{ "measurement": { "model.layers.0": { "accuracy": 0.9354426860809326, "total_bits": 361673760, "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.1": { "accuracy": 0.9299473762512207, "total_bits": 369407008, "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": [ 0.1, 0.9 ], "scale_bits": 4 } }, "model.layers.2": { "accuracy": 0.927403450012207, "total_bits": 508736544, "q_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "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": [ 0.25, 0.75 ], "scale_bits": 4 }, "up_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "gate_proj": { "group_size": { "4": 128, "2": 128 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "down_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, "model.layers.3": { "accuracy": 0.9244002103805542, "total_bits": 477279264, "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.4": { "accuracy": 0.937565803527832, "total_bits": 571651104, "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.5": { "accuracy": 0.9338212013244629, "total_bits": 571651104, "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.6": { "accuracy": 0.9320681095123291, "total_bits": 571651104, "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.7": { "accuracy": 0.9271104335784912, "total_bits": 571651104, "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.8": { "accuracy": 0.9257893562316895, "total_bits": 369013792, "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.9": { "accuracy": 0.9572443962097168, "total_bits": 361673760, "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.10": { "accuracy": 0.9548473358154297, "total_bits": 361673760, "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.11": { "accuracy": 0.9458401203155518, "total_bits": 361673760, "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.12": { "accuracy": 0.9409699440002441, "total_bits": 361673760, "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.13": { "accuracy": 0.9391531944274902, "total_bits": 361673760, "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.14": { "accuracy": 0.9374328851699829, "total_bits": 361673760, "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.15": { "accuracy": 0.9324246644973755, "total_bits": 361673760, "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.16": { "accuracy": 0.9298160076141357, "total_bits": 361673760, "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.17": { "accuracy": 0.9278309345245361, "total_bits": 379106336, "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.18": { "accuracy": 0.9249012470245361, "total_bits": 405582880, "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": 0.9240758419036865, "total_bits": 405582880, "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": 0.937448263168335, "total_bits": 477279264, "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": 0.9370510578155518, "total_bits": 477279264, "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.22": { "accuracy": 0.9278292655944824, "total_bits": 477279264, "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": 0.9289307594299316, "total_bits": 477279264, "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": 0.9291336536407471, "total_bits": 477279264, "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": 0.9336254596710205, "total_bits": 477279264, "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": 0.9338843822479248, "total_bits": 477279264, "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": 0.9345648288726807, "total_bits": 477279264, "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": 0.9210338592529297, "total_bits": 405582880, "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.29": { "accuracy": 0.9280407428741455, "total_bits": 405582880, "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.30": { "accuracy": 0.9309263229370117, "total_bits": 379106336, "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.31": { "accuracy": 0.9278252124786377, "total_bits": 373601312, "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 } } } }