615398, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9754001341368023, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9835824480182246, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9849148314250143, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9871398728144797, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913353402363626, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9921657286192241, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9915860020800641, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992624116571326, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956848891942125, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963029799492735, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9976377618548117, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977922498395568, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982951285415574, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993059389097126, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.7.self_attn": [ + { + "accuracy": 0.9697458995015997, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9710842339616073, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9746506198456413, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9820316747615212, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9850262278004697, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9853074660426692, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913638257666638, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9915724820212314, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922322630882263, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992594741676983, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9925041002662558, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929043101636987, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935428217837685, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940782746202067, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965331664210871, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970465518142048, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977031559928468, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9984738441478265, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993621891756591, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.8.mlp": [ + { + "accuracy": 0.9694317296931618, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9699630486337762, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9746997356414795, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9763235371363791, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9842288337255779, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9854673727562553, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9876511293022256, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917293826216146, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9925017592154051, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9919363832787463, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929136531917673, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958759283548907, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964541353677449, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977602331261886, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978916846999997, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983712717106468, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9993482398751535, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.8.self_attn": [ + { + "accuracy": 0.9739625061813154, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9752250031421059, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9774304800911954, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9821768403053284, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9874328045468581, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9876089103912052, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9925937495733562, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9928273323335146, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9932714418361062, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935960730439738, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.993676625976437, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940517709443444, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9944762058163944, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948539455470286, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969857288034338, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974161372765115, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.998045519974671, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985959218128732, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.999433560394927, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.9.mlp": [ + { + "accuracy": 0.9726729894939222, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 304011760, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9731899970456174, + "down_proj": { + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64, + "6": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 314579440, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9772774263432151, + "down_proj": { + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "3": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 351271040, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9786252300990256, + "down_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 394000512, + "up_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9859657601306313, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 128, + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 444300992, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9870555792984209, + "down_proj": { + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "3": 32, + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 455585920, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9889204807971653, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 489718464, + "up_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "3": 32, + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9926753083342, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 559812800, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933485467182962, + "down_proj": { + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "4": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 568053888, + "up_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.992825437533228, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 128, + "5": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 577208000, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.993691196174998, + "down_proj": { + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "4": 32, + "5": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 588492928, + "up_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963305443525314, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 128, + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 710115008, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9968376996877947, + "down_proj": { + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "group_size": { + "5": 32, + "6": 32, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 721399936, + "up_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980058499464863, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "6": 128, + "8": 32 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 821395136, + "up_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981187425161663, + "down_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 852622368, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985205263487602, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 929168416, + "up_proj": { + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "6": 128, + "8": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994038440973351, + "down_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "gate_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 1086979104, + "up_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.layers.9.self_attn": [ + { + "accuracy": 0.9794431956190812, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 80752640, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9800505936145782, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 82817024, + "v_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9819086319521854, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 84584960, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9850869225828272, + "k_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "2": 64, + "3": 64 + }, + "scale_bits": 4 + }, + "total_bits": 101098496, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9894285507892308, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119674880, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9896768786405262, + "k_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 128, + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 119709184, + "v_proj": { + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "3": 64, + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.993534910443582, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152705024, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9937916766656073, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 128 + }, + "scale_bits": 4 + }, + "total_bits": 152772096, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9943441913316124, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 64 + }, + "scale_bits": 4 + }, + "total_bits": 154024960, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9945483980210204, + "k_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + }, + "total_bits": 156262400, + "v_proj": { + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "4": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9946190981488479, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 128, + "5": 128 + }, + "scale_bits": 4 + }, + "total_bits": 157457920, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950928903724018, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 158137344, + "v_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951895765568081, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 64, + "5": 64 + }, + "scale_bits": 4 + }, + "total_bits": 159887360, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 64 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956747222887842, + "k_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "4": 32, + "5": 32 + }, + "scale_bits": 4 + }, + "total_bits": 161980416, + "v_proj": { + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "5": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974371545801037, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 128, + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 197007360, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.997843890009742, + "k_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "group_size": { + "5": 32, + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 199729152, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982697116701227, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + }, + "total_bits": 228202496, + "v_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 128 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9988866273901964, + "k_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "6": 32 + }, + "scale_bits": 4 + }, + "total_bits": 235954176, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 32 + }, + "scale_bits": 4 + } + }, + { + "accuracy": 0.9994748029367704, + "k_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "o_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "q_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + }, + "total_bits": 303699968, + "v_proj": { + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "group_size": { + "8": 128 + }, + "scale_bits": 4 + } + } + ], + "model.norm.norm": null + } +} \ No newline at end of file