{ "measurement": [ { "key": "model.layers.0.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.07641400396823883, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.07601680606603622, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.034384675323963165, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.03436262533068657, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.034337371587753296, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.018368782475590706, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.09288815408945084, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.07588130235671997, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.03438465669751167, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03431810811161995, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03734111785888672, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.040665626525878906, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.03430860489606857, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.02248825505375862, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.018359430134296417, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.022294435650110245, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.01835794933140278, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.016717171296477318, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.018357614055275917, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.016715509817004204, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.018047986552119255, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.018357248976826668, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.01636589877307415, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.01671483740210533, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.016717171296477318, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.01636589877307415, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.0.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.07152121514081955, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.07111835479736328, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.03490643948316574, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.03488767147064209, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.03486230596899986, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.017600582912564278, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.07552827894687653, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.07098104804754257, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.0348873995244503, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.03484304994344711, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.03669469803571701, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.040635284036397934, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.03483337536454201, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.02141687273979187, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.01759493350982666, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.02165341004729271, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.01759132742881775, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.01573777198791504, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.01759147085249424, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.015739748254418373, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.017431510612368584, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01759149879217148, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.01551962923258543, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.01573871821165085, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.017600582912564278, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.01551962923258543, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.0.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.11890317499637604, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.1085345670580864, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.06914885342121124, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.0670907273888588, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.06625178456306458, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.022176984697580338, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.1354496031999588, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.10617928951978683, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.06899571418762207, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.06484388560056686, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.0781732052564621, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.08880168944597244, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.06443270295858383, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.03663412854075432, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.019298143684864044, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.04825544357299805, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.017867444083094597, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.009044349193572998, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.01758347637951374, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.008198452182114124, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.029336679726839066, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.017510343343019485, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.00896428618580103, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.007821604609489441, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.017867444083094597, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.007821604609489441, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.0.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.02647339552640915, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.01618628017604351, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.008663843385875225, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.010659802705049515, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.010588687844574451, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.0047929547727108, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.017681635916233063, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.01575743407011032, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.012270588427782059, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.007846997119486332, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.008374808356165886, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.008873432874679565, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.007825515232980251, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.005670447368174791, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.005062310956418514, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.004619808867573738, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.003799515776336193, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.003510575508698821, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.003438309533521533, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.003100357251241803, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.002822467591613531, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.0034362562000751495, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.002215492771938443, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0029719779267907143, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.01618628017604351, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.0034362562000751495, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.0.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.1236632764339447, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.1158457100391388, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.10699900984764099, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.09618619829416275, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.05720817297697067, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.05159953236579895, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.07648096233606339, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.06857415288686752, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.058034930378198624, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.051996733993291855, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.04928875342011452, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.03778128698468208, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.03331073373556137, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.028501387685537338, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.027243779972195625, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.019731132313609123, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.016570165753364563, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.015834029763936996, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01570274494588375, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.014867385849356651, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012213376350700855, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.012897689826786518, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.010880104266107082, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010486340150237083, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.016570165753364563, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010486340150237083, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.0.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.16052651405334473, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.15057715773582458, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.14424799382686615, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1294216364622116, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.07423458993434906, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.0693059116601944, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.0891176238656044, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.08162614703178406, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.07527776807546616, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.06703750789165497, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.06319844722747803, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04513605684041977, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.038832202553749084, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.035556353628635406, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.03474479541182518, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.022629685699939728, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.01835818588733673, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.01789146475493908, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.016923315823078156, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.016326939687132835, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012000862509012222, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.011596111580729485, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.010809189639985561, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.007555660791695118, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.016923315823078156, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.007555660791695118, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.0.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.11449969559907913, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.09486611187458038, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.08697254210710526, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.07497461140155792, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.050039637833833694, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.0424744188785553, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.0620654933154583, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.05630214884877205, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.0528503842651844, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.04027929902076721, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.03767968341708183, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.03159726411104202, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.027188347652554512, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.02436627820134163, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.02367827109992504, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.015934979543089867, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.01333374809473753, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.013111221604049206, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.01156336534768343, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.011113027110695839, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.008780568838119507, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.00944916345179081, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.007831424474716187, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.007243161089718342, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.015934979543089867, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.007243161089718342, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.1.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.040272992104291916, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.037298526614904404, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.01740342192351818, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.017179163172841072, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.01693282462656498, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.008264021947979927, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.043522294610738754, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.03670565038919449, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.01780669204890728, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.01631326600909233, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.017343400046229362, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.018516669049859047, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.016208985820412636, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.009967708960175514, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.007875651121139526, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.010242336429655552, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.007548713590949774, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.006930855102837086, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.007480635307729244, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.006836962886154652, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.007286360487341881, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.007470409851521254, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006352486088871956, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0068010506220161915, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.01740342192351818, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006352486088871956, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.1.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.031036393716931343, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.029777899384498596, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.012429476715624332, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.012328345328569412, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.012214687652885914, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.005671610124409199, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.03793752193450928, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.02948402613401413, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.012546302750706673, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.01199065800756216, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.012233082205057144, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.014177754521369934, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.011945012956857681, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.007161174900829792, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.005516164004802704, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.007192548830062151, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.005396160297095776, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.0049020955339074135, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.005372721701860428, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.004868877585977316, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.0053033665753901005, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005368481855839491, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004701798781752586, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.004855500999838114, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.012429476715624332, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004701798781752586, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.1.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.18125450611114502, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.1347997933626175, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.10123487561941147, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.08918863534927368, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.07649372518062592, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.04894934594631195, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.1329571008682251, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.11041833460330963, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.0879664272069931, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.06158338114619255, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.06499896943569183, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.07195772975683212, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.05508772283792496, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.03873220086097717, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.032283343374729156, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.0385032519698143, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.020526252686977386, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.017339292913675308, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.017756015062332153, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.013398789800703526, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.02094927430152893, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.016636062413454056, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.011108752340078354, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.009930460713803768, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.017339292913675308, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.009930460713803768, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.1.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.1746339350938797, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.13253483176231384, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.10914776474237442, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.09009844809770584, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.07544469833374023, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.05547596514225006, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.11028004437685013, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.09751075506210327, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.08307540416717529, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.05666794627904892, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.05598682165145874, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.057414665818214417, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04805555194616318, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.038105327636003494, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.03541064262390137, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.029759464785456657, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.022409949451684952, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.0209815576672554, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.019350621849298477, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01771029829978943, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.017278308048844337, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.018115878105163574, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.013800781220197678, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0147591857239604, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01771029829978943, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.013800781220197678, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.1.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.19818857312202454, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.18529216945171356, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.1805700957775116, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.16125157475471497, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.0909760594367981, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.08658991008996964, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.1031600758433342, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.09463897347450256, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.09229158610105515, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.0817110538482666, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.07637243717908859, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.05229361727833748, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.045163728296756744, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.04358852654695511, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04320983216166496, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.026212178170681, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02275863289833069, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.022530365735292435, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.020963089540600777, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.020618915557861328, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.014189695939421654, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.014304596930742264, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.013617997989058495, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010059477761387825, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.014189695939421654, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010059477761387825, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.1.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.2325088381767273, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.219081312417984, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.2142696976661682, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.19247843325138092, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.1077970638871193, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10317306965589523, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12071733176708221, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11119914054870605, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10910562425851822, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09751714020967484, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09136271476745605, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.061231039464473724, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05296967178583145, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.051460493355989456, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05110475420951843, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03056972660124302, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.026308000087738037, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.026088561862707138, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.024230191484093666, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02388029918074608, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01603737473487854, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015674689784646034, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015470646321773529, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010084621608257294, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01603737473487854, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010084621608257294, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.1.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.18748915195465088, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.17079319059848785, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.16387629508972168, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.14997299015522003, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.08416969329118729, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.0778108760714531, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.09915880858898163, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.090552419424057, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.08645385503768921, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.07540430128574371, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.07305256277322769, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.050370536744594574, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.04329442232847214, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.04043077677488327, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.03973555192351341, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.02533666417002678, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.02138538472354412, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.02112228237092495, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.01972867175936699, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.019300101324915886, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.013743431307375431, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.014055748470127583, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.01273341104388237, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.010151514783501625, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.013743431307375431, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.010151514783501625, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.2.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.03954315185546875, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.03392047435045242, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.023342834785580635, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.020915143191814423, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.017361778765916824, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.01092896144837141, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.032490819692611694, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.029032697901129723, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.018645351752638817, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.015273070894181728, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.016026806086301804, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.01608453132212162, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.013761928305029869, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.008830062113702297, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.007222129963338375, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.008115334436297417, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.005261150188744068, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.00449743214994669, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.004922708962112665, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.00405163923278451, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.004511923994868994, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.004690840840339661, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.003024175763130188, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.003547379979863763, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.017361778765916824, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.003547379979863763, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.2.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.02788083255290985, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.024675384163856506, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.014896108768880367, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.013772028498351574, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.012191259302198887, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.006499377079308033, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.025540126487612724, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.022518575191497803, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.01290440745651722, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.011181464418768883, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.01182965561747551, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.012202979065477848, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.01059003733098507, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.006260172463953495, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.00472131185233593, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.006141013465821743, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.003786182263866067, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.0030821249820291996, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.0036214457359164953, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.0028608280699700117, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.0035616469103842974, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.003528701374307275, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.0024143902119249105, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.002645698608830571, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.014896108768880367, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.003528701374307275, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.2.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.20301119983196259, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.16779719293117523, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.14743757247924805, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.12691184878349304, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.09105831384658813, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.07302037626504898, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.13871115446090698, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.113014355301857, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.09732139855623245, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.07360643148422241, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.07174483686685562, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.07384175062179565, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.0545528344810009, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04385905712842941, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.041096281260252, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.03814317658543587, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.022726496681571007, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.021144913509488106, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.019207987934350967, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.017056375741958618, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.019417233765125275, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01555765513330698, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.012488839216530323, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00930135790258646, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.017056375741958618, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00930135790258646, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.2.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.18672971427440643, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.146916463971138, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.131308913230896, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.10773329436779022, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.08351820707321167, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.067070372402668, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.1081494614481926, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.09535501897335052, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.08814769238233566, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.061727624386548996, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.05811260640621185, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.055864039808511734, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.047136250883340836, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.041925009340047836, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.04061463102698326, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.028867896646261215, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.024083402007818222, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.023343227803707123, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.020350491628050804, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.019412169232964516, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.016822341829538345, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.018205802887678146, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.014936329796910286, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.014975976198911667, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.016822341829538345, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.014936329796910286, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.2.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.23229187726974487, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.21859438717365265, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.21378619968891144, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.19131670892238617, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10919062048196793, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10432571917772293, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12252125144004822, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11245699226856232, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.11036043614149094, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09820856899023056, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.0918712243437767, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06260313838720322, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.055254656821489334, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05389257147908211, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05356653034687042, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03188721835613251, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.030341746285557747, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.03015269711613655, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.028428316116333008, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.028117798268795013, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01836436428129673, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.021736018359661102, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.0178839024156332, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.018095849081873894, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.0178839024156332, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.0178839024156332, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.2.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.25964105129241943, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.2443583458662033, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.23937241733074188, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.21419896185398102, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.12195371091365814, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.11663971096277237, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.1364234983921051, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.12536285817623138, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.12352903932332993, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.10972459614276886, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.10266165435314178, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06995304673910141, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.06038236618041992, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.058881212025880814, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05850392207503319, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03512130677700043, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.031091896817088127, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.030880985781550407, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02871941216289997, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02833692729473114, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01932113617658615, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01991952210664749, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.01875084824860096, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.014612700790166855, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.014612700790166855, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.014612700790166855, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.2.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.059484001249074936, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.05719449371099472, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.03634709492325783, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.03448699414730072, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.029725853353738785, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.0148357218131423, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.04172908514738083, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.05247580260038376, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.029975131154060364, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.02868293598294258, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.02420075610280037, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.028612397611141205, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.027013519778847694, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.015926582738757133, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.010590994730591774, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.015444450080394745, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.009019628167152405, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.00411559920758009, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.008865071460604668, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.003779449500143528, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.009117858484387398, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.008572611957788467, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.0032547898590564728, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.002809440717101097, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.0148357218131423, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.003779449500143528, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } } }, { "key": "model.layers.3.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.05191606655716896, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.04648860916495323, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.04247572273015976, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.03728925436735153, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.023795537650585175, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.02073335088789463, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.03153562173247337, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.028628380969166756, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.024655772373080254, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.020567918196320534, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.019529225304722786, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.015887290239334106, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.013658764772117138, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.011497409082949162, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.010939168743789196, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.007949820719659328, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.006039395462721586, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.005729358643293381, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.005406915210187435, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.005013438407331705, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.004186646081507206, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.0041384631767869, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.003418977139517665, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.002752852626144886, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.015887290239334106, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.003418977139517665, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.3.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.04636731371283531, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.04073968902230263, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.03622349351644516, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.03155682981014252, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.020813612267374992, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.017478205263614655, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.028714269399642944, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.02629188820719719, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.021753722801804543, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.017821896821260452, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.017070936039090157, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.014450931921601295, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.01248122937977314, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.010046465322375298, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.009389384649693966, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.0071813552640378475, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.005229525733739138, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.004864929709583521, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.004625987261533737, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.004172487184405327, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.003730284981429577, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.00362378335557878, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.0028879165183752775, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0022621212992817163, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.017821896821260452, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.00362378335557878, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.3.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.20713359117507935, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.18582400679588318, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.1769651174545288, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.15486100316047668, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.09540457278490067, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.08674418181180954, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.11629216372966766, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.10424244403839111, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.09863444417715073, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08178184926509857, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.07663428783416748, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.05934591591358185, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.04983954504132271, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.0458013191819191, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.04478349909186363, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.029671238735318184, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.02331935055553913, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.02275952510535717, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.020500726997852325, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.01966525986790657, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.015113023109734058, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.014234110713005066, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.01310903113335371, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008566717617213726, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.015113023109734058, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008566717617213726, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.3.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.17725691199302673, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.1530110090970993, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.14291983842849731, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.11982709914445877, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.07914222031831741, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.07022394239902496, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.099281907081604, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.08958084881305695, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.0835113599896431, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.06522373855113983, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.06035463511943817, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.05086157098412514, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04328087717294693, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.038596972823143005, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.03741461783647537, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.025734243914484978, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.020902886986732483, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.02031814306974411, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.01832488738000393, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.017409414052963257, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.014250986278057098, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.014626400545239449, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.01257608737796545, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.01093136053532362, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.017409414052963257, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.01093136053532362, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.3.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.23329603672027588, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.2185467928647995, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.21356934309005737, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1905578076839447, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.1094917505979538, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10411247611045837, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12265373021364212, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11341510713100433, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.11114638298749924, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09787685424089432, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09103396534919739, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.0624130442738533, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05425383523106575, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.052500564604997635, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.0520910769701004, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.031196221709251404, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02697310969233513, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.02671147510409355, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.0245978981256485, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.024192046374082565, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01631963811814785, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01627175696194172, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015660203993320465, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010662727057933807, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01631963811814785, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010662727057933807, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.3.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.26487264037132263, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.248546302318573, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.24301742017269135, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.21710675954818726, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.1244521364569664, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.11840498447418213, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.13932916522026062, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.12872616946697235, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.12626710534095764, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.1113448515534401, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.10358885675668716, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.07087094336748123, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.06152251362800598, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05961151421070099, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05915248021483421, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.0354020856320858, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.030494602397084236, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.030224159359931946, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.027813296765089035, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.027359485626220703, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.018435483798384666, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.018211131915450096, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.017728785052895546, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.011696642264723778, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.017728785052895546, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.011696642264723778, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.3.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.22479397058486938, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.19734136760234833, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.1866205930709839, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.16586950421333313, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10145844519138336, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09090501070022583, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.121754951775074, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11084504425525665, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.10541293770074844, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.08649791777133942, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08246766030788422, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06184117868542671, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05314687266945839, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.04894224554300308, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.047885578125715256, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03110986389219761, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.02586018294095993, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.025522898882627487, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.022969307377934456, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.022307945415377617, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.016899893060326576, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.017061887308955193, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.015415196307003498, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012181871570646763, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.016899893060326576, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012181871570646763, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.4.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.071680948138237, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.06545154005289078, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.06123833358287811, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.054103363305330276, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.033163655549287796, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.02984917350113392, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.04233965277671814, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.03822862356901169, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.034074701368808746, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.02912500686943531, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.02750207856297493, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.021487044170498848, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.018311088904738426, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.016002394258975983, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.015420175157487392, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.010747873224318027, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.008339750580489635, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.008011299185454845, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.007538326550275087, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.007118252106010914, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.005594030953943729, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.0055227940902113914, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.004699446260929108, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.003638689639046788, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.016002394258975983, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.003638689639046788, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.4.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.06917345523834229, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.061973489820957184, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.05532582104206085, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.048534095287323, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.031145311892032623, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.026608578860759735, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04302160069346428, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.039186060428619385, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03227551653981209, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.027107015252113342, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.026005180552601814, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.021580392494797707, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.018575819209218025, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.014978116378188133, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.014049737714231014, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.010824264027178288, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.007801651023328304, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007233024574816227, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.007007946725934744, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006310007534921169, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005542327184230089, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005408565979450941, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004249237477779388, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.003315893467515707, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.014978116378188133, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004249237477779388, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.4.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.21408598124980927, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.1952296793460846, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.1873471587896347, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.16516627371311188, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.09889322519302368, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.09123232960700989, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.1189039796590805, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.10678048431873322, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.10126619040966034, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08629556000232697, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.08071161061525345, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.06070374697446823, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.05113578587770462, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04748021438717842, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.046615324914455414, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.030277226120233536, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.024106206372380257, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.023582499474287033, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.02150060422718525, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.020742101594805717, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.015495043247938156, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.014588851481676102, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.013738101348280907, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00879867747426033, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.015495043247938156, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00879867747426033, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.4.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.20795994997024536, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.16795603930950165, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.14975230395793915, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.12768468260765076, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.09221964329481125, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.07347989827394485, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.12123508751392365, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.10895390063524246, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.09870357066392899, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.07262913882732391, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.06742891669273376, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.06220729276537895, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.052621789276599884, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.04503115266561508, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.04307539761066437, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.031118661165237427, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.0244667436927557, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.0233685914427042, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.02082371897995472, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.0193922221660614, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0169193334877491, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.017507420852780342, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.014055078849196434, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.012949851341545582, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0169193334877491, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.012949851341545582, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.4.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.22937890887260437, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.21499572694301605, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.21005263924598694, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1875789612531662, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10784465819597244, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10264896601438522, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12065048515796661, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11143123358488083, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10940337926149368, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09640113264322281, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.08976632356643677, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06152825057506561, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.053328849375247955, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05172619968652725, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.051334451884031296, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.030749399214982986, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.026663929224014282, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.026428526267409325, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.024366213008761406, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.023966064676642418, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016242556273937225, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.016229363158345222, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015630582347512245, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010824738070368767, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016242556273937225, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010824738070368767, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.4.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.2659521698951721, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.249404639005661, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.24387477338314056, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.21796824038028717, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.12485403567552567, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.1188688799738884, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.1395857334136963, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.12893179059028625, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.1265876293182373, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.11176525056362152, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.10415484011173248, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.07097785174846649, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.061552561819553375, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.0597134567797184, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05927470698952675, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.035442762076854706, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.030447233468294144, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.030181147158145905, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.027757085859775543, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.0273063313215971, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.018438909202814102, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01802271045744419, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.017728179693222046, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.011352191679179668, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.017728179693222046, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.011352191679179668, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.4.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.2362481653690338, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.21104273200035095, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.20112773776054382, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.17813432216644287, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10788282006978989, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09802389144897461, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.12730684876441956, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11643677949905396, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.11136104166507721, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09267207235097885, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08742724359035492, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06476292014122009, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05577341467142105, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05190923437476158, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.05096716433763504, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03246337175369263, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.027236424386501312, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.02691628597676754, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.024280771613121033, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.023691939190030098, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.017460236325860023, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.017620453611016273, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016102580353617668, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012364394031465054, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.017460236325860023, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012364394031465054, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.5.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.06461597979068756, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.05899176001548767, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.05559881031513214, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.0491797998547554, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.02994704619050026, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.027180060744285583, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.03698652982711792, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.03383208066225052, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.03071858361363411, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.026225496083498, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.024647334590554237, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.01873099058866501, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.016134001314640045, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.014402657747268677, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.013974352739751339, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.009368342347443104, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.007475675083696842, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.0072302017360925674, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.006728782784193754, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.006408636923879385, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.00487153185531497, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.004830952733755112, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.004224914126098156, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0031511669512838125, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.016134001314640045, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.004224914126098156, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.5.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.05986633896827698, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.05380847305059433, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.04944925382733345, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.043349962681531906, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.027228038758039474, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.02394077181816101, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.03572779893875122, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.032351527363061905, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.02817464806139469, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.02358398772776127, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.02247355505824089, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.017998266965150833, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.01532571017742157, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.01312203612178564, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.01253435853868723, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.008957948535680771, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.006769008934497833, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.006430457811802626, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.006041296292096376, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.005615979433059692, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.004604815971106291, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.004469778388738632, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.003762957639992237, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0027623765636235476, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.01532571017742157, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.003762957639992237, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.5.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.21173132956027985, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.19277285039424896, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.1851925253868103, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.16345955431461334, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.09773465991020203, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.09029179811477661, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.11622066795825958, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.1053967997431755, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.1000983715057373, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08522097021341324, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.07940977811813354, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.05908330902457237, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.05029774084687233, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04681817814707756, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.046017684042453766, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.029548699036240578, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.02368839830160141, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.023193759843707085, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.02108132094144821, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.020349372178316116, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.015058505348861217, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01421766635030508, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.013546045869588852, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00838445220142603, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.015058505348861217, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00838445220142603, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.5.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.20165391266345978, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.16930881142616272, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.15643154084682465, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.1264498233795166, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.09131541848182678, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.07863206416368484, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.11321362853050232, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.10199302434921265, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.09537962079048157, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.07060094177722931, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.0635172501206398, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.058190420269966125, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04959972947835922, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.04480067268013954, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.0436568409204483, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.029457466676831245, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.024576954543590546, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.02396523207426071, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.020824048668146133, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01988721638917923, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.01655476540327072, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.0174611397087574, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.01488250121474266, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.013534263707697392, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.01655476540327072, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.013534263707697392, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.5.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.21949583292007446, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.2055906504392624, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.20073969662189484, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.17926201224327087, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10312359780073166, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.09799064695835114, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.11603136360645294, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.10696066170930862, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10466122627258301, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09214187413454056, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.08584684878587723, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.0591103658080101, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05121695622801781, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.04950611665844917, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04910392314195633, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.029598155990242958, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02554723434150219, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.025295497849583626, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.023332104086875916, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02294323220849037, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015703817829489708, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015600364655256271, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015054388903081417, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010437041521072388, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015703817829489708, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010437041521072388, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.5.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.26223304867744446, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.2461417317390442, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.2405865341424942, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.2150718718767166, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.12320555001497269, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.11728938668966293, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.13804848492145538, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.12743593752384186, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.12495648115873337, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.1102459728717804, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.10278908908367157, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.07016539573669434, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.060874152928590775, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.058971479535102844, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05852704495191574, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.0350528284907341, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.03005773015320301, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.0297907292842865, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02737955003976822, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.026927873492240906, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01821207068860531, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.017778487876057625, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.017475614324212074, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01118452288210392, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.017778487876057625, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01118452288210392, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.5.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.23529194295406342, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.21064376831054688, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.2011951506137848, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.17896293103694916, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10770674049854279, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09806618094444275, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.12614020705223083, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11596903204917908, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.11109218001365662, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.0929536521434784, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08815085142850876, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06437645852565765, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.0557062029838562, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05187131464481354, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.05092725530266762, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03232914209365845, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.02724326029419899, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.026929670944809914, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.02437930554151535, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.02378705143928528, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.01730322279036045, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.017670568078756332, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.01595635898411274, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.01244552806019783, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.01730322279036045, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.01244552806019783, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.6.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.058908093720674515, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.05265062302350998, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.04718849062919617, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.04167084023356438, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.0269036665558815, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.02290208637714386, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.03733281046152115, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.033796485513448715, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.027941500768065453, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.02344086952507496, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.022486763074994087, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.01896931417286396, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.016154512763023376, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.0130668506026268, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.01223305705934763, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.00949511956423521, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.006941249594092369, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.006488883402198553, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.006267949938774109, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.005712354090064764, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.004967418499290943, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.004953058902174234, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.0038264687173068523, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0033500315621495247, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.016154512763023376, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0033500315621495247, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.6.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.05528733506798744, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.04819094017148018, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.0412893146276474, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.036221716552972794, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.024473335593938828, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.019711937755346298, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.03660329058766365, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.0328444242477417, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.02575058676302433, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.021155934780836105, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.020686568692326546, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.018423620611429214, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.015647554770112038, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.011846187524497509, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.01076932530850172, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.009151794016361237, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.006232078652828932, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.005630962084978819, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.0055916584096848965, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.004847895819693804, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.004734247922897339, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.004565243609249592, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.003307417267933488, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0028238759841769934, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.015647554770112038, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.004565243609249592, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.6.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.1984224021434784, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.18046066164970398, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.17293912172317505, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.1519913524389267, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.09205237030982971, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.08460027724504471, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.11209984868764877, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.0996403843164444, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.09465774148702621, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08003765344619751, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.07473860681056976, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.0568377785384655, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.047710735350847244, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.044129807502031326, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.0433100201189518, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.02844414673745632, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.022335898131132126, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.02187325991690159, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.019870182499289513, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.01914804056286812, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.014457298442721367, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.013449725694954395, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.01267781388014555, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.007956142537295818, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.014457298442721367, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.007956142537295818, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.6.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.18162643909454346, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.154565691947937, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.1427106410264969, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.12317565083503723, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.08271317183971405, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.07081965357065201, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.1044599786400795, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.09360336512327194, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.08630602061748505, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.0670100674033165, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.06269947439432144, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.054119862616062164, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04551868513226509, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.040517933666706085, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.03926144540309906, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.02752615325152874, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.02220737375319004, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.021536381915211678, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.019315769895911217, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01842106319963932, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.015328609384596348, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.015806710347533226, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.013494345359504223, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.012110907584428787, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.015328609384596348, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.012110907584428787, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.6.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.20164436101913452, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.18885137140750885, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.18435940146446228, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.16489215195178986, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.0945446565747261, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.08982349932193756, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.1061173528432846, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.09805147349834442, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.09599608182907104, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.08454132080078125, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.07882186770439148, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.05395013466477394, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.04691511392593384, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.0453232079744339, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04494718462228775, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.0269713643938303, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02330925688147545, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.023084627464413643, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02128986455500126, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.020927278324961662, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.014086318202316761, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.0141350282356143, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.013461880385875702, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009307991713285446, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.014086318202316761, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009307991713285446, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.6.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.2546430230140686, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.23873767256736755, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.23332346975803375, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.20882010459899902, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.11951691657304764, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.1136946827173233, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.1340634524822235, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.12378092855215073, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.12128909677267075, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.1070365160703659, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09989146143198013, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06814972311258316, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05913078412413597, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05721623823046684, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05676577240228653, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03405007719993591, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.029185522347688675, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.028922896832227707, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.026627717539668083, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02617858164012432, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01765373907983303, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01733672060072422, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.016882820054888725, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010952268727123737, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01765373907983303, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010952268727123737, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.6.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.23903381824493408, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.21438582241535187, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.2050573229789734, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.18158556520938873, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.1099802702665329, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.1003882959485054, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.12976649403572083, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11780840903520584, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.11321799457073212, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09461702406406403, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08959096670150757, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06634578108787537, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05680570378899574, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05318741500377655, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.05232971906661987, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.033422209322452545, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.028289344161748886, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.02798948436975479, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.025341233238577843, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.024787871167063713, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.018254993483424187, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.018710188567638397, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016955401748418808, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.01371883973479271, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016955401748418808, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.01371883973479271, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.7.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.07894550263881683, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.07201741635799408, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.06669270247220993, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.05925511568784714, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.036514297127723694, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.032569076865911484, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.04761999100446701, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.04297289252281189, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.03758806735277176, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03216458857059479, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.030675427988171577, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.02423255518078804, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.020715247839689255, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.017794085666537285, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.017056869342923164, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.012188182212412357, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.00959891639649868, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.009200595319271088, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.008770046755671501, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.00828311312943697, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.006583418697118759, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.0067925360053777695, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.005546024069190025, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004951139912009239, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.017794085666537285, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004951139912009239, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.7.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.06604193150997162, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.0594271719455719, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.05459469556808472, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.04799826070666313, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.03005187213420868, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.026269100606441498, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.039209067821502686, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.03603224828839302, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03101734071969986, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.026131564751267433, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.024693386629223824, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.019712697714567184, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.017191845923662186, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.014465784654021263, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.013737663626670837, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.009871621616184711, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.0074494024738669395, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007059294730424881, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.006654025055468082, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006160328164696693, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005116934422403574, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.004960489924997091, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004193075932562351, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0030721311923116446, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.017191845923662186, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004193075932562351, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.7.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.24024450778961182, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.22253385186195374, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.21591155230998993, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.19134055078029633, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.11200666427612305, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.10541743785142899, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.12892061471939087, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.11674163490533829, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.11395087838172913, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.09876712411642075, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.09211452305316925, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.0656404122710228, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.05599436163902283, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.053735725581645966, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.053151752799749374, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.0328805036842823, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.027263794094324112, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.026941094547510147, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.024485712870955467, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.023990631103515625, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.016846762970089912, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01611372083425522, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.015680961310863495, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.009826699271798134, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.016846762970089912, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.009826699271798134, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.7.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.17682096362113953, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.1582251787185669, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.15104913711547852, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.12618666887283325, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.08172554522752762, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.07419116050004959, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.09786766767501831, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.08786149322986603, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.08396777510643005, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.06799157708883286, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.061619777232408524, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.05049304664134979, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.042536601424217224, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.03964734822511673, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.0389876514673233, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.025404617190361023, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.021098777651786804, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.020653383806347847, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.01843508891761303, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01785057596862316, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0136362100020051, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.01400740910321474, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.012506061233580112, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.010241280309855938, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01785057596862316, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.010241280309855938, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.7.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.18158012628555298, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.17045257985591888, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.16622042655944824, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.14899557828903198, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.08533395081758499, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.08102521300315857, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.09648964554071426, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.08898618072271347, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.0865337923169136, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.0764731615781784, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.07155872136354446, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04915613681077957, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.04264645278453827, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.04099813848733902, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04060351848602295, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.024562349542975426, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02128603123128414, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.021050479263067245, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.019527366384863853, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.019167326390743256, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01294011902064085, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.013251288793981075, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.012299614027142525, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009087489917874336, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01294011902064085, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009087489917874336, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.7.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.24013012647628784, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.2255372256040573, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.22019022703170776, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1975945234298706, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.11296318471431732, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10742955654859543, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.1276204138994217, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11743203550577164, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.11452624201774597, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.10139003396034241, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09484941512346268, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06502823531627655, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05612827092409134, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05410967394709587, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05362667888402939, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03249875828623772, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.027705056592822075, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.027429044246673584, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02533838152885437, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.024893682450056076, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01695927418768406, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01664491556584835, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.01615101657807827, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01071578823029995, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01695927418768406, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01071578823029995, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.7.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.07568853348493576, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.06884893774986267, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.05723872035741806, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.05101511999964714, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.03555870056152344, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.027140501886606216, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.05946781858801842, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.04938351362943649, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.03653363138437271, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.031725604087114334, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.031206656247377396, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.029713455587625504, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.02486962266266346, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.017600491642951965, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.015113051980733871, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.014842960983514786, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.00935880746692419, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.008043170906603336, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.008633201941847801, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.007114865351468325, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.007782664615660906, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.007391616702079773, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.005192015785723925, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.00457634124904871, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.017600491642951965, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.00457634124904871, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.8.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.07376240938901901, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.06763165444135666, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.059023600071668625, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.05266709625720978, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.03429071232676506, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.028417013585567474, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.0518815852701664, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.0455755814909935, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.03530583903193474, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.030638709664344788, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.030453942716121674, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.02736986055970192, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.022050637751817703, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.01667131669819355, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.015166392549872398, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013726952485740185, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.008872666396200657, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.008014990016818047, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.008172004483640194, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.007172078359872103, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.007195403799414635, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.0065895733423531055, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.005027592647820711, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004257477819919586, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.01667131669819355, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004257477819919586, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.8.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.06317956745624542, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.0575203076004982, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.04717642813920975, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.042119573801755905, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.02913561463356018, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.021977048367261887, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04868451505899429, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.04225355014204979, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.030095653608441353, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.02623915858566761, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.026374269276857376, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.025423789396882057, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.020703142508864403, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.014073830097913742, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.012069388292729855, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.01267818734049797, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.007351339794695377, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.006244122050702572, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.006774610374122858, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.005489780101925135, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.006510819308459759, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005709887482225895, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.003947425168007612, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0033183812629431486, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.014073830097913742, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.003947425168007612, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.8.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.19615301489830017, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.17818066477775574, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.16772907972335815, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.14854685962200165, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.0912046879529953, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.08177226036787033, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.12695279717445374, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.10358066856861115, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.09402618557214737, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.0799202173948288, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.0767359659075737, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.06739453226327896, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.050317179411649704, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04399409890174866, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.04241503030061722, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.033935580402612686, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.022506233304739, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.02155998907983303, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.02015789784491062, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.018977928906679153, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.01711393892765045, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.014395095407962799, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.012648550793528557, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008659769780933857, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.01711393892765045, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008659769780933857, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.8.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.18650712072849274, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.1527990996837616, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.13621176779270172, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.1147325187921524, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.08376172930002213, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.06870908290147781, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.11223258078098297, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.1002156063914299, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.08898552507162094, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.06595441699028015, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.062132786959409714, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.05788663402199745, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.0488070584833622, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.041490379720926285, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.039622627198696136, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.029476040974259377, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.023327661678195, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.022388963028788567, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.020167773589491844, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01888182945549488, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.016812652349472046, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.017494212836027145, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.014450832270085812, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.013726834207773209, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.016812652349472046, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.013726834207773209, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.8.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.19638420641422272, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.18448981642723083, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.18049871921539307, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1622515469789505, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.09222133457660675, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.08789784461259842, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.10329005122184753, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.09536722302436829, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.09355899691581726, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.08289965242147446, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.07745856046676636, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.052526235580444336, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.04560825973749161, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.04417125880718231, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04383709654211998, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.02626892551779747, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.022701876237988472, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.022498052567243576, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.020797835662961006, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02046157419681549, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01375532615929842, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.013688774779438972, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.013188489712774754, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008974823169410229, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01375532615929842, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008974823169410229, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.8.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.25334757566452026, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.23849844932556152, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.23350848257541656, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.2098134160041809, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.1192307248711586, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.11379565298557281, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.13332584500312805, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.1229524090886116, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.12089891731739044, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.10720135271549225, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.10038778185844421, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06790434569120407, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05883258953690529, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05713798478245735, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05673903971910477, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03397011756896973, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02932155504822731, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.02906930260360241, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02687186747789383, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.026453159749507904, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.017814449965953827, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.017606137320399284, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.0171329528093338, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01149559486657381, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.017814449965953827, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01149559486657381, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.8.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.23041924834251404, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.20564474165439606, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.19481873512268066, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.1729171872138977, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10579634457826614, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09524085372686386, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.12856583297252655, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11662711948156357, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.10936500877141953, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09125953912734985, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08675458282232285, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06591174751520157, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.056371357291936874, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.051251284778118134, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.04995248094201088, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03344331309199333, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.027341090142726898, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.026844702661037445, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.024548085406422615, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.023752771317958832, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.01829805225133896, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.01846323348581791, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016368113458156586, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013448633253574371, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016368113458156586, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013448633253574371, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.9.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.07658596336841583, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.07022621482610703, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.06604797393083572, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.058766383677721024, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.03558136522769928, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.03223097696900368, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.044677332043647766, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.04055106267333031, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.03644951432943344, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.031371619552373886, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.029832977801561356, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.02275845780968666, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.01944485865533352, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.01718456856906414, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.01662534847855568, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.01140405423939228, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.009032994508743286, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.008703825064003468, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.008206374011933804, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.007796408608555794, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.006039662752300501, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.006018610671162605, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.005225829780101776, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004101580008864403, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.01718456856906414, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004101580008864403, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.9.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.05952885374426842, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.05375979468226433, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.049185238778591156, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.04348701238632202, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.02698587439954281, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.023627936840057373, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.03570403531193733, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.03256489709019661, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.02780730091035366, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.02363423816859722, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.022458534687757492, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.018003346398472786, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.015466177836060524, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.01295536756515503, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.012314247898757458, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.008955356664955616, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.006670842412859201, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.006300549954175949, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.005991243291646242, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.005534976255148649, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.004621607717126608, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.0044276053085923195, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.003695230931043625, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0027103545144200325, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.015466177836060524, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.003695230931043625, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.9.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.2014760673046112, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.1864185482263565, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.1805848777294159, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.15959852933883667, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.09366763383150101, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.08785808086395264, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.1098155751824379, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.0990053117275238, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.09559590369462967, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08256156742572784, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.07755637168884277, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.056138500571250916, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.047286272048950195, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04486697167158127, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.04433917999267578, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.028117142617702484, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.022680243477225304, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.022353224456310272, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.020369356498122215, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.01987053081393242, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.014310372062027454, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.013280757702887058, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.01297970674932003, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.007794560398906469, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.014310372062027454, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.007794560398906469, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.9.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.19769489765167236, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.17813149094581604, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.17024071514606476, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.14381670951843262, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.0911317840218544, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.08294148743152618, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.10717473924160004, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.09804651886224747, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.09332386404275894, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.07654022425413132, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.06904304772615433, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.05480964854359627, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04694234952330589, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.04383467510342598, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.04309971258044243, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.0273855272680521, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.022812683135271072, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.022339200600981712, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.019912177696824074, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.019259469583630562, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.014462941326200962, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.014485049061477184, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.01328922901302576, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.009871280752122402, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.014462941326200962, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.009871280752122402, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.9.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.17505916953086853, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.16448529064655304, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.16066491603851318, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1443197876214981, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.0821596011519432, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07815467566251755, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.09235859662294388, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.08536860346794128, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.08340602368116379, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.07387857139110565, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.06913438439369202, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.047079600393772125, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.04088405519723892, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.039415840059518814, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.03907787427306175, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.02353287674486637, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.020343635231256485, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.020154407247900963, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.018671419471502304, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.018350359052419662, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012352561578154564, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.012444563210010529, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.011794865131378174, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008333816193044186, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012352561578154564, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008333816193044186, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.9.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.24205081164836884, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.22759711742401123, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.22270697355270386, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.20015476644039154, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.11372139304876328, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10837526619434357, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12739577889442444, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11761052161455154, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.11536284536123276, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.10230864584445953, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09567713737487793, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06475776433944702, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05620824173092842, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05442868173122406, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05400795489549637, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03232629597187042, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.027802210301160812, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.027551541104912758, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02545023337006569, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02504192292690277, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016734475269913673, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01654800772666931, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.016017675399780273, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010532396845519543, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016734475269913673, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010532396845519543, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.9.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.23668617010116577, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.21250751614570618, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.20258209109306335, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.1790475994348526, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10910452157258987, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09913266450166702, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.1300162672996521, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11848106980323792, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.11240802705287933, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09395135939121246, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08897940814495087, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06674333661794662, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05708920955657959, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05271906033158302, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.05167970806360245, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03358074650168419, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.027885954827070236, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.027496742084622383, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.02494960092008114, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.024275412783026695, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.018160726875066757, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.018423419445753098, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016614245250821114, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013204343616962433, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016614245250821114, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013204343616962433, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.10.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.07696564495563507, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.07111524045467377, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.06743773072957993, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.05998675897717476, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.03582006320357323, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.03288251534104347, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.044152598828077316, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.040129274129867554, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.036595311015844345, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03171956166625023, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03008752502501011, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.022487198933959007, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.01925632916390896, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.017335211858153343, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.01686113141477108, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.011266511864960194, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.009148032404482365, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.008866864256560802, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.008344425819814205, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.00799462292343378, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0060015227645635605, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.006086968816816807, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.005311129614710808, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004243319388478994, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.017335211858153343, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004243319388478994, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.10.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.067692831158638, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06188584119081497, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.05752699449658394, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.050745315849781036, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.030950356274843216, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.027633829042315483, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.03950728848576546, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.03627628833055496, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.031770020723342896, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.027146991342306137, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.025849783793091774, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.019924914464354515, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.017251942306756973, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.014860578812658787, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.014269650913774967, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.009957611560821533, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.007614234462380409, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007267731241881847, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.0068462626077234745, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006402214057743549, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005130068399012089, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.004935034085065126, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004262033849954605, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0029798850882798433, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.017251942306756973, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004262033849954605, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.10.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.21832726895809174, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.20089386403560638, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.19469179213047028, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.1732747107744217, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.1016671285033226, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.09507637470960617, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.12040680646896362, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.10770873725414276, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.10381072759628296, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.0899667739868164, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.08468891680240631, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.06170596554875374, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.05148814618587494, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.048656877130270004, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.047978725284338, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.031015533953905106, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.024598797783255577, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.024208080023527145, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.022085392847657204, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.02149602770805359, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.015813203528523445, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.014532903209328651, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.014160406775772572, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008573850616812706, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.015813203528523445, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008573850616812706, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.10.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.20522643625736237, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.17896310985088348, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.16858689486980438, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.14416278898715973, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.09360998123884201, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.08321549743413925, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.11227325350046158, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.10193822532892227, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.09692880511283875, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.07643056660890579, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.07064394652843475, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.057536669075489044, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04904589429497719, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.0452614426612854, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.044326554983854294, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.028949640691280365, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.023977724835276604, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.023453807458281517, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.020694732666015625, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01993284933269024, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.015608231537044048, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.015913670882582664, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.014214876107871532, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011440622620284557, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.015608231537044048, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011440622620284557, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.10.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.17696712911128998, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.16665469110012054, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.16291269659996033, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.14642365276813507, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.08310461789369583, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07927485555410385, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.09335901588201523, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.08611710369586945, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.08422651141881943, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.0747750848531723, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.07007414847612381, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04761062562465668, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.041202645748853683, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.03987050801515579, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.039550282061100006, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.02381688915193081, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.020604420453310013, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.02040734328329563, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.018932409584522247, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.01863117516040802, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012609351426362991, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01262075174599886, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.012102728709578514, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008504820987582207, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012609351426362991, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008504820987582207, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.10.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.23851294815540314, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.22460515797138214, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.22003173828125, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.19785866141319275, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.11219850927591324, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10716252028942108, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12574191391468048, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.1157047301530838, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.11374787241220474, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.10106957703828812, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09486175328493118, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06414715945720673, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05538233742117882, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05379693955183029, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05341104790568352, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.032046034932136536, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.027659809216856956, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.027421126142144203, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.025389550253748894, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.025009313598275185, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016924962401390076, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01669575646519661, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.016290659084916115, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.011012551374733448, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016924962401390076, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.011012551374733448, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.10.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.22752216458320618, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.20492644608020782, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.1953667253255844, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.1733052283525467, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10519790649414062, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09570758789777756, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.12518510222434998, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11457321047782898, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.10825782269239426, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09095725417137146, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.0863393023610115, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06415543705224991, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.055153924971818924, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05082549899816513, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.049777545034885406, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.0323973223567009, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.02690129354596138, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.026508962735533714, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.02414580248296261, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.023478087037801743, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.0177060104906559, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.01776580512523651, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016232263296842575, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012747075408697128, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.0177060104906559, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012747075408697128, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.11.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.08853564411401749, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.08040380477905273, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.07553204149007797, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.0667848289012909, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.04081530123949051, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.03671658784151077, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.051802270114421844, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.046553291380405426, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.04195333644747734, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03567918762564659, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03399302810430527, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.0264865905046463, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.022413285449147224, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.019744619727134705, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.01909220777451992, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013364491984248161, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.010473404079675674, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.010103575885295868, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.009489336051046848, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.009009947068989277, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.007092870771884918, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.007096005603671074, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.0060371351428329945, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004989235196262598, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013364491984248161, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004989235196262598, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.11.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.07306505739688873, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06568550318479538, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.06101816147565842, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.05367094650864601, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.0331585519015789, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.029411161318421364, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04216334968805313, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.03845185041427612, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03410971164703369, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.028671981766819954, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.027122292667627335, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.021389860659837723, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.018303900957107544, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.01590709201991558, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.015293850563466549, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.010625392198562622, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.008192429319024086, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007837754674255848, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.007291399408131838, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.0068411254324018955, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005506413523107767, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005313384812325239, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.00459546223282814, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0032938215881586075, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.01590709201991558, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.00459546223282814, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.11.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.2030339539051056, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.18877381086349487, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.18300355970859528, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.16303183138370514, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.09495687484741211, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.08934108912944794, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.10954312980175018, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.10000055283308029, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.09661553800106049, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08399683237075806, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.07842886447906494, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.05564359575510025, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.04760827496647835, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04529361426830292, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.0447549968957901, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.027781764045357704, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.022963646799325943, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.022652117535471916, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.020727403461933136, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.020210694521665573, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.01421246025711298, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01351187750697136, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.01311440672725439, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008009436540305614, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.01421246025711298, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008009436540305614, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.11.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.21294386684894562, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.18808415532112122, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.17725138366222382, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.1535702496767044, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.09739597141742706, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.08696462213993073, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.12037006765604019, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.10858158022165298, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.10150676220655441, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.08212867379188538, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.07711663097143173, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.06180594861507416, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.05234472453594208, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.04717453569173813, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.04595140367746353, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.03100569359958172, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.02504877746105194, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.024368880316615105, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.02207934856414795, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.021113872528076172, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.016701336950063705, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.016804572194814682, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.014877427369356155, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011981128714978695, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.016701336950063705, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011981128714978695, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.11.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.174952432513237, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.1645057499408722, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.16078534722328186, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.14432057738304138, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.08237812668085098, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07839137315750122, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.09286748617887497, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.08553721755743027, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.08353377133607864, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.07408817112445831, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.06947192549705505, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04747416451573372, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.04105789214372635, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.039651934057474136, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.039319656789302826, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.023753467947244644, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.020664088428020477, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.020454078912734985, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01900813728570938, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.018703525885939598, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012698418460786343, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.012921866960823536, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.012172077782452106, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009016570635139942, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012698418460786343, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009016570635139942, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.11.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.23841765522956848, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.22440242767333984, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.219559445977211, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1972460299730301, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.11235513538122177, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.1071019247174263, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12594397366046906, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.1159704253077507, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.11385714262723923, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.10102039575576782, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09471730142831802, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06427861005067825, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05549948662519455, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05384492501616478, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05343743413686752, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03213229030370712, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.027652809396386147, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.027400020509958267, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.025342948734760284, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.024933205917477608, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01687845215201378, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.016643589362502098, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.01621636189520359, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010921102948486805, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01687845215201378, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010921102948486805, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.11.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.24206119775772095, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.21834677457809448, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.20810581743717194, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.18445676565170288, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.11220324784517288, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.10200805962085724, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.13334569334983826, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.12217450886964798, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.1154855266213417, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09687595814466476, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.091764435172081, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06817131489515305, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.058836743235588074, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05420072749257088, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.053076840937137604, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.034182798117399216, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.028612343594431877, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.028166942298412323, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.02563357725739479, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.024907633662223816, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.01830093190073967, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.018810681998729706, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016684893518686295, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013367570005357265, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016684893518686295, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013367570005357265, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.12.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.08824422955513, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.08169136941432953, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.07709873467683792, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.06832405179738998, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.04112520441412926, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.03755614534020424, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.05116342380642891, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.04653366282582283, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.041931238025426865, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.036352500319480896, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.034493960440158844, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.026089293882250786, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.022305874153971672, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.019863255321979523, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.019260767847299576, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013052468188107014, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.01041276566684246, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.010064458474516869, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.009476441890001297, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.00903945229947567, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.006920505780726671, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.006894128397107124, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006067677401006222, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0047022816725075245, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013052468188107014, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0047022816725075245, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.12.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.06494270265102386, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.05987720936536789, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.05517742037773132, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.04880556836724281, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.02987390197813511, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.02657068707048893, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.03894912078976631, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.035860706120729446, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.030545711517333984, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.02641970105469227, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.02510211244225502, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.019621331244707108, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.01700805500149727, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.014380284585058689, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.013694306835532188, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.009791508316993713, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.007429995574057102, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007022550329566002, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.006731813307851553, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.0062442077323794365, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005102416034787893, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.004929742775857449, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004190413746982813, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0030410662293434143, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.01700805500149727, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004190413746982813, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.12.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.23760174214839935, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.22065365314483643, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.21484941244125366, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.1894947588443756, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.11164209991693497, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.10495954751968384, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.12923602759838104, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.11576820909976959, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.11342969536781311, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.0979815125465393, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.09157367795705795, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.06650002300739288, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.055393341928720474, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.05337260663509369, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.05289117246866226, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.03321947529911995, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.0268543753772974, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.02653106115758419, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.024027151986956596, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.023535728454589844, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.016925547271966934, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.015533487312495708, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.015480168163776398, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00903639942407608, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.016925547271966934, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00903639942407608, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.12.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.2146478295326233, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.19194087386131287, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.18414756655693054, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.15990839898586273, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.09882879257202148, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.09026553481817245, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.11606556922197342, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.10497596114873886, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.10142960399389267, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.08341991901397705, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.07720406353473663, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.059607259929180145, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.050641532987356186, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.047852374613285065, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.047155071049928665, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.030017904937267303, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.025179557502269745, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.02479463815689087, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.022139739245176315, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.02156151831150055, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.01635170914232731, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.01628727838397026, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.015325717628002167, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011654836125671864, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.01635170914232731, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011654836125671864, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.12.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.15936221182346344, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.149363175034523, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.1454567313194275, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.13062454760074615, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.07502136379480362, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07098203897476196, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.08550472557544708, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.07856776565313339, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.07623152434825897, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.0672721415758133, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.06320061534643173, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04371200501918793, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.03774687647819519, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.03615209087729454, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.035784829407930374, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.021923968568444252, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.01898045465350151, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.01877826638519764, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01745687797665596, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.017136704176664352, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.011737026274204254, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.012115538120269775, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.01114162802696228, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008642829954624176, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01745687797665596, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008642829954624176, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.12.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.22989997267723083, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.21580621600151062, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.21068055927753448, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1891532838344574, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10825503617525101, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10282247513532639, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12253306806087494, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11248139292001724, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10991474241018295, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09702008217573166, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09094163775444031, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06245777755975723, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05382391810417175, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.051913198083639145, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.051456667482852936, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.031231718137860298, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.026667198166251183, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.026408672332763672, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02438535913825035, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02395380102097988, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016307933256030083, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.016135958954691887, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.01553959958255291, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01057128794491291, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016307933256030083, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01057128794491291, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.12.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.23873639106750488, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.21382397413253784, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.20344123244285583, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.18006254732608795, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.1104382798075676, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.0999787300825119, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.13201647996902466, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.12006891518831253, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.11385148018598557, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09480553865432739, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08955436199903488, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06761997938156128, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.057871878147125244, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.053457990288734436, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.05238351970911026, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03399099409580231, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.02841014601290226, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.02800830826163292, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.025401854887604713, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.024707768112421036, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.018454886972904205, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.01889825612306595, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016900494694709778, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013703471049666405, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016900494694709778, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013703471049666405, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.13.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.08873449265956879, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.08214510232210159, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.07791417092084885, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.06898818910121918, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.04142104089260101, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.037986818701028824, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.05116357281804085, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.04642527550458908, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.04221522435545921, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.036564901471138, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03443138301372528, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.02599632367491722, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.022198744118213654, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.019990531727671623, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.019433358684182167, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013035216368734837, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.01042403094470501, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.010103161446750164, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.00946087297052145, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.009060208685696125, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.006853952072560787, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.00678840558975935, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006028696428984404, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004585066344588995, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013035216368734837, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004585066344588995, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.13.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.07096856087446213, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06525837630033493, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.06095761060714722, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.05358108878135681, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.03251301124691963, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.029295148327946663, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.041066136211156845, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.037781741470098495, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03319159150123596, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.028503218665719032, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.02686268463730812, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.020702874287962914, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.017835019156336784, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.015594948083162308, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.015035499818623066, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.010349022224545479, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.008041824214160442, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007677865214645863, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.007245425134897232, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006817952264100313, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005377056542783976, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005214720498770475, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004538815002888441, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0032548464369028807, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.017835019156336784, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004538815002888441, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.13.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.20759595930576324, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.19280259311199188, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.18714922666549683, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.16669680178165436, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.09720079600811005, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.09135555475950241, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.11415714025497437, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.1021384447813034, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.09874693304300308, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08576970547437668, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.0810391828417778, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.05852780118584633, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.04869571328163147, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04645770043134689, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.04587692767381668, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.02930641919374466, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.023480169475078583, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.023106906563043594, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.02113388292491436, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.020597092807292938, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.014920464716851711, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.013759315013885498, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.013531352393329144, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008092960342764854, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.014920464716851711, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008092960342764854, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.13.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.1949779987335205, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.17623424530029297, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.16878727078437805, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.14562317728996277, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.08999787271022797, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.0823369175195694, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.10656236112117767, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.09657958894968033, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.0920817032456398, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.07613322138786316, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.07020971179008484, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.054764021188020706, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04644929617643356, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.04346972703933716, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.042724136263132095, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.02746928296983242, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.02275409735739231, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.022283099591732025, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.01998453587293625, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01937197893857956, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.014704450033605099, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.014646901749074459, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.013556438498198986, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.010244362987577915, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.014704450033605099, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.010244362987577915, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.13.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.15908195078372955, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.14924508333206177, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.14523908495903015, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.13062894344329834, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.07486269623041153, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07095726579427719, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.08541657775640488, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.07853321731090546, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.07608408480882645, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.06728833168745041, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.063165083527565, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.043726641684770584, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.037767671048641205, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.036129310727119446, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.035736795514822006, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.02193974144756794, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.018984537571668625, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.0187669824808836, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.017499329522252083, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.01716097630560398, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.011758673936128616, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01215178333222866, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.011135746724903584, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008689402602612972, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.017499329522252083, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008689402602612972, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.13.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.22526849806308746, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.2116212099790573, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.20663832128047943, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1857680231332779, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10603316128253937, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10080520808696747, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12010925263166428, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11023743450641632, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10766144841909409, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09529397636651993, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.08948853611946106, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06125388294458389, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05278109386563301, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05085691437125206, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.050399862229824066, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.030616125091910362, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.026130255311727524, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.02586572989821434, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.023943306878209114, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.023513633757829666, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01602274551987648, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015829388052225113, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015236863866448402, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010371540673077106, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01602274551987648, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010371540673077106, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.13.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.23451736569404602, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.20942653715610504, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.1984981745481491, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.17570076882839203, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10836287587881088, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09750501066446304, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.13054367899894714, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11900956183671951, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.11203811317682266, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09282100945711136, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08807723224163055, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06716274470090866, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05753640457987785, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05249956622719765, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.05128861963748932, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03384467959403992, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.028036391362547874, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.027573134750127792, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.02505933679640293, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.024287419393658638, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.01829284429550171, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.018897341564297676, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.01646127551794052, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.01380949281156063, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.01646127551794052, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.01380949281156063, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.14.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.07808871567249298, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.07186546921730042, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.06805726140737534, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.06033574789762497, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.036230430006980896, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.03308665007352829, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.045214440673589706, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.040728434920310974, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.037037283182144165, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03196876496076584, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.030394118279218674, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.022959286347031593, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.019518522545695305, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.017502710223197937, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.017008859664201736, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.011550959199666977, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.009188949130475521, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.00890356581658125, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.008349637500941753, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.007992668077349663, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.006144748535007238, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.00607781857252121, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.0053963493555784225, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0041947923600673676, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.017502710223197937, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0041947923600673676, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.14.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.06923692673444748, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06343915313482285, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.058972492814064026, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.05199217051267624, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.031627167016267776, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.028285101056098938, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.0403892882168293, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.03683950752019882, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.0322832353413105, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.027750108391046524, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.02615073323249817, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.020322958007454872, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.017438892275094986, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.015185032039880753, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.014578699134290218, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.010155259631574154, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.007813775911927223, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007476800587028265, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.007047149818390608, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006623591296374798, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005257642827928066, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005082824733108282, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.0044026365503668785, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.00318423705175519, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.017438892275094986, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.0044026365503668785, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.14.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.1994561105966568, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.18386210501194, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.17722900211811066, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.15766514837741852, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.092807337641716, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.08588072657585144, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.1127234548330307, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.09973672032356262, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.09505489468574524, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08163368701934814, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.0773436427116394, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.057656191289424896, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.04777521267533302, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04436220973730087, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.043543700128793716, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.028879282996058464, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.02250979281961918, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.022038035094738007, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.02025999128818512, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.019582629203796387, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.014717409387230873, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.013557671569287777, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.0128623116761446, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008181730285286903, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.014717409387230873, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008181730285286903, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.14.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.19077496230602264, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.1695936620235443, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.158842071890831, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.13595236837863922, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.08629409968852997, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.07729801535606384, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.11040481179952621, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.09930679947137833, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.09066670387983322, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.07434006780385971, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.06908246874809265, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.056868452578783035, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04819968715310097, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.04227418452501297, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.040796928107738495, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.028786631301045418, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.023028522729873657, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.022337118163704872, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.020762931555509567, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.0197394210845232, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0156694445759058, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.01632041484117508, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.013698818162083626, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.01230511162430048, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0156694445759058, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.01230511162430048, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.14.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.1617497354745865, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.15194809436798096, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.14794577658176422, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.13321655988693237, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.0760767012834549, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07215900719165802, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.08686186373233795, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.07972797751426697, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.07726886868476868, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.06848018616437912, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.06444579362869263, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04434110224246979, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.03833746537566185, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.03667381405830383, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.036278028041124344, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.022245321422815323, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.01922224648296833, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.01899968646466732, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01773110032081604, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.017402876168489456, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01185335498303175, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.012254996225237846, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.011214804835617542, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008714534342288971, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01773110032081604, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008714534342288971, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.14.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.2232183963060379, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.20987017452716827, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.20496222376823425, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.18469619750976562, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10500164330005646, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.09988678246736526, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.11894591897726059, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.1091846451163292, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10655859112739563, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.0945124626159668, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.08880453556776047, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06056715175509453, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05225462093949318, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05030988156795502, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.0498393289744854, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.030288441106677055, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.025800397619605064, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.025523139163851738, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.023653317242860794, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.023236382752656937, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015763571485877037, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015537922270596027, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.014944962225854397, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.0100520309060812, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015763571485877037, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.0100520309060812, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.14.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.2183724045753479, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.19307291507720947, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.18192529678344727, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.1607758104801178, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10040226578712463, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.08962827175855637, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.12369217723608017, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11154372245073318, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.10445813834667206, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.08551298081874847, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08131443709135056, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06329377740621567, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05385927855968475, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.0487079992890358, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.04740404337644577, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.031880732625722885, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.02595234476029873, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.02545664645731449, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.023082654923200607, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.022243835031986237, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.0171784870326519, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.017530294135212898, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.015249300748109818, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012648312374949455, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.0171784870326519, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012648312374949455, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.15.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.08902566134929657, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.08193453401327133, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.07701075822114944, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.06831473112106323, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.041234612464904785, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.037285175174474716, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.05269539728760719, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.04729874059557915, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.0422481968998909, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03650933876633644, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.035176925361156464, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.026988646015524864, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.02272872067987919, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.019933011382818222, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.019225353375077248, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013563274405896664, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.010499782860279083, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.01010597962886095, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.009590070694684982, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.009096624329686165, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.007158900145441294, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.007071777246892452, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006087352521717548, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004876851569861174, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013563274405896664, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004876851569861174, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.15.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.0704452395439148, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06434216350317001, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.0591854490339756, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.05207274481654167, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.031909096986055374, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.028163257986307144, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04213213175535202, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.0385182723402977, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03278670459985733, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.028168242424726486, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.026889657601714134, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.021187573671340942, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.018295837566256523, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.015370333567261696, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.014593087136745453, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.010636165738105774, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.007986974902451038, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007529096677899361, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.0072286417707800865, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006688354071229696, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005532470997422934, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005367236211895943, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004474195651710033, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0033779959194362164, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.015370333567261696, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0033779959194362164, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.15.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.2167956680059433, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.19911512732505798, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.18874585628509521, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.16894616186618805, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.1007959321141243, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.09213639795780182, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.13039566576480865, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.1127495989203453, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.10380564630031586, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.08925939351320267, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.08639343082904816, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.0682171881198883, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.054247938096523285, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.04843981936573982, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.0470227487385273, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.034379370510578156, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.024802913889288902, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.023952223360538483, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.022448627278208733, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.021328030154109, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.017576642334461212, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.015661882236599922, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.014355078339576721, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.009492740966379642, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.017576642334461212, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.009492740966379642, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.15.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.18938668072223663, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.1603924185037613, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.15202823281288147, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.13226990401744843, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.08670149743556976, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.07530499249696732, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.10225309431552887, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.09187224507331848, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.08849035948514938, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.07039175182580948, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.06498156487941742, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.05269072577357292, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04419514909386635, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.041909947991371155, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.04137562960386276, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.026463929563760757, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.021991277113556862, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.021534856408834457, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.01876264438033104, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.01826128177344799, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.014263221062719822, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.014079645276069641, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.013354032300412655, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.00997629202902317, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.014263221062719822, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.00997629202902317, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.15.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.17230252921581268, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.1617264747619629, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.15730229020118713, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.14171428978443146, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.08071902394294739, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07656896859407425, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.09231603145599365, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.08500846475362778, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.08204054832458496, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.07278331369161606, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.0684659332036972, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.047022998332977295, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.040731217712163925, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.038774993270635605, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.03829508647322655, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.023507168516516685, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.020108437165617943, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.019851461052894592, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01851123943924904, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.018127046525478363, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01233611349016428, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.012519112788140774, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.011592352762818336, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.0084928497672081, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.01233611349016428, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.0084928497672081, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.15.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.22579824924468994, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.2124321460723877, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.20732733607292175, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.18695750832557678, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10583984851837158, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10071242600679398, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12022414803504944, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.1105738952755928, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10752926021814346, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09556438773870468, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.08986730873584747, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.061204925179481506, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.052867330610752106, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05071699991822243, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.050202567130327225, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.030591091141104698, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02599276974797249, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.02569725178182125, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.023884108290076256, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.023418553173542023, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015927564352750778, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015682850033044815, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.01506655290722847, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010087556205689907, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015927564352750778, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010087556205689907, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.15.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.23653830587863922, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.2057902216911316, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.19079789519309998, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.1683369278907776, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10837417095899582, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09440533071756363, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.13665834069252014, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.12374180555343628, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.11323035508394241, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09119227528572083, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08690880984067917, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.07021312415599823, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.059960249811410904, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05296077951788902, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.051202766597270966, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03555088862776756, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.02881009690463543, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.02813579887151718, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.025579998269677162, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.02448316477239132, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.019616995006799698, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.0203251875936985, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.017219573259353638, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.015269416384398937, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.017219573259353638, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.015269416384398937, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.16.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.09051358699798584, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.08335965126752853, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.07824231684207916, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.0699508786201477, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.041924525052309036, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.038015205413103104, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.05428306385874748, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.04832920804619789, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.04297205060720444, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03732442855834961, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03603942692279816, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.02778131142258644, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.02316093072295189, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.020297663286328316, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.019566470757126808, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013955621048808098, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.01070686150342226, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.010301870293915272, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.009815405122935772, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.009306736290454865, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.007425868883728981, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.007250932510942221, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.00627764081582427, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.005029548890888691, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013955621048808098, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.005029548890888691, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.16.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.06978973001241684, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06439562886953354, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.059366680681705475, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.05263258144259453, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.03201623633503914, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.028360644355416298, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.0414627343416214, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.03821950778365135, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03276264667510986, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.02834577113389969, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.027078140527009964, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.021056700497865677, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.018297918140888214, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.015404138714075089, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.0145853441208601, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.010525099001824856, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.007898928597569466, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007433217018842697, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.007178090512752533, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006615375634282827, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005440779961645603, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005246246233582497, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004428056068718433, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0031467454973608255, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.015404138714075089, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004428056068718433, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } }, { "key": "model.layers.16.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.23360057175159454, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.21668371558189392, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.2098279893398285, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.18730750679969788, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.10918975621461868, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.10238325595855713, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.1305924654006958, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.11621858179569244, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.1113923043012619, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.09727385640144348, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.09182602167129517, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.06742867082357407, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.055672787129879, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.052362535148859024, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.05158174782991409, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.03371589630842209, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.02650890126824379, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.026032060384750366, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.023936191573739052, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.023236528038978577, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.017222873866558075, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01570899412035942, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.015242948196828365, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0093062249943614, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.017222873866558075, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0093062249943614, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.16.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.20410895347595215, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.18082600831985474, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.17260673642158508, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.14858324825763702, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.09318111836910248, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.08414885401725769, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.11072409152984619, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.10027173161506653, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.09635580331087112, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.0784849300980568, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.07209569960832596, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.05671384558081627, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.04805774241685867, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.044865451753139496, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.04408903792500496, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.028394276276230812, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.023458633571863174, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.022999580949544907, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.02056138962507248, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.019874388352036476, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.015108149498701096, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.015049047768115997, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.013851352967321873, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.010427546687424183, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.015108149498701096, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.010427546687424183, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.16.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.16607528924942017, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.155772864818573, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.15147949755191803, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.13640815019607544, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.07785177230834961, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07373341172933578, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.08885916322469711, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.08190746605396271, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.07916849106550217, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.07010906934738159, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.06584630161523819, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04535979777574539, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.039293840527534485, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.037429358810186386, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.03697968274354935, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.022711627185344696, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.019451631233096123, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.01919981837272644, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01789858378469944, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.01753031648695469, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012034503743052483, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01218862272799015, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.011334775947034359, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008364005014300346, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.01753031648695469, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.008364005014300346, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.16.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.2222261279821396, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.20904017984867096, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.20399361848831177, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1838533580303192, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10435359179973602, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.0991833508014679, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.1184067502617836, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.10881875455379486, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10591746866703033, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09403710067272186, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.08844292163848877, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06025952845811844, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.0520426444709301, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05001853406429291, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04953015595674515, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.030155880376696587, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02573329582810402, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.025447610765695572, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.023627419024705887, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.023192627355456352, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015868769958615303, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015639180317521095, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015081532299518585, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01027147751301527, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015868769958615303, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.01027147751301527, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.16.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.2284240573644638, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.19889238476753235, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.18580308556556702, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.16330298781394958, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10445106029510498, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09163589030504227, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.12931664288043976, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11721732467412949, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.10915658622980118, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.08754108101129532, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08331170678138733, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.066319540143013, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.056634992361068726, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05079161748290062, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.04933009296655655, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03349778801202774, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.027323495596647263, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.026818882673978806, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.024117030203342438, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.023207761347293854, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.01829661801457405, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.018828537315130234, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016245592385530472, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013927271589636803, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016245592385530472, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013927271589636803, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.17.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.08609452843666077, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.07930634915828705, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.07426901906728745, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.06639142334461212, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.039905156940221786, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.03608275577425957, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.05094384402036667, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.04614987596869469, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.040811341255903244, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03545384854078293, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03403119742870331, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.025852443650364876, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.022067898884415627, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.019274571910500526, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.018581796437501907, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.012998268008232117, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.010122111067175865, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.009720254689455032, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.009244143962860107, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.008752990514039993, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.006900970824062824, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.0067919110879302025, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.005904010031372309, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004598971921950579, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.012998268008232117, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004598971921950579, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.17.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.07331952452659607, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06739787757396698, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.061043135821819305, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.05427735671401024, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.03356958553195, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.029173118993639946, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04532215744256973, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.04180871695280075, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03443407639861107, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.02987603098154068, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.02864890545606613, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.022827336564660072, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.019918017089366913, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.016171161085367203, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.015144547447562218, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.011419941671192646, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.008374289609491825, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007820693776011467, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.00763739226385951, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006981030106544495, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.005933663807809353, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.0057555376552045345, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004704755265265703, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.003587496466934681, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.016171161085367203, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.003587496466934681, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.17.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.2505655288696289, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.22637389600276947, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.2168620079755783, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.1924564391374588, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.11520225554704666, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.105710469186306, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.14171425998210907, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.1250450164079666, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.11902107298374176, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.10043040663003922, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.09568897634744644, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.07422015070915222, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.05992814525961876, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.055328767746686935, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.05417207255959511, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.037863582372665405, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.02825218439102173, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.027650294825434685, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.025170933455228806, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.024307187646627426, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.019492406398057938, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.017321689054369926, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.01634405367076397, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.010624280199408531, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.017321689054369926, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.010624280199408531, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.17.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.25437015295028687, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.2342698872089386, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.22714121639728546, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.19676826894283295, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.11808344721794128, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.11052130162715912, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.13652287423610687, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.12358418107032776, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.1202869638800621, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.10233244299888611, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.09284516423940659, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.06981439888477325, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.05934765934944153, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.05691729113459587, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.05632075294852257, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.03503841161727905, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.02972814440727234, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.029365573078393936, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.026589922606945038, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.026035191491246223, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.018661610782146454, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.018786704167723656, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.017708169296383858, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.01318918913602829, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.017708169296383858, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.01318918913602829, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.17.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.18881285190582275, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.17713825404644012, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.1725974977016449, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.15529461205005646, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.08866565674543381, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.08416878432035446, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.1006070226430893, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.09278403967618942, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.09020709246397018, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.07979033142328262, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.07490666955709457, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.05136535316705704, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.044468171894550323, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.042628537863492966, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.042197853326797485, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.025675665587186813, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.022124839946627617, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.02188458852469921, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.020338671281933784, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.01995902694761753, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.013593938201665878, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.013766787946224213, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.01291207130998373, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009427569806575775, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.013593938201665878, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009427569806575775, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.17.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.23479865491390228, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.22061367332935333, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.21536947786808014, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1936798393726349, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.11024600267410278, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10491153597831726, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12449268251657486, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11470408737659454, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.11196303367614746, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09924131631851196, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09309391677379608, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06340227276086807, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05486740916967392, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.052836280316114426, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05234574154019356, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03169301152229309, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02709317021071911, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.02680901810526848, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.024835191667079926, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02439313381910324, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016612624749541283, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.016326315701007843, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015823356807231903, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010582499206066132, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.016612624749541283, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.010582499206066132, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.17.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.2393425852060318, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.21189624071121216, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.19944116473197937, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.17566430568695068, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.11017400771379471, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.09810642898082733, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.13479091227054596, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.12241110950708389, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.1141652762889862, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.09336990118026733, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08833776414394379, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.0687876045703888, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05877509340643883, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.05338754877448082, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.05201577767729759, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03464944288134575, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.028298893943428993, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.027775589376688004, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.02503848448395729, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.02418353594839573, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.018802236765623093, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.018908917903900146, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016844239085912704, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013535435311496258, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.016844239085912704, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013535435311496258, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.18.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.0965581014752388, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.08867204934358597, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.08282095938920975, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.07394351810216904, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.04481750354170799, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.04022907093167305, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.05759906396269798, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.05202246084809303, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.04590421915054321, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03970695659518242, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03813212364912033, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.02942134439945221, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.02501502074301243, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.02165505848824978, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.020824726670980453, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.014774623326957226, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.011425292119383812, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.010963043197989464, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.010432618670165539, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.00985571276396513, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.007824793457984924, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.007759558502584696, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006624432746320963, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.005335692316293716, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.014774623326957226, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.005335692316293716, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.18.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.07397061586380005, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06770434230566025, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.06009544059634209, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.053564902395009995, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.033784493803977966, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.02861819788813591, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04782922565937042, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.043865855783224106, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03466726094484329, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.0300358347594738, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.028923295438289642, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.024107584729790688, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.020791955292224884, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.01626831665635109, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.015060309320688248, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.01196870394051075, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.008448806591331959, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.0078025637194514275, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.007694236468523741, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006934538949280977, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.006184347439557314, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005944343283772469, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004752002656459808, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.003690265817567706, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.01626831665635109, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.003690265817567706, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.18.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.24798622727394104, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.2217305451631546, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.20462335646152496, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.18147149682044983, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.11399855464696884, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.09994564950466156, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.15382152795791626, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.1342855840921402, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.11861671507358551, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.0988697037100792, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.0968712642788887, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.08143101632595062, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.06470746546983719, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.05513746663928032, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.052604977041482925, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.04235402122139931, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.028422128409147263, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.02700994350016117, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.02533973939716816, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.02355276420712471, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.02193322591483593, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01882334053516388, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.016280660405755043, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.011732153594493866, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.016280660405755043, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.011732153594493866, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.18.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.23454731702804565, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.20686131715774536, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.19656537473201752, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.16660547256469727, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.1071988120675087, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.09669793397188187, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.12893490493297577, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.11627607047557831, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.11096019297838211, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.08866742998361588, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.0824364647269249, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.06609871983528137, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.05582617595791817, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.05160663276910782, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.05056853964924812, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.033154167234897614, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.026958372443914413, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.026411769911646843, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.023290684446692467, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.022428715601563454, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.01776222325861454, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.017362160608172417, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.016132544726133347, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011964328587055206, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.01776222325861454, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011964328587055206, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.18.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.18873390555381775, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.17696711421012878, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.1726064831018448, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.15503807365894318, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.08870134502649307, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.08419272303581238, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.10007806122303009, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.09257936477661133, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.09015509486198425, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.07963290810585022, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.07459384202957153, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.05098758265376091, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.04435281455516815, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.0426141656935215, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04221248999238014, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.02548859268426895, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02208283171057701, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.021848119795322418, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02025770954787731, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.01989130862057209, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.013408171944320202, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.013682072050869465, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.012759901583194733, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009316680952906609, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.013408171944320202, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009316680952906609, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.18.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.22890900075435638, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.21494317054748535, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.21001088619232178, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1886664628982544, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.1076110303401947, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10233230888843536, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12076281756162643, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.1116754338145256, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.1092485636472702, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09665657579898834, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09044251590967178, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06142386794090271, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05335614085197449, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.05152781307697296, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05107617378234863, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.030668115243315697, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.02631688304245472, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.026058951392769814, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02406277507543564, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02363726682960987, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015897363424301147, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015691449865698814, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015188782475888729, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009963384829461575, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015897363424301147, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009963384829461575, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.18.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.21183991432189941, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.18725889921188354, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.1729976385831833, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.15228937566280365, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.09763455390930176, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.08495322614908218, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.12401765584945679, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.11328740417957306, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.10123562067747116, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.08295779675245285, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.07921212166547775, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06376495957374573, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05484269931912422, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.04750272259116173, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.04562348127365112, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.03241215646266937, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.025512689724564552, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.024720730260014534, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.02273913286626339, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.021571775898337364, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.017761899158358574, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.01785150170326233, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.015246357768774033, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012942363508045673, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.017761899158358574, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.012942363508045673, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.19.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.09254539757966995, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.08537981659173965, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.07926955074071884, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.07096382975578308, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.04294288530945778, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.03845599666237831, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.056038983166217804, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.05083755776286125, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.043942973017692566, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.03827793896198273, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03682571277022362, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.028421675786376, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.024322696030139923, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.020860731601715088, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.019956205040216446, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.014329789206385612, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.011064007878303528, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.010552646592259407, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.01015193946659565, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.009530046954751015, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.007708795368671417, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.007644368335604668, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006536745931953192, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.005284867249429226, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.014329789206385612, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.005284867249429226, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.19.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.07234105467796326, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.0665181428194046, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.059166986495256424, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.0527472198009491, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.03321991115808487, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.02826128527522087, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04647521302103996, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.04284759610891342, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03424566239118576, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.02960384264588356, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.028490863740444183, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.023379424586892128, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.020536351948976517, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.016057409346103668, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.014824863523244858, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.01181721966713667, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.008432689122855663, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.00778167461976409, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.007712685503065586, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.006933773402124643, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.006254907231777906, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.006023885682225227, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004896244965493679, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0038855017628520727, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.016057409346103668, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0038855017628520727, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.19.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.2535112500190735, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.22693127393722534, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.21317334473133087, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.18936575949192047, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.11601315438747406, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.10386576503515244, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.1529800295829773, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.1319086104631424, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.12070005387067795, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.10035423189401627, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.09814103692770004, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.08077064156532288, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.06300455331802368, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.055796705186367035, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.053939543664455414, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.04105648398399353, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.028705744072794914, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.027701590210199356, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.025564802810549736, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.024220524355769157, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.021188851445913315, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.018408382311463356, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.0164247527718544, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.011441363953053951, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.0164247527718544, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.011441363953053951, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.19.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.23377248644828796, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.21221093833446503, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.20428676903247833, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.17040479183197021, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.10799207538366318, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.0997127816081047, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.12617740035057068, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.11449336260557175, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.11044637113809586, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.09066895395517349, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.08097320050001144, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.06486279517412186, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.054942063987255096, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.051915064454078674, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.051207516342401505, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.032635949552059174, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.0270429328083992, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.0266044232994318, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.023594053462147713, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.022903982549905777, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.017405258491635323, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.01706724613904953, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.016210023313760757, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011721539311110973, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.017405258491635323, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.011721539311110973, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.19.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.20084422826766968, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.18839626014232635, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.18384265899658203, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.16518685221672058, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.09446901082992554, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.08972548693418503, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.10644800215959549, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.09832964837551117, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.09591173380613327, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.08474458754062653, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.07941524684429169, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.054240789264440536, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.04712928459048271, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.045340247452259064, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.0449339784681797, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.0271297600120306, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.023446686565876007, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.023205190896987915, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.02149149775505066, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.021096963435411453, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.014285854995250702, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.014422878623008728, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.01363296341150999, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009719878435134888, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.014285854995250702, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009719878435134888, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.19.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.22861148416996002, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.21490947902202606, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.2099001258611679, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1887497901916504, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10751687735319138, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.10230819135904312, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.12043189257383347, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.11144820600748062, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10911738872528076, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09652405232191086, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.09041514247655869, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.06118154153227806, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.05322124809026718, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.051439862698316574, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.05102851986885071, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.03055436722934246, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.026197882369160652, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.025950787588953972, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.023932043462991714, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.02351747825741768, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015762904658913612, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015480058267712593, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.015079965814948082, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009665642865002155, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015762904658913612, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009665642865002155, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.19.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.21961461007595062, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.1937168836593628, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.1799711138010025, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.15830349922180176, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.10085498541593552, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.08834909647703171, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.1268080174922943, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.1158079132437706, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.10481962561607361, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.08575725555419922, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.08152052760124207, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.06495814770460129, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.05602612718939781, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.049081139266490936, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.04730769246816635, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.032733816653490067, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.026314182206988335, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.025586770847439766, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.02343052439391613, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.02233489416539669, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.01765349507331848, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.018247194588184357, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.01525655947625637, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013240252621471882, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.01765349507331848, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.013240252621471882, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.20.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.09055797755718231, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.0835648849606514, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.07818116992712021, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.07000069320201874, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.04229063540697098, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.03798605501651764, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.05391007661819458, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.04879644513130188, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.04318755865097046, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.0373896062374115, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.03566858917474747, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.027402177453041077, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.023476507514715195, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.020446164533495903, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.01966278627514839, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013791520148515701, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.010719215497374535, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.01029153447598219, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.00975100975483656, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.009222441352903843, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0073420400731265545, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.007192474789917469, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006322287023067474, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004870173055678606, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.013791520148515701, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.004870173055678606, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.20.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.07434702664613724, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.06838523596525192, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.06185737997293472, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.05526057630777359, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.034120943397283554, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.02958369255065918, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04632598161697388, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.04278397187590599, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03504989668726921, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.03049582429230213, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.029205847531557083, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.023437129333615303, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.020464863628149033, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.016466697677969933, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.01541350781917572, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.011665173806250095, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.008566438220441341, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.007961856201291084, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.00783600378781557, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.007102637086063623, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.006120550446212292, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.005940485280007124, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.004817802459001541, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0036785027477890253, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.016466697677969933, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0036785027477890253, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.20.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.2597622871398926, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.23850829899311066, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.23117554187774658, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.2056090235710144, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.12041549384593964, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.11251021176576614, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.1416347175836563, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.12756295502185822, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.12332884967327118, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.10607590526342392, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.10001280158758163, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.07269343733787537, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.061057016253471375, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.057842522859573364, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.057046424597501755, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.03650328516960144, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.029224704951047897, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.028798576444387436, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.026185661554336548, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.02547885663807392, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.018623102456331253, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01725327968597412, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.016704028472304344, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.010281406342983246, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.01725327968597412, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.010281406342983246, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.20.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.21343764662742615, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.19408854842185974, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.1875257045030594, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.1600670963525772, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.09871424734592438, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.09149204939603806, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.11358579993247986, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.10384078323841095, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.10086540132761002, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.08458180725574493, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.07657651603221893, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.05828336626291275, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.049844056367874146, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.04748888686299324, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.046940941363573074, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.029263729229569435, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.02468193881213665, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.02433440089225769, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.021823154762387276, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.021282903850078583, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0155717171728611, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.01541689969599247, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.014681187458336353, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.010537690483033657, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.0155717171728611, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.010537690483033657, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.20.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.20206475257873535, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.18955789506435394, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.18496128916740417, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.1663462072610855, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.09500789642333984, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.09013687819242477, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.10703817754983902, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.0989479050040245, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.09646449238061905, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.08523014932870865, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.0799204558134079, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.054634179919958115, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.0474463514983654, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.04560977220535278, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04515134543180466, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.02733631059527397, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.023613613098859787, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.0233574490994215, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.021653592586517334, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.021265966817736626, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.014419515617191792, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.014588966965675354, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.013745608739554882, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009889463894069195, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.014419515617191792, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009889463894069195, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.20.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.21856465935707092, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.20539942383766174, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.20078836381435394, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.18054021894931793, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.10274854302406311, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.09775645285844803, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.11510305106639862, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.10653620213270187, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.10428325831890106, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.09231974184513092, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.08647698909044266, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.05853936821222305, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.050886210054159164, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.049180809408426285, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04876798763871193, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.029220853000879288, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.025074683129787445, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.024830473586916924, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.022917812690138817, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.022530842572450638, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015105370432138443, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.014870370738208294, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.014451604336500168, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009355560876429081, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.015105370432138443, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.009355560876429081, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.20.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.20231890678405762, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.17919528484344482, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.16878463327884674, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.1485150307416916, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.09308306127786636, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.08296985924243927, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.11395411193370819, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.10363680124282837, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.0964999720454216, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.07915902137756348, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.07477372884750366, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.05815155431628227, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.04994678124785423, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.04513169825077057, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.043934401124715805, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.029301952570676804, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.0239910576492548, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.02352767251431942, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.021302588284015656, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.020538482815027237, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.015830732882022858, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.016142671927809715, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.014132672920823097, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.01159900426864624, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.015830732882022858, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.01159900426864624, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.21.self_attn.q_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.09773210436105728, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.08981361240148544, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.08352431654930115, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.07481899857521057, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.04546906426548958, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.04052582383155823, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.05904197692871094, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.05305253341794014, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.046457767486572266, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.04014260321855545, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.038336507976055145, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.030017780140042305, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.025428619235754013, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.021983515471220016, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.021094920113682747, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.015121301636099815, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.011541378684341908, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.01104668341577053, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.010491725988686085, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.0098892105743289, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.008073962293565273, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.007794014178216457, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.006894586607813835, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0052864099852740765, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.015121301636099815, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.0052864099852740765, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.21.self_attn.k_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.07934994995594025, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.07313751429319382, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.06712152808904648, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.05990375950932503, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.036528099328279495, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.032318778336048126, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.04815559461712837, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.044180918484926224, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.03746139258146286, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.032550256699323654, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.03109017014503479, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.024310331791639328, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.021151047199964523, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.017649147659540176, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.01676904782652855, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.012187093496322632, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.009183169342577457, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.00864462275058031, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.008382524363696575, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.00772849190980196, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.006437118165194988, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.006232742220163345, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.005263691768050194, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0039332290180027485, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.017649147659540176, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.0039332290180027485, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.21.self_attn.v_proj", "numel": 524288, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.193359375, "total_bits": 1149952.0, "err": 0.2297985851764679, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.380859375, "total_bits": 1248256.0, "err": 0.21172471344470978, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.630859375, "total_bits": 1379328.0, "err": 0.20556676387786865, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.755859375, "total_bits": 1444864.0, "err": 0.1824997067451477, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.240234375, "total_bits": 1698816.0, "err": 0.10660169273614883, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.740234375, "total_bits": 1960960.0, "err": 0.099824458360672, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03271484375, "total_bits": 1590016.0, "err": 0.12344856560230255, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.130859375, "total_bits": 1641472.0, "err": 0.11147505789995193, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.193359375, "total_bits": 1674240.0, "err": 0.1090221032500267, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.537109375, "total_bits": 1854464.0, "err": 0.09375341981649399, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6904296875, "total_bits": 1934848.0, "err": 0.08782900124788284, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03271484375, "total_bits": 2114304.0, "err": 0.0630631372332573, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.130859375, "total_bits": 2165760.0, "err": 0.05335693433880806, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.240234375, "total_bits": 2223104.0, "err": 0.051092684268951416, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.349609375, "total_bits": 2280448.0, "err": 0.05051395669579506, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03271484375, "total_bits": 2638592.0, "err": 0.03147696703672409, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.240234375, "total_bits": 2747392.0, "err": 0.02579505741596222, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.380859375, "total_bits": 2821120.0, "err": 0.025438755750656128, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.537109375, "total_bits": 2903040.0, "err": 0.023016976192593575, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.771484375, "total_bits": 3025920.0, "err": 0.022536972537636757, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.01597975380718708, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.130859375, "total_bits": 3214336.0, "err": 0.014939794316887856, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28271484375, "total_bits": 3293952.0, "err": 0.014715605415403843, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008628816343843937, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.03271484375, "total_bits": 3162880.0, "err": 0.01597975380718708, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.130859375, "total_bits": 4262912.0, "err": 0.008628816343843937, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.21.self_attn.o_proj", "numel": 4194304, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.188232421875, "total_bits": 9178112.0, "err": 0.14543882012367249, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 9964544.0, "err": 0.12311279028654099, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 11013120.0, "err": 0.11397860944271088, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750732421875, "total_bits": 11537408.0, "err": 0.0982792004942894, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.235107421875, "total_bits": 13569024.0, "err": 0.06572482734918594, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.735107421875, "total_bits": 15666176.0, "err": 0.0555277094244957, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 12714752.0, "err": 0.08120936155319214, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 13110272.0, "err": 0.07420604676008224, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.188232421875, "total_bits": 13372416.0, "err": 0.06845703721046448, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531982421875, "total_bits": 14814208.0, "err": 0.05329635366797447, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6878662109375, "total_bits": 15468032.0, "err": 0.050284769386053085, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 16909056.0, "err": 0.041796326637268066, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 17304576.0, "err": 0.03590569272637367, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.235107421875, "total_bits": 17763328.0, "err": 0.03209973871707916, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.344482421875, "total_bits": 18222080.0, "err": 0.031136775389313698, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 21103360.0, "err": 0.021126756444573402, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.017478113994002342, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.375732421875, "total_bits": 22547456.0, "err": 0.01692202128469944, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.531982421875, "total_bits": 23202816.0, "err": 0.01517355814576149, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.766357421875, "total_bits": 24185856.0, "err": 0.014489390887320042, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 25297664.0, "err": 0.01165548525750637, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 25693184.0, "err": 0.012307245284318924, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.28143310546875, "total_bits": 26346240.0, "err": 0.010329877026379108, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.009343513287603855, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.235107421875, "total_bits": 21957632.0, "err": 0.017478113994002342, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 34081792.0, "err": 0.009343513287603855, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.21.mlp.gate_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.17439089715480804, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.16360001266002655, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.1594538390636444, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.14320924878120422, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.08250416070222855, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.07831677794456482, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.09320559352636337, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.08611070364713669, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.08379119634628296, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.0740482434630394, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.06950295716524124, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04820223152637482, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.04201437532901764, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.04044162482023239, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.04004780203104019, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.02438148856163025, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.022165272384881973, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.021959472447633743, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.020584534853696823, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.020279599353671074, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.013761231675744057, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.015309982933104038, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.013190933503210545, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.012171515263617039, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.013761231675744057, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.012171515263617039, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.21.mlp.up_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.187766335227273, "total_bits": 25234432.000000004, "err": 0.14697527885437012, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375266335227273, "total_bits": 27397120.000000004, "err": 0.13764142990112305, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625266335227273, "total_bits": 30280704.000000004, "err": 0.13451288640499115, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.750266335227273, "total_bits": 31722496.000000004, "err": 0.12031248211860657, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.234641335227273, "total_bits": 37309440.0, "err": 0.06967594474554062, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.734641335227273, "total_bits": 43076608.0, "err": 0.0660889744758606, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.0313165838068183, "total_bits": 34964224.0, "err": 0.0788373351097107, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125266335227273, "total_bits": 36047872.0, "err": 0.0725119560956955, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.187766335227273, "total_bits": 36768768.0, "err": 0.07082797586917877, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.531516335227273, "total_bits": 40733696.0, "err": 0.0625605434179306, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.6876331676136362, "total_bits": 42534400.0, "err": 0.05882653966546059, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.031316583806818, "total_bits": 46498560.0, "err": 0.04080734774470329, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.1252663352272725, "total_bits": 47582208.0, "err": 0.03572273999452591, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2346413352272725, "total_bits": 48843776.0, "err": 0.034472573548555374, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.3440163352272725, "total_bits": 50105344.0, "err": 0.03417988494038582, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.031316583806818, "total_bits": 58032896.0, "err": 0.020771032199263573, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2346413352272725, "total_bits": 60378112.0, "err": 0.019385099411010742, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.3752663352272725, "total_bits": 62000128.0, "err": 0.019229706376791, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.5315163352272725, "total_bits": 63802368.0, "err": 0.01811143569648266, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.017867352813482285, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.031316583806818, "total_bits": 69567232.0, "err": 0.012105725705623627, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.1252663352272725, "total_bits": 70650880.0, "err": 0.01391355600208044, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.281316583806818, "total_bits": 72450816.0, "err": 0.011673692613840103, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.011503072455525398, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.7658913352272725, "total_bits": 66505728.0, "err": 0.017867352813482285, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125266335227273, "total_bits": 93719552.0, "err": 0.011503072455525398, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } }, { "key": "model.layers.21.mlp.down_proj", "numel": 11534336, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1768687855113638, "total_bits": 25108736.0, "err": 0.14164666831493378, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.25:3b_32g/0.75:2b_32g s4", "bpw": 2.375732421875, "total_bits": 27402496.0, "err": 0.12597806751728058, "qparams": { "group_size": { "3": 32, "2": 32 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.25:4b_32g/0.75:2b_32g s4", "bpw": 2.625732421875, "total_bits": 30286080.0, "err": 0.11828383058309555, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", "bpw": 2.7336869673295454, "total_bits": 31531264.0, "err": 0.1034436821937561, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 2 ], "bits_prop": [ 0.1, 0.4, 0.5 ], "scale_bits": 4 } }, { "desc": "0.1:4b_32g/0.9:3b_32g s4", "bpw": 3.228005149147727, "total_bits": 37232896.0, "err": 0.06498751789331436, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7393687855113638, "total_bits": 43131136.0, "err": 0.05808405950665474, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, { "desc": "1.0:3b_128g s4", "bpw": 3.03143310546875, "total_bits": 34965568.0, "err": 0.08359568566083908, "qparams": { "group_size": { "3": 128 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:3b_32g s4", "bpw": 3.125732421875, "total_bits": 36053248.0, "err": 0.07361141592264175, "qparams": { "group_size": { "3": 32 }, "bits": [ 3 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.05:4b_32g/0.95:3b_32g s4", "bpw": 3.1768687855113638, "total_bits": 36643072.0, "err": 0.06764072179794312, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "desc": "0.4:4b_32g/0.6:3b_32g s4", "bpw": 3.529141512784091, "total_bits": 40706304.0, "err": 0.055920910090208054, "qparams": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.6:4b_64g/0.4:3b_64g s4", "bpw": 3.665138938210227, "total_bits": 42274944.0, "err": 0.05322260037064552, "qparams": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.6, 0.4 ], "scale_bits": 4 } }, { "desc": "1.0:4b_128g s4", "bpw": 4.03143310546875, "total_bits": 46499904.0, "err": 0.042626332491636276, "qparams": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:4b_32g s4", "bpw": 4.125732421875, "total_bits": 47587584.0, "err": 0.03587917611002922, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.2280051491477275, "total_bits": 48767232.0, "err": 0.03196005895733833, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.330277876420454, "total_bits": 49946879.99999999, "err": 0.03095974586904049, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:5b_128g s4", "bpw": 5.03143310546875, "total_bits": 58034240.0, "err": 0.02172343246638775, "qparams": { "group_size": { "5": 128 }, "bits": [ 5 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.01758378930389881, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", "bpw": 5.330277876420454, "total_bits": 61481215.99999999, "err": 0.017208879813551903, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.05, 0.9 ], "scale_bits": 4 } }, { "desc": "0.4:6b_32g/0.6:5b_32g s4", "bpw": 5.529141512784091, "total_bits": 63774976.0, "err": 0.015886075794696808, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.733686967329546, "total_bits": 66134272.00000001, "err": 0.015286102890968323, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.1, 0.3, 0.6 ], "scale_bits": 4 } }, { "desc": "1.0:6b_128g s4", "bpw": 6.03143310546875, "total_bits": 69568576.0, "err": 0.012238754890859127, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "1.0:6b_32g s4", "bpw": 6.125732421875, "total_bits": 70656256.0, "err": 0.012561766430735588, "qparams": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, { "desc": "0.1:8b_128g/0.9:6b_128g s4", "bpw": 6.2587058327414775, "total_bits": 72190016.0, "err": 0.01082510408014059, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.009679592214524746, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.1:6b_32g/0.9:5b_32g s4", "bpw": 5.2280051491477275, "total_bits": 60301568.0, "err": 0.01758378930389881, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.125732421875, "total_bits": 93724928.0, "err": 0.009679592214524746, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } } ], "last_module_idx": 46, "base_perplexity": 2.411064600164806 }