{ "measurement": { "model.layers.0": { "accuracy": 0.9382696151733398, "total_bits": 446084128, "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.1": { "accuracy": 0.9415526390075684, "total_bits": 578335776, "q_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.5, 0.5 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "up_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "gate_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "down_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.5, 0.5 ], "scale_bits": 4 } }, "model.layers.2": { "accuracy": 0.9650976061820984, "total_bits": 424064032, "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.3": { "accuracy": 0.9562180042266846, "total_bits": 424064032, "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": 0.9498416185379028, "total_bits": 424064032, "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.5": { "accuracy": 0.9452450275421143, "total_bits": 424064032, "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.6": { "accuracy": 0.9379130601882935, "total_bits": 424064032, "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.7": { "accuracy": 0.932453989982605, "total_bits": 424064032, "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.8": { "accuracy": 0.9342072010040283, "total_bits": 446084128, "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.9": { "accuracy": 0.932488203048706, "total_bits": 446084128, "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.10": { "accuracy": 0.9288740158081055, "total_bits": 446084128, "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.11": { "accuracy": 0.9333758354187012, "total_bits": 478983200, "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.12": { "accuracy": 0.9322373867034912, "total_bits": 478983200, "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.13": { "accuracy": 0.9323720932006836, "total_bits": 478983200, "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.14": { "accuracy": 0.9289882183074951, "total_bits": 478983200, "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.15": { "accuracy": 0.9291183948516846, "total_bits": 478983200, "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.16": { "accuracy": 0.9419481754302979, "total_bits": 569947168, "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.17": { "accuracy": 0.9265720844268799, "total_bits": 478983200, "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.18": { "accuracy": 0.9266581535339355, "total_bits": 478983200, "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.19": { "accuracy": 0.9268605709075928, "total_bits": 478983200, "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.20": { "accuracy": 0.9419999122619629, "total_bits": 569947168, "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.21": { "accuracy": 0.928084135055542, "total_bits": 478983200, "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.22": { "accuracy": 0.941460132598877, "total_bits": 569947168, "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.23": { "accuracy": 0.9415285587310791, "total_bits": 569947168, "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.24": { "accuracy": 0.9416172504425049, "total_bits": 569947168, "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.25": { "accuracy": 0.9402132034301758, "total_bits": 569947168, "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.26": { "accuracy": 0.939831018447876, "total_bits": 569947168, "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.27": { "accuracy": 0.9388439655303955, "total_bits": 569947168, "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.28": { "accuracy": 0.9380154609680176, "total_bits": 578335776, "q_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.5, 0.5 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "up_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "gate_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "down_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.5, 0.5 ], "scale_bits": 4 } }, "model.layers.29": { "accuracy": 0.9371981620788574, "total_bits": 578335776, "q_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.5, 0.5 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "up_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "gate_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "down_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.5, 0.5 ], "scale_bits": 4 } }, "model.layers.30": { "accuracy": 0.93943190574646, "total_bits": 675329056, "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.31": { "accuracy": 0.9260892868041992, "total_bits": 578335776, "q_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.5, 0.5 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "up_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "gate_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.35, 0.65 ], "scale_bits": 4 }, "down_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.5, 0.5 ], "scale_bits": 4 } } } }