diff --git "a/quant_strategy.json" "b/quant_strategy.json" new file mode 100644--- /dev/null +++ "b/quant_strategy.json" @@ -0,0 +1,4265 @@ +{ + "measurement": { + "model.layers.0": { + "accuracy": 0.7789640426635742, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.1": { + "accuracy": 0.8816347122192383, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.2": { + "accuracy": 0.8823585510253906, + "total_bits": 171831456, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "model.layers.3": { + "accuracy": 0.9009003639221191, + "total_bits": 168554656, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + } + }, + "model.layers.4": { + "accuracy": 0.8899164199829102, + "total_bits": 169210016, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "model.layers.5": { + "accuracy": 0.8945960998535156, + "total_bits": 169865376, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "model.layers.6": { + "accuracy": 0.8889718055725098, + "total_bits": 171176096, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "model.layers.7": { + "accuracy": 0.8746490478515625, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.8": { + "accuracy": 0.8656606674194336, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.9": { + "accuracy": 0.8623347282409668, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.10": { + "accuracy": 0.8454461097717285, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.11": { + "accuracy": 0.8986392021179199, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.12": { + "accuracy": 0.9072887897491455, + "total_bits": 169865376, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "model.layers.13": { + "accuracy": 0.9071612358093262, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.14": { + "accuracy": 0.9054374694824219, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.15": { + "accuracy": 0.9041454792022705, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.16": { + "accuracy": 0.9000320434570312, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.17": { + "accuracy": 0.8985891342163086, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.18": { + "accuracy": 0.8946540355682373, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.19": { + "accuracy": 0.8946354389190674, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.20": { + "accuracy": 0.8893041610717773, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.21": { + "accuracy": 0.8887925148010254, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.22": { + "accuracy": 0.8898129463195801, + "total_bits": 176418976, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.02, + 0.98 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + "model.layers.23": { + "accuracy": 0.8921952247619629, + "total_bits": 186249376, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "model.layers.24": { + "accuracy": 0.89935302734375, + "total_bits": 186249376, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "model.layers.25": { + "accuracy": 0.9103565216064453, + "total_bits": 171176096, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "model.layers.26": { + "accuracy": 0.937091588973999, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.27": { + "accuracy": 0.9371011257171631, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.28": { + "accuracy": 0.9361209869384766, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.29": { + "accuracy": 0.9362266063690186, + "total_bits": 174452896, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.01, + 0.99 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + "model.layers.30": { + "accuracy": 0.9399299621582031, + "total_bits": 186249376, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "model.layers.31": { + "accuracy": 0.9511301517486572, + "total_bits": 220590240, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "model.layers.32": { + "accuracy": 0.9504289627075195, + "total_bits": 220590240, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "model.layers.33": { + "accuracy": 0.9642099142074585, + "total_bits": 264630432, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "model.layers.34": { + "accuracy": 0.963350772857666, + "total_bits": 264630432, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.5, + 0.5 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "model.layers.35": { + "accuracy": 0.9725379943847656, + "total_bits": 292155552, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "model.layers.36": { + "accuracy": 0.9723954200744629, + "total_bits": 292155552, + "q_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128, + "2": 128 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.75, + 0.25 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "model.layers.37": { + "accuracy": 0.9867923259735107, + "total_bits": 319680672, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.38": { + "accuracy": 0.9684550762176514, + "total_bits": 319680672, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + "model.layers.39": { + "accuracy": 0.9736276865005493, + "total_bits": 319680672, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + } +} \ No newline at end of file