{ "measurement": { "model.layers.0.self_attn": [ { "accuracy": 0.973719643894583, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9794381251558661, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.981044398387894, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9876877887872979, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9881932975258678, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9888243598397821, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9895069503691047, "total_bits": 355869696, "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 } }, { "accuracy": 0.9901029403554276, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9911176259629428, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9915056428872049, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943335170391947, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947512160870247, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946930714068003, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952097241184674, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968332842981908, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997482674109051, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970618706138339, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979524720984045, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990962060837774, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.0.block_sparse_moe": [ { "accuracy": 0.9401493659242988, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9474926968105137, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9493989497423172, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9494462641887367, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9897591817425564, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9938924019224942, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9939397787093185, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976530134008499, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981476864777505, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998737956288096, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990669398175669, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989867361291545, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995565570534382, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994281941853842, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99955531339765, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995861186489492, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999466610914169, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.1.self_attn": [ { "accuracy": 0.9916500975377858, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926550908130594, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931528515880927, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957825755700469, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961089556745719, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961993579927366, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973892237758264, "total_bits": 355869696, "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 } }, { "accuracy": 0.997520193021046, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.997734246251639, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99785023961158, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9980453397147357, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998172385545331, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982051278930157, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983589744078927, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989611127457465, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991425891348626, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999257760355249, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995122082436865, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997710563220608, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.1.block_sparse_moe": [ { "accuracy": 0.8984257634729147, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9141951370984316, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9193237163126469, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9193277722224593, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9838477098383009, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9923319689696655, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9923634060251061, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974782160061295, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976116467223619, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980333055937081, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997809303473332, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998613696618122, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986873789130186, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986472474120092, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.999240644465317, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992508911345794, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995269606479269, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.2.self_attn": [ { "accuracy": 0.996395785594359, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996627684449777, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969140161119867, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979865725472337, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980877204361605, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982422228931682, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985947446402861, "total_bits": 355869696, "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 } }, { "accuracy": 0.9987538797431625, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989531943429029, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989866522155353, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991016031126492, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991900031818659, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991683056141483, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992603858481743, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995068162897951, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996242883098603, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996190397469036, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998006579426146, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998957241132302, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.2.block_sparse_moe": [ { "accuracy": 0.9959539762348868, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9960642032674514, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9965382121445145, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9966885848843958, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981212187849451, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982341377035482, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984382152324542, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990318398486124, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991175930917962, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990608389562112, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99917690102302, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995134743876406, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995887761797349, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999726500680481, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9997429418963293, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997850501731591, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999200571346591, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.3.self_attn": [ { "accuracy": 0.9952593620982952, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9956582925515249, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961202817794401, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972391510382295, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975743942777626, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977553481585346, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981705765821971, "total_bits": 355869696, "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 } }, { "accuracy": 0.9984076297405409, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986166253802367, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986776430596365, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988475214486243, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989523422409547, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989504610421136, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990716549873468, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993951180222211, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99951926320864, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995319442132313, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997300035211083, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998596276545868, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.3.block_sparse_moe": [ { "accuracy": 0.9931382146314718, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9934657393023372, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9939135110471398, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9940690597286448, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9953043172718026, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9915942427469417, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.991669415612705, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996028795663733, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917726236162707, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980024852557108, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978713128075469, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976734749798197, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998304115186329, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977375948656118, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991326428353204, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991549457117799, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991945664514787, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.4.self_attn": [ { "accuracy": 0.9918317746487446, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917559548630379, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942852982203476, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957831191713922, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959835969784763, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960126580845099, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977866400731727, "total_bits": 355869696, "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 } }, { "accuracy": 0.9979559146304382, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981072488299105, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981869145849487, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982728416071041, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984662972856313, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985779699636623, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987218131573172, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991979252081364, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999302957112377, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994135429005837, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996423596967361, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998334071406134, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.4.block_sparse_moe": [ { "accuracy": 0.9931561208795756, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9933602716191672, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9941966400365345, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9944501546560787, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9966842654685024, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969423397269566, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972968996153213, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982938341854606, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984351187740685, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983160924166441, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985243714036187, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990406901633833, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992659492927487, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993668680326664, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995548214610608, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996265153349668, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998544390746247, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.5.self_attn": [ { "accuracy": 0.9926811900222674, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9930985138053074, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9938506953185424, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953896069782786, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961048979894258, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962614748219494, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973398502042983, "total_bits": 355869696, "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 } }, { "accuracy": 0.997548992105294, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9977085157879628, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9978364217386115, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.997965215385193, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981498382985592, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983614463853883, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985121924837586, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990786285416107, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992461201836704, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993196903960779, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995724580221577, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997897057655791, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.5.block_sparse_moe": [ { "accuracy": 0.9917308809817769, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9919728420791216, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9930336032994092, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9933628516155295, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9959417105419561, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962627270724624, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967154026089702, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997900229122024, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980912688333774, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99794020569243, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982001283351565, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989499161747517, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999102063186001, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994306683947798, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9994622738886392, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995547567850736, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998389647716976, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.6.self_attn": [ { "accuracy": 0.9916050465544686, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9925691977259703, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930020215688273, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9945441489107907, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958863380015828, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959470493486151, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973198994412087, "total_bits": 355869696, "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 } }, { "accuracy": 0.9973942312644795, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9975890364730731, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9976582917734049, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9978233384317718, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979365082981531, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982410151133081, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983820865745656, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990115827895352, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991685515560675, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992636592505733, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995161982078571, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997704317502212, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.6.block_sparse_moe": [ { "accuracy": 0.9888724300544709, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.989212658139877, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.990676419227384, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911273104371503, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9945723557611927, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950037469970994, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956311854766682, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971934125642292, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974337409657892, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997240724216681, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975864832522348, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985862711328082, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987906840251526, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992271646042354, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992777956940699, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994075033027912, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997844211138727, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.7.self_attn": [ { "accuracy": 0.9896782299038023, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9904253190616146, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909430299885571, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929927795310505, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941737889312208, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943199090193957, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966665752290282, "total_bits": 355869696, "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 } }, { "accuracy": 0.9967161471140571, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9970097086334135, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9971885181148537, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972495849942788, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974370923882816, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997748218011111, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979396704293322, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987520926952129, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989506101701409, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991039547312539, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993844864293351, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997246396778792, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.7.block_sparse_moe": [ { "accuracy": 0.9861582483863458, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9865659897914156, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9884724263101816, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989058691309765, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932166094658896, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.993752350914292, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9945651100133546, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996491995640099, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968053105985746, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965530289046001, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969827400636859, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982401905290317, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984945983305806, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990485042508226, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990996490887483, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992702650633873, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997343691211427, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.8.self_attn": [ { "accuracy": 0.9882878720527515, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9893808903871104, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9901964284945279, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929238315671682, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935778080835007, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9939908984233625, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963635508320294, "total_bits": 355869696, "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 } }, { "accuracy": 0.9964677078532986, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9967099939531181, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9968259264423978, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9968124493316282, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970754480164032, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973910504486412, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976709459733684, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985860703891376, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988120612833882, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989894011960132, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993605644704076, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997171479153621, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.8.block_sparse_moe": [ { "accuracy": 0.9843489901395515, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9848112479085103, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9869046255480498, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9875469988910481, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923497142153792, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9929443872533739, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9938452643691562, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960411039937753, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9963737338257488, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960787742456887, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965942918206565, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980037495406577, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983026857371442, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989069805742474, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9989846211610711, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991730203182669, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996979639945494, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.9.self_attn": [ { "accuracy": 0.984103512368165, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9862865785835311, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9876481133978814, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9911444369936362, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924177289940417, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9925178327248432, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9949980060919188, "total_bits": 355869696, "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 } }, { "accuracy": 0.9950839747325517, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9955377061269246, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995729211543221, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9960900048608892, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964420997130219, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967815406271257, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970698349643499, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981864541914547, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998513136190013, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987567493371898, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991899452506914, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996436865440046, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.9.block_sparse_moe": [ { "accuracy": 0.9812290843110532, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9817838175222278, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9844120587222278, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9852256767917424, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907401365926489, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9914848516928032, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9926111847744323, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9951990176341496, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.995631652243901, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9952911202562973, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958794162375852, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975975806009956, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979436767607694, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986958235094789, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987716131581692, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990112320083426, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996355575494817, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.10.self_attn": [ { "accuracy": 0.9843496754765511, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9857475481694564, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9868397096870467, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900230860803276, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9918688266188838, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920568243833259, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9949573355843313, "total_bits": 355869696, "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 } }, { "accuracy": 0.9952224837034009, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956043674610555, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958458917681128, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958707781042904, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961723740561865, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966718807118014, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969623592915013, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998122571822023, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984336437482852, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986585242586443, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991446661369991, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996283102009329, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.10.block_sparse_moe": [ { "accuracy": 0.9786292091012001, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9792771795764565, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9823354708496481, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.983264469075948, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9894522238755599, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9902893230319023, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9915869090473279, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9945112534915097, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.995003743446432, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994636484712828, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995304306037724, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972587983938865, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976550959108863, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985122851358028, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9986001680226764, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988766503374791, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995910119650944, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.11.self_attn": [ { "accuracy": 0.9820184677373618, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9829046044033021, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9843332796590403, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.988038198905997, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907613408286124, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909689459018409, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942319274414331, "total_bits": 355869696, "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 } }, { "accuracy": 0.9944732754956931, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948527517262846, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951552322600037, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952403281931765, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995541719137691, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996018716978142, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996304365224205, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977743928320706, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981546952476492, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984828153683338, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989588839234784, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995717936217261, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.11.block_sparse_moe": [ { "accuracy": 0.9759307778440416, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9766345131210983, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.980015923967585, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9810476729180664, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.988064359058626, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9890113447327167, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9904395529301837, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938055359525606, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994357067567762, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9939252902986482, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9946821884950623, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996900776430266, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997347933560377, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998323007428553, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984173469565576, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987199593306286, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995430240887799, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.12.self_attn": [ { "accuracy": 0.976000941824168, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9777310157660395, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9794058536645025, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9844446826027706, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9875601028325036, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879832556471229, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927486827364191, "total_bits": 355869696, "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 } }, { "accuracy": 0.9928569766343571, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932501800358295, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938043319853023, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939655006164685, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943528063013218, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9949027543771081, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953364313114434, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997196826065192, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976450852700509, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980909827281721, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986522361141397, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994402943120804, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.12.block_sparse_moe": [ { "accuracy": 0.9732591195497662, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9740158813074231, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9779601865448058, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9791718141641468, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9867176077095792, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9877674826420844, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9894444813253358, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9931030529551208, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937231780495495, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932427178719081, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9940825109370053, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965548638137989, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997048541350523, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981354575429577, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982417052815435, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985945414373418, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994893747243623, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.13.self_attn": [ { "accuracy": 0.9748206920921803, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9757225846406072, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9776651407591999, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9838092703139409, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9864006551215425, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9866252384381369, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9923593901330605, "total_bits": 355869696, "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 } }, { "accuracy": 0.9924705093144439, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9926701345830224, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935652484418824, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931022410746664, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935350427986123, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994136948720552, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947718686889857, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968659784353804, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973776535480283, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979663588019321, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9985600443033036, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994380470889155, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.13.block_sparse_moe": [ { "accuracy": 0.9704063427634537, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9713004773948342, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9756564958952367, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.976957930251956, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9854278799612075, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9865821324056014, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.988406561431475, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9924318412668072, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.993108895199839, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9925814591697417, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.993507523147855, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962115383532364, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967589949083049, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979461711336626, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980611534556374, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984421611443395, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994300844628015, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.14.self_attn": [ { "accuracy": 0.9713667889591306, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9725863009225577, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9747465765103698, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9812633327674121, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9851514748297632, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9854043558007106, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914394017541781, "total_bits": 355869696, "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 } }, { "accuracy": 0.9916485782014206, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9921812160173431, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9927700157277286, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.992504570807796, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930634860647842, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.993698064237833, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941395184141584, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965180973231327, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969361130788457, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977696200367063, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9983049488364486, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993748165361467, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.14.block_sparse_moe": [ { "accuracy": 0.9683715074788779, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.969330680090934, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.973947603488341, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9753206393215805, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9844108216930181, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9856440962757915, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.987575021921657, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918921951320954, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926129542873241, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9920518174185418, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9930423421319574, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959368488634937, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965269809181336, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977976916707121, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979223286354681, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983261907618726, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993990038492484, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.15.self_attn": [ { "accuracy": 0.967776637757197, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9689149453770369, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9716730681248009, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.97805135557428, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9834884674055502, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9838872303953394, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910781319485977, "total_bits": 355869696, "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 } }, { "accuracy": 0.9911683950340375, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9917910853982903, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.992269204754848, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9920589426765218, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926872146315873, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9932510215439834, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9938015782972798, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962926311709452, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969139701861423, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976277402311098, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982630238228012, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993480346893193, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.15.block_sparse_moe": [ { "accuracy": 0.965417695697397, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9664727859199047, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9717281286139041, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9733062551822513, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9829341054428369, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9842927993740886, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.986502897227183, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911032012896612, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9919080713880248, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912996545899659, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9923922980669886, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955540997907519, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962040259269997, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975863210565876, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977273327822331, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981839707179461, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993431148832315, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.16.self_attn": [ { "accuracy": 0.9599521332420409, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9601033087819815, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.962146318051964, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9672264188993722, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9820787494536489, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9822285114787519, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900003822986037, "total_bits": 355869696, "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 } }, { "accuracy": 0.9902491831453517, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9906184058636427, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914659161586314, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9911792647326365, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9911609614500776, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991903179208748, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.993092885997612, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958839159808122, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965309223334771, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974383822409436, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9980901092349086, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992771726901992, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.16.block_sparse_moe": [ { "accuracy": 0.9630389926023781, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9641888434998691, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9698657200206071, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9715564749203622, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9817663112189621, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.983224775409326, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9855999121209607, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9904909225879237, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9913651263341308, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907136737601832, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.991880739456974, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952595789218321, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959467829612549, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974280300084502, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975771072786301, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980665816838155, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992929452782846, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.17.self_attn": [ { "accuracy": 0.9640472093597054, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9652763935737312, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9679816400166601, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9764737589284778, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9815218881703913, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9817342676687986, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897926209960133, "total_bits": 355869696, "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 } }, { "accuracy": 0.9899911822285503, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9905787097522989, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9911715039052069, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9906888217665255, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912705867318437, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919983659056015, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926692162407562, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955902892979793, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963253128516953, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973122062510811, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9980396151659079, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992685150355101, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.17.block_sparse_moe": [ { "accuracy": 0.9617686849087477, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9629443190060556, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9686820388305932, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9703945235814899, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9811446426901966, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9826286181341857, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.98504815378692, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901715975720435, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910595461260527, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9904048583703116, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.991597686544992, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995106928050518, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995807902305387, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973431075923145, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975018253899179, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980051454767818, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992711735831108, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.18.self_attn": [ { "accuracy": 0.9609708348289132, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9622493968345225, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9652338935993612, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9736187404487282, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9795365163590759, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9800808769650757, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.988523177918978, "total_bits": 355869696, "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 } }, { "accuracy": 0.9888855087338015, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9897511965828016, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9905439659487456, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9899780629202724, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907545428723097, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9915054470766336, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9921958897612058, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952578634256497, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960808835749049, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970618478255346, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9978651599376462, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992049475840759, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.18.block_sparse_moe": [ { "accuracy": 0.9605018114671111, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9617222459055483, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9676584796980023, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9694292107596993, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.980506684165448, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9820493862498552, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9845328034134582, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9898208394879475, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.990736072533764, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9900782139739022, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.991312641883269, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949395424337126, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956656178110279, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972446645842865, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974186525214463, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979341145372018, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992445061216131, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.19.self_attn": [ { "accuracy": 0.9570305058732629, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9587050136178732, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9616323197260499, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9699855267535895, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9771643439307809, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9773946579080075, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.986969810212031, "total_bits": 355869696, "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 } }, { "accuracy": 0.9870926297735423, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9883447979809716, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9889916040701792, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.988633205415681, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9895930116763338, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9904401312815025, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9913151778746396, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947695219889283, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956095084780827, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966598478786182, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9975374845380429, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990892872374388, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.19.block_sparse_moe": [ { "accuracy": 0.9588211444206536, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9600797817111015, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9662497863173485, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9680995286908001, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9795682250987738, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9811910265125334, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9837764706462622, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9893300653202459, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9902926301583648, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9895857652882114, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9908832464134321, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9946833429276012, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954495338024572, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971120498084929, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9972853125364054, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978208338725381, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992135285719996, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.20.self_attn": [ { "accuracy": 0.9557601055130363, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9576921472325921, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9613528489135206, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9699599801097065, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9779774092603475, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9781769446562976, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9871278777718544, "total_bits": 355869696, "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 } }, { "accuracy": 0.9873884741682559, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9870958368992433, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9887391057563946, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9887579150963575, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989587641088292, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9905195524916053, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907283441862091, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947886498994194, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956725314259529, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965944057621527, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9975431880739052, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990719386987621, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.20.block_sparse_moe": [ { "accuracy": 0.9551001088693738, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9564834926277399, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9633405688218772, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9654281013645232, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9776763843838125, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9794639912433922, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9823353653773665, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9883022542344406, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9893645090050995, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9886250720592216, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9900495570618659, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9941970216459595, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950337277259678, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968296959996223, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9970409778470639, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976396976853721, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991215574555099, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.21.self_attn": [ { "accuracy": 0.9520502067171037, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9538233382627368, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9574317308142781, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9671131481882185, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9755651322193444, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9757262079510838, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9858048758469522, "total_bits": 355869696, "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 } }, { "accuracy": 0.9860857286257669, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9869651978369802, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9875420444877818, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9873310662806034, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9885607964824885, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9894763580523431, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9904063238063827, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941289144917391, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951683569815941, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962352972943336, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9973092258442193, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989942258544033, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.21.block_sparse_moe": [ { "accuracy": 0.954626533202827, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9560268227942288, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9628162649460137, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9648632570169866, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9775655490811914, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9793480592779815, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.982138458872214, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9882613003719598, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9893176659243181, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9885782115161419, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9899976467713714, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9941712091094814, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950046935700811, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968185143952724, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997025210497668, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976106700487435, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991147090404411, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.22.self_attn": [ { "accuracy": 0.949972253292799, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9517655246891081, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9564938526600599, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9673811255488545, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9747711047530174, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9748781626112759, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.98569403111469, "total_bits": 355869696, "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 } }, { "accuracy": 0.9860030920244753, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9870262944605201, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9878319537965581, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.987338452367112, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882308720843866, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893311038613319, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9902778660180047, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942011352977715, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951201216317713, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963077088759746, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972950564115308, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990022219571983, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.22.block_sparse_moe": [ { "accuracy": 0.9570436249487102, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9584419294260442, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9646603735163808, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9665086879394948, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9789894185960293, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9806561088189483, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9830835673492402, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890083306236193, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899708066368476, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9893066416261718, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9906417563324794, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9945444034528919, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953275716397911, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970126853731927, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9972146470099688, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977250351221301, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991713527560933, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.23.self_attn": [ { "accuracy": 0.9625110924243927, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9637769577093422, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.966564605012536, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9738302598707378, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9800550136715174, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9807105236686766, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879675803240389, "total_bits": 355869696, "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 } }, { "accuracy": 0.9883405048167333, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9899132796563208, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9905470544472337, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9902394263772294, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909509517019615, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919448642758653, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9925700563471764, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955944587709382, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962919488025364, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970529670536052, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979113442532253, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991648048817297, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.23.block_sparse_moe": [ { "accuracy": 0.9607007699087262, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9619574476964772, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9674513232894242, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9690929267089814, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9806298506446183, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9822025645989925, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9843694533919916, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899175954051316, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907916390802711, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901461307890713, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9913945348234847, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949757147114724, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995706012705341, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972520888841245, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974306373915169, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978723651147448, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999219491808617, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.24.self_attn": [ { "accuracy": 0.9664455398451537, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9672865895554423, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9700764727313071, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9767538434825838, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9829392600804567, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9834318978246301, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9896894725970924, "total_bits": 355869696, "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 } }, { "accuracy": 0.9902962889755145, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9909171301405877, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9911826094612479, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914430857170373, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9918265676824376, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924440390313976, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9933729203767143, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960237946070265, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966719633084722, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973707739263773, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9980602317809826, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992639154515928, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.24.block_sparse_moe": [ { "accuracy": 0.9616416334174573, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9628558503463864, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9681100733578205, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9696621894836426, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9810919526498765, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9826096969190985, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9846815709024668, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901692303828895, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910194807453081, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9903617023956031, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9915768408682197, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950784971588291, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957940273743588, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973149679135531, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974803202203475, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979004947672365, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992430431157118, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.25.self_attn": [ { "accuracy": 0.9638154790736735, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9648589286953211, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9687254650052637, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9744850778952241, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9814297519624233, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9816410709172487, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9894651662325487, "total_bits": 355869696, "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 } }, { "accuracy": 0.9897387054516003, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9906011609127745, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914427160983905, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9910254400456324, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920985630596988, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928687102510594, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9934953977935947, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960172942955978, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967187583097257, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973323218291625, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998118134928518, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992479261636618, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.25.block_sparse_moe": [ { "accuracy": 0.960497529245913, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9617349430918694, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9669919717125595, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9686024936381727, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9804621993098408, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.982014431618154, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9841588394483551, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9897974486229941, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9906876081367955, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9900451535359025, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9912895348388702, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949218714609742, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956507445313036, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972244567470625, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974054566700943, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978490395587869, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992119001763058, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.26.self_attn": [ { "accuracy": 0.9631168600171804, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9636643985286355, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9663194555323571, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9731329123023897, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9801708133891225, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9809051679912955, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9889795810449868, "total_bits": 355869696, "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 } }, { "accuracy": 0.9894046327099204, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.98994840902742, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.990435883984901, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.990632833680138, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991473633388523, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9922426737030037, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928709110827185, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956546968896873, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964123999234289, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970072669966612, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9977708974038251, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991890938035795, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.26.block_sparse_moe": [ { "accuracy": 0.9603411490097642, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.961542584002018, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9668214973062277, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9684556976426393, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9803956034593284, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9819333362393081, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.984100311761722, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9897457711631432, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9906384864589199, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9900107397697866, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9912467135582119, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949007508694194, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956275673175696, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997209267661674, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973937429313082, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997849253952154, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992036705152714, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.27.self_attn": [ { "accuracy": 0.9664644510485232, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9668303208891302, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9687591835390776, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.973463659407571, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9835765545722097, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.983623806387186, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9901358917122707, "total_bits": 355869696, "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 } }, { "accuracy": 0.9905279915546998, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9911560716573149, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9915970010915771, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9916943847783841, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920543613261543, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928927542641759, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935763159883209, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958971608430147, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964870704861823, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973794833640568, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981689975684276, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992649610649096, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.27.block_sparse_moe": [ { "accuracy": 0.9610110628418624, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9621670809574425, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9674255645368248, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9690947344060987, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9806451522745192, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9821676639840007, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9844130493002012, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9898728803964332, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907639977755025, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9901462187990546, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9913651956594549, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949734540423378, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956885155988857, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972512225504033, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974338922766037, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979015955323121, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992203419169527, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.28.self_attn": [ { "accuracy": 0.9657252850010991, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9669495562557131, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9704823719803244, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.976862822426483, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9819612221326679, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.981981192715466, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893937631277367, "total_bits": 355869696, "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 } }, { "accuracy": 0.9900287213968113, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9907992165535688, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9913255834253505, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.991307764314115, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9917931819218211, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926052223308943, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9932311059674248, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959957426472101, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966421108983923, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972992668044753, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981029340851819, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992317547803395, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.28.block_sparse_moe": [ { "accuracy": 0.9605085947550833, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9616565611213446, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9671408920548856, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9689630176872015, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9802696029655635, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9818392156157643, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9842506794957444, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9895961355650797, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9905420636059716, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9899510035756975, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9911970508983359, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948726415750571, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956019914825447, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971801178180613, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973873279523104, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978960124135483, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991938233579276, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.29.self_attn": [ { "accuracy": 0.96618148451671, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.967889778316021, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9708085299935192, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9772881150711328, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9824537257663906, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.982924735872075, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989325049216859, "total_bits": 355869696, "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 } }, { "accuracy": 0.9895870137261227, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9905557526508346, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9909560506930575, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9905088256346062, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919135076343082, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928404567763209, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9933210429735482, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959641398163512, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967009698448237, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972468229825608, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9975466504110955, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992354061905644, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.29.block_sparse_moe": [ { "accuracy": 0.964088742621243, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9651157399639487, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9699720349162817, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9715870376676321, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9820392064284533, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9834509214852005, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9855626114876941, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9905578994657844, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914089061203413, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.990849380614236, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9919790522544645, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953309952397831, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959970532217994, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997439689876046, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976195238123182, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980610982311191, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999273438435921, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.30.self_attn": [ { "accuracy": 0.9736473697703332, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9750442071817815, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9785423984285444, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.982845441205427, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9866152859758586, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9864614575635642, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919204036123119, "total_bits": 355869696, "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 } }, { "accuracy": 0.99216842133319, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9929463041480631, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932276124600321, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9927670055767521, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935037377872504, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9945110395783558, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950499127735384, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996987998514669, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99749459917075, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997835732443491, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998661172678112, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994023876352003, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.30.block_sparse_moe": [ { "accuracy": 0.9631482176482677, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.964197691064328, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9690836167428643, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9707611403428018, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9815215454436839, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9829739489359781, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9851436807075515, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9902646031696349, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9911487664794549, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9905887034256011, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9917616490856744, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952005541999824, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9958842295163777, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973628149018623, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975585275387857, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980081124813296, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992503805478918, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.31.self_attn": [ { "accuracy": 0.9746476996224374, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9747080609668046, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9785332644823939, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9816192148718983, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9862032291712239, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9863394993590191, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9916566653409973, "total_bits": 355869696, "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 } }, { "accuracy": 0.9920320453238674, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.992928643419873, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932576064020395, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99331119726412, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937463416717947, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947749942657538, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953339895582758, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997116257000016, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975795118371025, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978665089001879, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986981224064948, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994067886073026, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.31.block_sparse_moe": [ { "accuracy": 0.9648436498828232, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.965879671741277, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9702413955237716, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9717190251685679, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.982488278998062, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.983850893448107, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9857509180437773, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.990797852515243, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9916162637528032, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910973244113848, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9921930736163631, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954650985891931, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961035196611192, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975114330882207, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997690429125214, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980848008708563, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992858397963573, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.32.self_attn": [ { "accuracy": 0.9716836870647967, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9726009203586727, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9757514726370573, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9821357703767717, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9852843066910282, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9854605954606086, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991447392385453, "total_bits": 355869696, "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 } }, { "accuracy": 0.9916101196431555, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9922549140173942, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9926399880787358, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9925338943721727, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930604131659493, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9938865875010379, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944028447498567, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99662088722107, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971792021242436, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977853945747484, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984490426140837, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993802298777155, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.32.block_sparse_moe": [ { "accuracy": 0.9653742285445333, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9663617305923253, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9706091203261167, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9720806626137346, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9826854281127453, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9840436624363065, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9859079672023654, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9909025133820251, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991719437239226, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912002230994403, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9922874680487439, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955183831043541, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961512508743908, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975389041064773, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977184429881163, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980995394289494, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992920587465051, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.33.self_attn": [ { "accuracy": 0.9728645388968289, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9737720871344209, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9765683389268816, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9818953818175942, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9859420980792493, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9861697143642232, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9917335549253039, "total_bits": 355869696, "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 } }, { "accuracy": 0.9920160094625317, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9925237980205566, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9930047715315595, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9930061782361008, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9934711568057537, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994178930122871, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994661457312759, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967678760585841, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972922828455921, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978144946217071, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984822630940471, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993723916559247, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.33.block_sparse_moe": [ { "accuracy": 0.9661223469302058, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9670809032395482, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.971171441487968, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9725918178446591, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9830557283712551, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9843870368786156, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9861671619582921, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910986373433843, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918983834795654, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9913807459524833, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9924491847632453, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956088345497847, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962341400096193, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975962817843538, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977653897658456, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981310377479531, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993213288544212, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.34.self_attn": [ { "accuracy": 0.9750087480060756, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9760705474764109, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9786439470481128, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9827800134662539, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9865027533378452, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9869530660798773, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919057592051104, "total_bits": 355869696, "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 } }, { "accuracy": 0.9923130607348867, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934667921043001, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937579997931607, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937260445440188, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943268067436293, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950785199180245, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953760701464489, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972570162790362, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976573825697415, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980858896888094, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987017637904501, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994333984723198, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.34.block_sparse_moe": [ { "accuracy": 0.9686819782946259, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9695732318796217, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9731907844543457, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9744658786803484, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9843345721019432, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9855656964937225, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9871361595578492, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917722376994789, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.992501079570502, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.992038473719731, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9930214597261511, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959440908569377, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965145503228996, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977681389718782, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979335698881187, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982525429950329, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993539171773591, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.35.self_attn": [ { "accuracy": 0.980748048517853, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9815634288825095, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9852022381965071, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9883458570111543, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9904194154078141, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9905000928556547, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994338053104002, "total_bits": 355869696, "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 } }, { "accuracy": 0.9945792850921862, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949691623914987, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951922564650886, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951190919382498, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955055067548528, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962561921274755, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965279757161625, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979152371670352, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982406551280292, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985063681669999, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990679923575954, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995700567960739, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.35.block_sparse_moe": [ { "accuracy": 0.9690400795079768, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9699326227419078, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9733830271288753, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9745970265939832, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9845760142197832, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9857913993764669, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9872660749824718, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991889838303905, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926044496824034, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921595221967436, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9931281066965312, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9960057754069567, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965688718075398, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977993867942132, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979614910844248, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982640553935198, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993518529081484, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.36.self_attn": [ { "accuracy": 0.9787098532542586, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9793474259786308, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9819066780619323, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9860884549561888, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.98881882999558, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9891756628639996, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9934267548960634, "total_bits": 355869696, "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 } }, { "accuracy": 0.9936794530949555, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941596776479855, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944502438884228, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944322719820775, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948283254634589, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954769511241466, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958198785316199, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974976122903172, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978923742310144, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982687165902462, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988244162377669, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994992773608828, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.36.block_sparse_moe": [ { "accuracy": 0.970039184903726, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9709472642280161, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9742502318695188, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9753717759158462, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9851824273355305, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9864117606775835, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9877992336405441, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9922791110584512, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9929407733725384, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9925190504873171, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9934463142999448, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961951860168483, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967315350659192, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979108223487856, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980602922005346, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983409546111943, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993953744706232, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.37.self_attn": [ { "accuracy": 0.9766775544267148, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9775830616708845, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9799012634903193, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9848740205634385, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9881345726316795, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.988331546774134, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928575190133415, "total_bits": 355869696, "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 } }, { "accuracy": 0.9931452636956237, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936328720068559, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939902697806247, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994035251729656, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944245326332748, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9949739472358488, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953495360678062, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997194050316466, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976542755030096, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980831871798728, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986705896735657, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994484289345564, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.37.block_sparse_moe": [ { "accuracy": 0.9718282299581915, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9726907671429217, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9757621355820447, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9768147191498429, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9860690091736615, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9872248615138233, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9884939588373527, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9927196801872924, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9933551102876663, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.992953478533309, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9938279053312726, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964117159543093, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969158831809182, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980156959500164, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981679740158143, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984235071169678, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999406645722047, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.38.self_attn": [ { "accuracy": 0.9826144999824464, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9834842614363879, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.985779325128533, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884607271524146, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910596602130681, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912410049000755, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947742980439216, "total_bits": 355869696, "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 } }, { "accuracy": 0.9948853309615515, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954190818825737, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956049016327597, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9955141157261096, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958994355692994, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965409505821299, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968965119333006, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980959575041197, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984225692314794, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985960813937709, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999114496204129, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995918927670573, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.38.block_sparse_moe": [ { "accuracy": 0.9740485057700425, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9748191628605127, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9776724940165877, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9786265112925321, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9870927860029042, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9881172673776746, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9892781473463401, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932349390583113, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938229133258574, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9934377040481195, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9942538236500695, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966559602180496, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971288368396927, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981570821109926, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982886713114567, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985226527351188, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994496677609277, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.39.self_attn": [ { "accuracy": 0.9809031940530986, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9820912096183747, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9845717053394765, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9885258339345455, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9903906016843393, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9906337910797447, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941531883087009, "total_bits": 355869696, "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 } }, { "accuracy": 0.9943814373691566, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948219265206717, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995043873321265, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995131537609268, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954289168817922, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996115850401111, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963876332622021, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997840528289089, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981884626467945, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984484759916086, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989574724895647, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995396429767425, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.39.block_sparse_moe": [ { "accuracy": 0.9741069986484945, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9748540909495205, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9776839406695217, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9786184879485518, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9871093513211235, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9881234295899048, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9892806890420616, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932625559158623, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938486877363175, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.993450301408302, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9942593463347293, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966664802632295, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971363874792587, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981745113182114, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982983096706448, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985300477419514, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994666700549715, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.40.self_attn": [ { "accuracy": 0.9817298171110451, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9828213618602604, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9852342277299613, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9886497312691063, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910185522167012, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9911473740357906, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943662867299281, "total_bits": 355869696, "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 } }, { "accuracy": 0.994545585825108, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949609903269447, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951689832960255, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953921020496637, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956920019467361, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962606577028055, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996555879333755, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979310634953436, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982715588266728, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984815378702478, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989852060461999, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995591341321415, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.40.block_sparse_moe": [ { "accuracy": 0.9745844439603388, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9753250412177294, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9780422747135162, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9789424340706319, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9873185579199344, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9883290349971503, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9894527739379555, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9933787939953618, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9939540688064881, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9935509389033541, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943554404308088, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967115427425597, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971758114988916, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981882304273313, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983168426842894, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985370088979835, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994597534459899, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.41.self_attn": [ { "accuracy": 0.9819913599640131, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9827944303397089, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9856092397822067, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9889750484144315, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9908515304559842, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909662363352254, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943283952306956, "total_bits": 355869696, "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 } }, { "accuracy": 0.994608890148811, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949892205768265, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952005377854221, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953645113273524, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956557700061239, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996331151254708, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966245229006745, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979645662533585, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998299869170296, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984989631921053, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990091939762351, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995355255450704, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.41.block_sparse_moe": [ { "accuracy": 0.976340051041916, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9770164974033833, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9796015124302357, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9804350072517991, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9882111654151231, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9891492052702233, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9902009866200387, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938580869347788, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943919795332476, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9940119410748594, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9947534637176432, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969510827213526, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973815183911938, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983326322108041, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984370497113559, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986443083325867, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994957271192106, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.42.self_attn": [ { "accuracy": 0.9820759920403361, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9832755448296666, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9851293171523139, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9883353175828233, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907899901736528, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910336091415957, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943909504218027, "total_bits": 355869696, "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 } }, { "accuracy": 0.9946770104579628, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950174447149038, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952226686291397, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954102131305262, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956949090119451, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962256934377365, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965625661134254, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979277100937907, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982802388258278, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985197637870442, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990085075478419, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995596593980736, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.42.block_sparse_moe": [ { "accuracy": 0.9767163896467537, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9773783308919519, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9798822763841599, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9806803199462593, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9883973402902484, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9893150610150769, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9903318993747234, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9939549383125268, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9944832478067838, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9941048828768544, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.994833063276019, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969994190905709, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974211423250381, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983577179955319, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984619617316639, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986624698503874, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995035939609807, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.43.self_attn": [ { "accuracy": 0.9882487936411053, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989091973635368, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910719625186175, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926537038409151, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9939520877087489, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941815062775277, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961151315364987, "total_bits": 355869696, "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 } }, { "accuracy": 0.9964045941014774, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996690860891249, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9968309347459581, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9969616262242198, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972299717774149, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977226696209982, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979429855011404, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987212123960489, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998944149716408, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998980562617362, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993828012302401, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996858962040278, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.43.block_sparse_moe": [ { "accuracy": 0.9777305603493005, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9783826407510787, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9807939259335399, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9815523845609277, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.988925356999971, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9898038069950417, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9907643113983795, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942391471704468, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947339760838076, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943753504776396, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950702171772718, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971376073663123, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975399676477537, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984336935158353, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985320441192016, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998718352697324, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995263804958086, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.44.self_attn": [ { "accuracy": 0.9822271273005754, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9831328967120498, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9859025798505172, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9885525583522394, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910926000447944, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9913306009257212, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944758285419084, "total_bits": 355869696, "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 } }, { "accuracy": 0.99458870961098, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950703355134465, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950762676890008, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951590820564888, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956555483513512, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963773259369191, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967237449018285, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979428953811293, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982380965957418, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984830435714684, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990330993823591, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995378492858435, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.44.block_sparse_moe": [ { "accuracy": 0.9783985109534115, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9790288840886205, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9813161508645862, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9820305661996827, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989275686093606, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9901187854120508, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9910359803470783, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9944210237590596, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9948999449261464, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9945538921747357, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9952266049222089, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972298052744009, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976183466787916, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984861861157697, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985812987142708, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987599740852602, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995515489972604, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.45.self_attn": [ { "accuracy": 0.9829988728743047, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9836993946228176, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9859675165498629, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9891548833111301, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914310746826231, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9916628350620158, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947941385325976, "total_bits": 355869696, "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 } }, { "accuracy": 0.9950700798071921, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954315123613924, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956778314081021, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956593003007583, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960046112828422, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996420787909301, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967356213892344, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980148767062929, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983402353827842, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986093298211927, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990754486352671, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995896062391694, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.45.block_sparse_moe": [ { "accuracy": 0.979063643142581, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9796762252226472, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9818517647217959, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9825388208264485, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9895945460302755, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9904099741252139, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.991289759404026, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9945816745166667, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9950510297203436, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947150347870775, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995364953414537, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973085985984653, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976855020795483, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985259723034687, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998616720069549, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987881825945806, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999543739562796, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.46.self_attn": [ { "accuracy": 0.9849754574242979, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9861899997340515, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.988571178750135, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9911931931274012, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9925308405654505, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928777790628374, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955011433339678, "total_bits": 355869696, "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 } }, { "accuracy": 0.995772430265788, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9960861176077742, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996265982335899, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9963275799818803, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965634563704953, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971396258915775, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973638587980531, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983962041005725, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986517421202734, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988001988385804, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992309665904031, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996243754903844, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.46.block_sparse_moe": [ { "accuracy": 0.979904098669067, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9804892879910767, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9825840581906959, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9832379176514223, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989992810646072, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9907836191123351, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9916339737246744, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994795388658531, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9952472658478655, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9949185288860463, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955442786449566, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974110060720704, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977726729703136, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985808596829884, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9986682763556018, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988324041914893, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995605831391003, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.47.self_attn": [ { "accuracy": 0.9871510601369664, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9874224353116006, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9902189879212528, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.992191236349754, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9934498383081518, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9933943059295416, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960922978643794, "total_bits": 355869696, "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 } }, { "accuracy": 0.9962950361077674, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965638032008428, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996667616564082, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9966824934817851, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968675305426586, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974928718584124, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976877492154017, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985997663607122, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988355453679105, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989727188803954, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993901626949082, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996787096897606, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.47.block_sparse_moe": [ { "accuracy": 0.9790978257078677, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9797169589437544, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9818678845185786, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9825541579630226, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9896087794331834, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9904229479143396, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9912985457340255, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9945945600047708, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9950602181488648, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947273848229088, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953737792675383, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973194012127351, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976924668007996, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985356349352514, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9986263432510896, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987966550979763, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995638118671195, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.48.self_attn": [ { "accuracy": 0.9826987439300865, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9834984469925985, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9863935533212498, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899518068414181, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9911719892406836, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912532815942541, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947807799326256, "total_bits": 355869696, "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 } }, { "accuracy": 0.9949583102716133, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953925940208137, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956273713032715, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956049073953182, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995922694040928, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996541720291134, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968339938786812, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980544489371823, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983857266342966, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998588227754226, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991104086875566, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999589000861306, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.48.block_sparse_moe": [ { "accuracy": 0.9779906200710684, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9786708005703986, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9808174604550004, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9815340654458851, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890727452002466, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9899335211375728, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.990819436730817, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942743937135674, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947672212147154, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9944478385150433, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995131301227957, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971677358262241, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975614554423373, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984288946579909, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998545849826769, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987210096005583, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999514576717047, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.49.self_attn": [ { "accuracy": 0.9827718904707581, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9838012986583635, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9870656467974186, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9894167347811162, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909910717979074, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9913414549082518, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946123625850305, "total_bits": 355869696, "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 } }, { "accuracy": 0.9947871219483204, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953576276893727, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954864645842463, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9955389287788421, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958627856103703, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966686260595452, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969413998478558, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981400713586481, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984269702108577, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985940713813761, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991327270836337, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995787860207201, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.49.block_sparse_moe": [ { "accuracy": 0.9767281964886934, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9774407141376287, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9796593249775469, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9804324705619365, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9884455156279728, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9893553431611508, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9902831260114908, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9939246828434989, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9944545715115964, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9941229277173989, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948493631673045, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970066238020081, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974238691502251, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983418279589387, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984663534851279, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986520147358533, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994925657229032, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.50.self_attn": [ { "accuracy": 0.9789161623921245, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9796130401082337, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9820957095362246, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9859595614252612, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9890330131165683, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893829048378393, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937431616708636, "total_bits": 355869696, "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 } }, { "accuracy": 0.9939472871483304, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944591191597283, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947365029365756, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945732899359427, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9949129170272499, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955843454808928, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960222022782546, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975842376297805, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980326539080124, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983028724818723, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989198195980862, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995103082510468, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.50.block_sparse_moe": [ { "accuracy": 0.9750268717762083, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9758076802827418, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9781274527776986, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9789733760990202, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9876220637233928, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9886026318417862, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9896144844824448, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9934755603317171, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9940416814060882, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937155377701856, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944909288315102, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967967561678961, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972438388795126, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982160001964075, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983580716798315, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985639500664547, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994523539062357, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.51.self_attn": [ { "accuracy": 0.9756529489532113, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9782411225605756, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.980779470410198, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9851913121528924, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879155851667747, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884233712218702, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931295011774637, "total_bits": 355869696, "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 } }, { "accuracy": 0.9932475524256006, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937648325576447, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9940173911163583, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939813126111403, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943303095642477, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951738138333894, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955395310535096, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973364872275852, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977486838470213, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981559684238164, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987840297690127, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994642042584019, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.51.block_sparse_moe": [ { "accuracy": 0.9751359764486551, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9759043126832694, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9782213703729212, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9790870877914131, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9875751959625632, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9885662683518603, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9896139582851902, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9934256189735606, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9940047271666117, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936709988396615, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944707279209979, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967802545870654, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997235466784332, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982030774990562, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983575978985755, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985730178887025, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994614482711768, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.52.self_attn": [ { "accuracy": 0.9771843722555786, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9789023534394801, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9816723491530865, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.985873308731243, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9887709486065432, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9891982865519822, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.993301008711569, "total_bits": 355869696, "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 } }, { "accuracy": 0.9936058775056154, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9940576872322708, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943766398937441, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944161715102382, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948422277229838, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954178995685652, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957960325409658, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974713142728433, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978700995561667, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998262464243453, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988150858407607, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994765778683359, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.52.block_sparse_moe": [ { "accuracy": 0.9749560605268925, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9757238975726068, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9780476205050945, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9789254041388631, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9874316218774766, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9884324158774689, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9895010163309053, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9933177912607789, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938980612787418, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9935903485165909, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943896543118171, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996724695462035, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971918342635036, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998153393418761, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983237811975414, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985468418308301, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994259617160424, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.53.self_attn": [ { "accuracy": 0.9739071875810623, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9757708827964962, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9795227183494717, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9840898814145476, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9873202784219757, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9874675835017115, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924947217223234, "total_bits": 355869696, "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 } }, { "accuracy": 0.9926836787490174, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9933232925832272, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934583074063994, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935902151628397, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941173683619127, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9949566589202732, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953625329071656, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971791064017452, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976601832313463, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979717844107654, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987131761445198, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999423460001708, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.53.block_sparse_moe": [ { "accuracy": 0.9760291697457433, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.976734175812453, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9789497246965766, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.979790651705116, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9879873177269474, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9889464265434071, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9899752706987783, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936471035471186, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942208313150331, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938883560826071, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9946485536056571, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968757320311852, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973188158473931, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982609740109183, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983966287836665, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998614490905311, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994665934500517, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.54.self_attn": [ { "accuracy": 0.974317968590185, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9755392840597779, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9804388643242419, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9852105572354048, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9877969052176923, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9880441161803901, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929181933403015, "total_bits": 355869696, "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 } }, { "accuracy": 0.9931263336911798, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936451630783267, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9940020316280425, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938440104015172, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942599774803966, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950565628823824, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954780618427321, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972792616754305, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977194181992672, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981600997271016, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987883786670864, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994572844952927, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.54.block_sparse_moe": [ { "accuracy": 0.9751338993664831, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9758391480427235, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9779100951272994, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9787347104866058, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9877152645494789, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9887175095500425, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.989714840427041, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9934140174882486, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9940451321890578, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937366690137424, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9945201243972406, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967814334086142, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972423756844364, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981807564035989, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99834740908409, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985682198603172, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994250476302113, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.55.self_attn": [ { "accuracy": 0.975923465564847, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9746030271053314, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9799166347365826, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9825188165996224, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882242061430588, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9890852348180488, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926541475579143, "total_bits": 355869696, "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 } }, { "accuracy": 0.9925869273138233, "total_bits": 356068608, "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": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932403223938309, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934500402887352, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994682012940757, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951698750373907, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956098768743686, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958959943032824, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974180734716356, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976703880820423, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982345232274383, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986328049999429, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993934345984599, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.55.block_sparse_moe": [ { "accuracy": 0.975224549183622, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9759399483446032, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9779096976853907, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9787495380733162, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9881880135508254, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.989127003820613, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9900289266370237, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9935723193339072, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942190403235145, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9939712976920418, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.994718203204684, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996847878646804, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972838017274626, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981321484374348, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983498699148186, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985684413113631, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992750603123568, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.norm.norm": null, "lm_head.linear": null }, "last_module_idx": 114 }