diff --git "a/measurement.json" "b/measurement.json" new file mode 100644--- /dev/null +++ "b/measurement.json" @@ -0,0 +1,230826 @@ +{ + "measurement": [ + { + "key": "model.layers.0.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.024900425225496292, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.02462487295269966, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.011244457215070724, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.011175799183547497, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.011117233894765377, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0073049976490437984, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.03750455006957054, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.024356521666049957, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.011137228459119797, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.011082216165959835, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.011622770689427853, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.014176304452121258, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.011065805330872536, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.008161813952028751, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0072753895074129105, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.008475743234157562, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.007267041131854057, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.006945888511836529, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.007265355903655291, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.006944175809621811, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00752494391053915, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00726509653031826, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0071802702732384205, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0069430191069841385, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.024900425225496292, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.024900425225496292, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.02365073747932911, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.023318925872445107, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.01105935126543045, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.010972218587994576, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.010892768390476704, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.00553315132856369, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.030020078644156456, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.02305687963962555, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.010932372882962227, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.010837794281542301, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.011061464436352253, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.011772763915359974, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.010814491659402847, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.006795276887714863, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.005475160665810108, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.006628116592764854, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.005452047102153301, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.004955265671014786, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.005448649637401104, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.004949879366904497, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.004936956334859133, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0054468815214931965, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.004266067873686552, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004946091212332249, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.02365073747932911, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.02365073747932911, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.06823264062404633, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.056940000504255295, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.04363870620727539, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.037094514816999435, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.030637212097644806, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.020976170897483826, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05956115573644638, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0453193373978138, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.032601889222860336, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.02537151612341404, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.026713043451309204, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03175647556781769, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02226794883608818, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.015301859937608242, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.013159258291125298, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01581532694399357, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.00871690921485424, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.007716379594057798, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.007807397283613682, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0065299575217068195, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00817715935409069, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007311289198696613, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.004525299649685621, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005326750222593546, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.04363870620727539, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.04363870620727539, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.028085041791200638, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.016999278217554092, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.010736239142715931, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.011756548658013344, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.011245564557611942, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.005482736974954605, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.017517324537038803, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.015414099209010601, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.012900961562991142, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.007942341268062592, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.008345249108970165, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.008843840099871159, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.00767862144857645, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.005942679010331631, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.005471551325172186, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.004681576043367386, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.003967470955103636, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0038317255675792694, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.003553071990609169, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.003314433852210641, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0029612374491989613, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0035257863346487284, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0025126105174422264, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0031293847132474184, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.028085041791200638, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.028085041791200638, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.07062546908855438, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.06495901197195053, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.06311721354722977, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.056414250284433365, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.03197522461414337, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.030079564079642296, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.03617497906088829, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.03336106240749359, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.03257914260029793, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.02835276536643505, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.02674270235002041, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.01847589574754238, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.016159744933247566, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.015542453154921532, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.015399524010717869, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.009346536360681057, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.008556006476283073, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.008507495746016502, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.007909061387181282, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.007818956859409809, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.00530457217246294, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.005945325363427401, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.0051033468917012215, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.004756676033139229, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.03617497906088829, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.03617497906088829, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.10542310774326324, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.09710267931222916, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.0944291427731514, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.08445338904857635, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.047664571553468704, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.04490114003419876, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.0537107028067112, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.04954797029495239, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.04855921491980553, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.04223452880978584, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.03982030600309372, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.02723478153347969, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.02360468916594982, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.022776639088988304, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.022575322538614273, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.013612636364996433, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.011886218562722206, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01182567235082388, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.010831516236066818, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.010710548609495163, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.00725608691573143, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.00747351860627532, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.006966793444007635, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.005257736891508102, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.04223452880978584, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.04223452880978584, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.07068964838981628, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.05770494416356087, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.04105576127767563, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.036157701164484024, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.02826974354684353, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.020824840292334557, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.0517863929271698, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.04649737849831581, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.030565649271011353, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.02297123149037361, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.023951757699251175, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.023684905841946602, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.019320866093039513, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.017178161069750786, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.013739943504333496, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.014187254942953587, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.009783057495951653, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.007878473959863186, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.009051410481333733, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.006799021270126104, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.008462208323180676, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.008608736097812653, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.005628132726997137, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.005661563016474247, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.04105576127767563, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.04105576127767563, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.015660276636481285, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.012246087193489075, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.009342585690319538, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.008200121112167835, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0066651408560574055, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.004439749289304018, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0119024021551013, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.009646180085837841, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.007310854271054268, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.005407495889812708, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.005811365786939859, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.006313812453299761, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.00472997734323144, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0033146364148706198, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0028698157984763384, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0032313531264662743, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.001864365884102881, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.00166955660097301, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0016483472427353263, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0013745559845119715, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0017141528660431504, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0015402943827211857, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0010647540912032127, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0011014891788363457, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.015660276636481285, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.015660276636481285, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.015370645560324192, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.011869578622281551, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.00900510884821415, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.00790916197001934, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.00642035948112607, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.004310176242142916, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.010751301422715187, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.009339066222310066, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.007118169218301773, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.005143764428794384, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.005299827083945274, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.005480526015162468, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.004475991707295179, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0031821420416235924, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.002798268338665366, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.002758651040494442, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0017981575801968575, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0016374527476727962, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0015778271481394768, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0013409591047093272, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0014837357448413968, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.001472442876547575, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.001011531101539731, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.001074042753316462, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.015370645560324192, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.015370645560324192, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08823852986097336, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07329204678535461, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.06657329201698303, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.057004332542419434, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.039422255009412766, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03299984708428383, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04999721795320511, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0456753671169281, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0416451096534729, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03137427568435669, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.02950281836092472, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02556028962135315, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.021951714530587196, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.019101109355688095, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.018364597111940384, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012835883535444736, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010129041969776154, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.009862648323178291, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008593265898525715, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.008104909211397171, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006782230455428362, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00693585816770792, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005781196523457766, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004863433539867401, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0456753671169281, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0456753671169281, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.17400513589382172, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1062103658914566, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.06915820389986038, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07592030614614487, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0742606371641159, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04067498818039894, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10703511536121368, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0954640731215477, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08340960741043091, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04869253933429718, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05175185203552246, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05545547604560852, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04728170111775398, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03784956410527229, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03528311103582382, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0284847691655159, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022745413705706596, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.021990643814206123, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.018822697922587395, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.017276406288146973, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016460301354527473, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0186393354088068, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013527674600481987, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015509377233684063, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04067498818039894, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04067498818039894, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.12883614003658295, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.12087565660476685, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.11854201555252075, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.10717169940471649, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0577382855117321, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.05540307238698006, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.06411426514387131, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.05915556102991104, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.05844938009977341, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.05245312303304672, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.049923624843358994, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03261170908808708, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.028295762836933136, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.027689967304468155, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.027543241158127785, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.01637144573032856, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.014822226949036121, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.014781542122364044, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.013852478004992008, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.013765115290880203, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.009075777605175972, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.009765575639903545, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.00888247974216938, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.007458175998181105, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03261170908808708, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03261170908808708, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.15926095843315125, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.14997026324272156, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1472110152244568, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13365910947322845, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07156527042388916, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.06878350675106049, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.0793209820985794, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07321511209011078, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07242008298635483, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06522633135318756, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06201313063502312, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.039910368621349335, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03462564945220947, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.033908452838659286, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03373965993523598, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.01987510919570923, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.017268365249037743, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01722118631005287, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01600707694888115, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.015902047976851463, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.010311674326658249, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.010168405249714851, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.010069217532873154, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.006468949373811483, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.039910368621349335, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.039910368621349335, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.14467984437942505, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.1284809410572052, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.11937002092599869, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.10652126371860504, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.06608612835407257, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.05725393071770668, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.0815073698759079, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.07549813389778137, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.06826818734407425, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.05768589302897453, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.055564507842063904, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.043306462466716766, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.03861799091100693, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.032667480409145355, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.030742408707737923, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.02297757938504219, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.01809939742088318, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.016954921185970306, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.016627583652734756, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.01522457879036665, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01350295078009367, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.013575638644397259, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01127404160797596, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.009852076880633831, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.043306462466716766, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.043306462466716766, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.03791269659996033, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.03446435555815697, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.03310666233301163, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.029389305040240288, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.017242765054106712, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.015957560390233994, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.020827624946832657, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.018432127311825752, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.017630554735660553, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.01506369560956955, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.014545360580086708, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.010683782398700714, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.008828505873680115, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.008312959223985672, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.008186915889382362, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.005384913180023432, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0043329475447535515, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.004283800255507231, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.003907398320734501, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.003828891785815358, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0028325580060482025, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0027411612682044506, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.002665700390934944, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.001876956899650395, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.03791269659996033, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.03791269659996033, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.03057878278195858, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.027682548388838768, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.026457563042640686, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0234098881483078, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.013805232010781765, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.012673846445977688, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.01647629588842392, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.014949062839150429, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.014129284769296646, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.01199869904667139, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.011327983811497688, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.008338441140949726, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0071250940673053265, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0066129290498793125, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0064889308996498585, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.00418152566999197, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.003446870483458042, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0033962696325033903, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.003094895975664258, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0030149081721901894, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0022162371315062046, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0022062344942241907, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0020290629472583532, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0015083705075085163, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.03057878278195858, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.03057878278195858, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1366012841463089, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12596023082733154, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12246476113796234, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10972391813993454, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06285402923822403, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.059158459305763245, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07101643085479736, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06538757681846619, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06403848528862, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05559698864817619, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05240226536989212, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036016568541526794, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03113798052072525, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029996821656823158, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029734907671809196, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01796874962747097, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015308110043406487, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01521964929997921, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013826324604451656, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013649583794176579, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009285662323236465, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009120416827499866, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008871985599398613, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005751997698098421, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036016568541526794, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036016568541526794, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.17603187263011932, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1397150307893753, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12485452741384506, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10236348956823349, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07880093902349472, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06356444209814072, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10127410292625427, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09088312089443207, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08408016711473465, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05868472903966904, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.054921627044677734, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.052208274602890015, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.044531937688589096, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.039118118584156036, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.037758395075798035, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026746269315481186, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0222683846950531, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.021844198927283287, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01885002851486206, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.017985740676522255, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015609919093549252, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01676454022526741, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014000470750033855, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01358905341476202, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.044531937688589096, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.044531937688589096, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16232417523860931, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15310928225517273, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1503237634897232, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13725094497203827, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07448019832372665, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07155297696590424, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08272626996040344, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07621025294065475, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07537022978067398, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06798406690359116, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06522101163864136, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04221388325095177, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.036436937749385834, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.035678111016750336, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.035506170243024826, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021163443103432655, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0187135711312294, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01866711862385273, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01746869832277298, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.01736268401145935, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011541726998984814, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011815913021564484, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011303811334073544, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00852925144135952, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04221388325095177, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04221388325095177, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20247410237789154, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.19122041761875153, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18785879015922546, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.17164653539657593, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09311875700950623, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08950629830360413, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10348823666572571, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.0952618271112442, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09424713999032974, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08508367091417313, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08159895986318588, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05248262733221054, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045330148190259933, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04438948258757591, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.044171132147312164, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026210876181721687, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.022723017260432243, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02266433648765087, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021111058071255684, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020973563194274902, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013842526823282242, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013543672859668732, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013542717322707176, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008824646472930908, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045330148190259933, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045330148190259933, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.20903810858726501, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.18752282857894897, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.17890337109565735, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16023726761341095, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.09530147165060043, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.08690676838159561, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.11254561692476273, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.10335908830165863, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.09833060950040817, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08278941363096237, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.0787888839840889, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.057394832372665405, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.04954644665122032, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.045908696949481964, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04501919075846672, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.028862686827778816, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.024214617908000946, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.023897942155599594, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02179906517267227, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02121691405773163, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.015693942084908485, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.015864508226513863, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014540915377438068, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011282749474048615, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04501919075846672, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04501919075846672, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.05185345932841301, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.04719923436641693, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.04519946873188019, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.04031410440802574, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.023669660091400146, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.021813875064253807, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.028894802555441856, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.025637667626142502, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.02446795627474785, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0208281222730875, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.019874243065714836, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.014878107234835625, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.012337449006736279, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.011372309178113937, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.011125385761260986, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.007496417500078678, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.005917070899158716, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.005843722261488438, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.005355468485504389, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.005197302903980017, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0039475723169744015, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0037921620532870293, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.003476022044196725, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0025562127120792866, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.04519946873188019, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.04519946873188019, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.04235349968075752, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.03876810520887375, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.03705023229122162, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.03299863263964653, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.019389156252145767, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.017897294834256172, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.023188047111034393, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.02107229083776474, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.019826212897896767, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.017041558399796486, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.016222914680838585, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.011777647770941257, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.010078520514070988, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.009334114380180836, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.009153480641543865, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.005901704076677561, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.004907215945422649, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.00483235577121377, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.004449155647307634, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.004337231628596783, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0031516216695308685, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.003196816658601165, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.002894952893257141, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0022504040971398354, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.04235349968075752, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.04235349968075752, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18337827920913696, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.16982707381248474, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.165360689163208, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.14849907159805298, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08514439314603806, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08030910789966583, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09676376730203629, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.08843616396188736, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08668725937604904, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07553431391716003, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07149364799261093, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.049151308834552765, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.042204901576042175, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.040709931403398514, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04035753011703491, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.024539658799767494, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.020764879882335663, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.020650843158364296, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.018789030611515045, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01855500042438507, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.012721888720989227, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.012331809848546982, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012158431112766266, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.007754776161164045, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.042204901576042175, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.042204901576042175, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.16824091970920563, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1404009759426117, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12616965174674988, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1017991229891777, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0763804018497467, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06389492750167847, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10492002964019775, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09072428941726685, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07963809370994568, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06006394326686859, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05551648139953613, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.053226761519908905, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04570390656590462, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03943958878517151, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03779958188533783, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.028513172641396523, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02437674067914486, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.023762907832860947, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.021884581074118614, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02097768895328045, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017941951751708984, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020007917657494545, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016066793352365494, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01738683134317398, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04570390656590462, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04570390656590462, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20486481487751007, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1926676630973816, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18893443048000336, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.17189179360866547, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09609905630350113, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0919579267501831, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1068822368979454, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09860783070325851, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09741038084030151, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08708394318819046, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08307784050703049, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0549030527472496, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04767584055662155, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04654709994792938, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04629192873835564, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02753591723740101, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024990206584334373, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024916432797908783, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02328149601817131, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02312212809920311, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015213009901344776, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016539441421628, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.0148587292060256, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012658458203077316, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02753591723740101, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02753591723740101, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.23720277845859528, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22319301962852478, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21892061829566956, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19929896295070648, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11099300533533096, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10614825785160065, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12349076569080353, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11382277309894562, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11246079951524734, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10049303621053696, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09580466151237488, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06302329897880554, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.054453324526548386, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05316462740302086, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05286480486392975, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03149754926562309, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0275109950453043, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0274287611246109, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02541448548436165, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.025231925770640373, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016771649941802025, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01682928204536438, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016346104443073273, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011489647440612316, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03149754926562309, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03149754926562309, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.04057672247290611, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.03628110885620117, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.029535584151744843, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.025871695950627327, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.01678905449807644, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.014954051934182644, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.02830415405333042, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.02541378326714039, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.017406944185495377, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.014085669070482254, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.013252400793135166, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.01061186008155346, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.00913564395159483, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.008270004764199257, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.00818908866494894, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.005477720871567726, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.004794950131326914, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.0044104899279773235, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.0043218075297772884, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.0038472977466881275, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.003534339601173997, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.0035675112158060074, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.0028007477521896362, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.002410768996924162, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.04057672247290611, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.04057672247290611, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.06802140921354294, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.06185317039489746, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.05932411551475525, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05280259996652603, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03121691197156906, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.028793297708034515, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.036546796560287476, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.033551882952451706, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.031954411417245865, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.027226997539401054, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.02570486068725586, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.018549954518675804, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.016023041680455208, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.014981724321842194, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.014730929397046566, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.009270134381949902, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.007769101765006781, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.007670868653804064, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.006976384669542313, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.006814697291702032, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0048570698127150536, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0048972973600029945, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.004509618505835533, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.003294609487056732, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.036546796560287476, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.036546796560287476, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.05969574674963951, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.054239239543676376, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0519106388092041, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.046200212091207504, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.027370501309633255, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.025188785046339035, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.03231668844819069, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.029567161574959755, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0280285831540823, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.023861512541770935, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.022590193897485733, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.01638929545879364, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.014107389375567436, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.013136859983205795, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.012899341061711311, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.008196202106773853, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.006800740025937557, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.006705775856971741, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0061028883792459965, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.005949399434030056, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.004289957694709301, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.004281290806829929, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.003958042245358229, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0028518664184957743, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.03231668844819069, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.03231668844819069, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.19189788401126862, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.17703129351139069, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1720646619796753, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15445537865161896, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0894722044467926, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08404400199651718, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10147964954376221, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09322547167539597, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09118306636810303, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07900489121675491, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07450595498085022, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.051586370915174484, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.044510357081890106, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.042796604335308075, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04238678142428398, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.025751352310180664, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02183552086353302, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.021708708256483078, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.019685404375195503, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01941804401576519, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013293273746967316, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.013021658174693584, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012664555571973324, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.008195354603230953, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.044510357081890106, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.044510357081890106, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1921081840991974, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1620737761259079, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14754262566566467, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1254829615354538, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08962292969226837, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07547061145305634, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11361818760633469, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10280504077672958, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09331086277961731, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0703839585185051, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06639263778924942, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05926152691245079, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05044787377119064, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04434806853532791, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04279506951570511, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030087675899267197, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02480752021074295, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024201147258281708, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02138356678187847, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02041790261864662, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01701258309185505, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018280688673257828, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015084261074662209, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014460260048508644, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04434806853532791, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04434806853532791, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21599748730659485, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20252850651741028, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1984284371137619, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1801484227180481, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10129519551992416, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09660668671131134, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11259352415800095, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10407939553260803, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10276809334754944, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09125761687755585, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08659656345844269, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.057402390986680984, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.049780718982219696, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04851917922496796, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04822508618235588, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028642985969781876, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02503836713731289, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02495640516281128, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0230195764452219, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022827928885817528, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015057935379445553, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015232334844768047, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014643209055066109, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010247326456010342, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028642985969781876, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028642985969781876, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2490445375442505, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.233660027384758, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22894904017448425, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20791460573673248, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11674238741397858, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11135952174663544, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1298394501209259, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11990918964147568, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11843188852071762, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10519439727067947, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09987063705921173, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06598417460918427, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.057234637439250946, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05580367147922516, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.055463314056396484, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.032915420830249786, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02840307541191578, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02831517532467842, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02602136880159378, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02580972947180271, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017021603882312775, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01669299043715, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01654033362865448, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010465314611792564, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.032915420830249786, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.032915420830249786, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2560821771621704, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.23009999096393585, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.22051246464252472, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.19461233913898468, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.11875905096530914, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.10878690332174301, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1381164938211441, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.12697482109069824, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12221168726682663, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10192594677209854, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.09561078995466232, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07073277235031128, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06095147132873535, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.057230543345212936, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.056317903101444244, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03547146916389465, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02992868795990944, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.029623717069625854, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.026605812832713127, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.026000238955020905, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.019139563664793968, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01914084516465664, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.017940912395715714, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013314539566636086, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03547146916389465, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03547146916389465, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.07989566028118134, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07274160534143448, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.06960595399141312, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06191140413284302, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.036777935922145844, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.033852726221084595, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04314282909035683, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.039741165935993195, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.03763621672987938, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.032070450484752655, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.030246511101722717, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02192327193915844, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.01900540105998516, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.017688220366835594, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.017368242144584656, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.010967201553285122, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.009237012825906277, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.009107418358325958, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008312014862895012, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.008107040077447891, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005783803295344114, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005924303084611893, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00534060038626194, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004088525660336018, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04314282909035683, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04314282909035683, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.06946519762277603, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.06325475126504898, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.060390274971723557, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05374712496995926, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.031992994248867035, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.02939046174287796, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.038003064692020416, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.034788090735673904, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.032743144780397415, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.027903839945793152, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.02644449658691883, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.019295385107398033, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.016624853014945984, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.01537884771823883, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.01507872249931097, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.009657888673245907, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.008007482625544071, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.007882919162511826, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.007197997998446226, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.007002964150160551, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0050788032822310925, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005114185158163309, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.004658810328692198, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0034712543711066246, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.038003064692020416, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.038003064692020416, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.19984009861946106, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.18441124260425568, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1791190356016159, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16063307225704193, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09335442632436752, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08762026578187943, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10615657269954681, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09738928824663162, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09516817331314087, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08232954144477844, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07761746644973755, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05401588976383209, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04652368277311325, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04468501731753349, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04425185173749924, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026964152231812477, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02280684933066368, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02266502007842064, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.020538097247481346, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.020256970077753067, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013947341591119766, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0136131402105093, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013264857232570648, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.008579010143876076, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04468501731753349, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04468501731753349, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.19775232672691345, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.17410361766815186, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1644924283027649, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1367560774087906, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0923662781715393, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08279867470264435, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11088012903928757, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10082745552062988, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09564351290464401, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0753738135099411, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06781109422445297, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05746222659945488, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04924855753779411, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04538898169994354, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04444888234138489, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02927042730152607, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.024996737018227577, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024688705801963806, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02188492938876152, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021299906075000763, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016619911417365074, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017674332484602928, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015406832098960876, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013905821368098259, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04538898169994354, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04538898169994354, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19429819285869598, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18185053765773773, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17794838547706604, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16125062108039856, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09132584929466248, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08692395687103271, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10183271020650864, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09399432688951492, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09268423169851303, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08200319111347198, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07781219482421875, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.052065420895814896, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045087914913892746, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0438711941242218, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04358743131160736, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02602704055607319, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02284390665590763, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.022761942818760872, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02099001780152321, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02080911211669445, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013883918523788452, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01420542597770691, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013488121330738068, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009911693632602692, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045087914913892746, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045087914913892746, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25108203291893005, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.235168918967247, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23031216859817505, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20874090492725372, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11794835329055786, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11231587827205658, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1315440535545349, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1213437020778656, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11970092356204987, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10594069212675095, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10051924735307693, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06687448173761368, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05794357508420944, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.056411005556583405, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.056036774069070816, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03339743986725807, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02872651070356369, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.028624799102544785, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0262470506131649, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02601630985736847, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01734539307653904, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016912657767534256, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016817264258861542, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010602129623293877, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03339743986725807, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03339743986725807, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2512768507003784, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.22530192136764526, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.21543756127357483, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.18966403603553772, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.11664469540119171, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1064477264881134, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.13626094162464142, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.12523253262043, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12004589289426804, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09971853345632553, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.0932636484503746, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06975293159484863, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06018700823187828, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05628325790166855, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05534457042813301, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03506321832537651, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.0295726265758276, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.029249563813209534, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.026253025978803635, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.025623571127653122, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.019048551097512245, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01911821775138378, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.017817113548517227, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01347265113145113, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03506321832537651, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03506321832537651, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09342074394226074, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08596014231443405, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08290400356054306, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07418781518936157, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04341007396578789, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0404345877468586, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.050247762352228165, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04620054364204407, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04429296776652336, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03824499249458313, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03611163794994354, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02555239014327526, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02213199809193611, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.020887229591608047, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.020590482279658318, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012800530530512333, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010899175889790058, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010780710726976395, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009872202761471272, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009685307741165161, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006759651470929384, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00691310316324234, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006337600760161877, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004780549556016922, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04429296776652336, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04429296776652336, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.07777926325798035, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07159759104251862, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.06890153884887695, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06175904721021652, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03616051748394966, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03361446037888527, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04236523434519768, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.03874180465936661, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.03690281882882118, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03188832104206085, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03025646321475506, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02152993157505989, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.01853025332093239, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.017374595627188683, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.01709558255970478, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.010771818459033966, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.009012225084006786, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.008898084983229637, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008156399242579937, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.007980176247656345, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005655617453157902, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00566718028858304, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005264543928205967, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0038147657178342342, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04236523434519768, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04236523434519768, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21795782446861267, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.20187292993068695, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.196472629904747, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.176476389169693, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10211807489395142, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09618857502937317, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11575832962989807, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10624463111162186, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10399461537599564, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0904296338558197, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08536628633737564, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05886729806661606, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05074343457818031, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04886016622185707, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.048412103205919266, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029368555173277855, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.024877425283193588, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024732869118452072, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02245517447590828, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02216355875134468, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01512114703655243, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014728743582963943, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014408944174647331, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009145919233560562, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029368555173277855, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029368555173277855, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2189275324344635, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.18272723257541656, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.16757158935070038, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.13911324739456177, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10109258443117142, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08529244363307953, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12503904104232788, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11374403536319733, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10557595640420914, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07765942066907883, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07122000306844711, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06471056491136551, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05535588040947914, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0495736300945282, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04814007878303528, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03258540481328964, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02723502553999424, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026723047718405724, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.022984763607382774, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.022064270451664925, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018207529559731483, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019413238391280174, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016385238617658615, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015016155317425728, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03258540481328964, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03258540481328964, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20153743028640747, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18851664662361145, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1843298077583313, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16674545407295227, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09483902156352997, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09012675285339355, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10593617707490921, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09781967848539352, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09628626704216003, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08497647196054459, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08050037920475006, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0540010929107666, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04682955890893936, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.045476529747247696, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04515392333269119, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026978952810168266, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023431820794939995, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023334382101893425, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021433016285300255, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021227583289146423, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014168049208819866, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014234275557100773, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013714038766920567, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009488156996667385, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.045476529747247696, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.045476529747247696, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25892654061317444, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.242341086268425, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23709000647068024, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21466562151908875, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12187721580266953, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11592841148376465, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1361589878797531, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12562349438667297, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12373150885105133, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10929025709629059, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10361145436763763, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06937889754772186, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.060043394565582275, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05834856629371643, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05795224756002426, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03465110436081886, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029777899384498596, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029660357162356377, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027180060744285583, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026921402662992477, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01807631552219391, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017635686323046684, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017496872693300247, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011170043610036373, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03465110436081886, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03465110436081886, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2639017403125763, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.23614276945590973, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.22541052103042603, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.19847679138183594, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12250929325819016, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.111509308218956, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1439986228942871, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1322837769985199, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12636543810367584, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10471688956022263, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.09812995046377182, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07373453676700592, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06364434957504272, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05921094864606857, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05812234804034233, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.037028200924396515, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03120032697916031, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.030821677297353745, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02771713398396969, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.0269988514482975, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02006096951663494, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.020346883684396744, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01865379884839058, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014430559240281582, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.037028200924396515, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.037028200924396515, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09632465988397598, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08879877626895905, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08568572998046875, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07667210698127747, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.044806428253650665, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04175376147031784, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.051858045160770416, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.047687504440546036, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04568152874708176, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03949415683746338, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03733139485120773, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02637634240090847, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.022813718765974045, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02151467092335224, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.021209105849266052, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013188492506742477, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011142122559249401, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011015201918780804, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010076837614178658, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009878472425043583, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006906233262270689, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006960798986256123, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006460003554821014, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004669055342674255, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04568152874708176, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04568152874708176, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08182700723409653, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07547298073768616, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.07261616736650467, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06501644104719162, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03802809491753578, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03536989167332649, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04444113373756409, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04087189584970474, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.03876248002052307, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.033587321639060974, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03182363137602806, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02260255441069603, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.019539866596460342, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.018267419189214706, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.017962096258997917, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.011295142583549023, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.009453092701733112, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.00932040810585022, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008557567372918129, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.008360382169485092, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005907882936298847, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005923627410084009, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005482418462634087, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0039366367273032665, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04444113373756409, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04444113373756409, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2144889086484909, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19871997833251953, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1933867484331131, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1738104671239853, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10055331885814667, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09469331800937653, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11438227444887161, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10466358065605164, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10233313590288162, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08901773393154144, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08417291939258575, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05823187157511711, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04999694228172302, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04812197387218475, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04767283424735069, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02906237170100212, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02450665831565857, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02435685135424137, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.022119924426078796, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021827826276421547, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014970541000366211, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014529786072671413, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014215321280062199, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009033918380737305, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02906237170100212, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02906237170100212, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22690607607364655, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1987263411283493, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1880594789981842, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1558893322944641, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10537312179803848, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09416312724351883, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12500913441181183, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11414141952991486, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10909385979175568, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08525621145963669, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07700890302658081, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06470014154911041, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05574289709329605, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.051787711679935455, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.050847478210926056, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0328216515481472, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028716176748275757, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02841464802622795, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02508723922073841, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02449544332921505, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018701840192079544, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020420968532562256, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017533322796225548, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01628854312002659, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0328216515481472, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0328216515481472, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.18069122731685638, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.16885876655578613, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1649756282567978, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14925037324428558, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08509469777345657, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08077280968427658, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09528975188732147, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08790664374828339, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08642090111970901, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0761522501707077, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07217637449502945, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0486113503575325, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.042148407548666, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.040867507457733154, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.040563762187957764, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.024330688640475273, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.021177196875214577, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.021082038059830666, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019380349665880203, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019189834594726562, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012907501310110092, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013045677915215492, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012484082020819187, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00890553742647171, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.042148407548666, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.042148407548666, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2515871524810791, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23530030250549316, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23013737797737122, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20830963551998138, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11869160085916519, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11276856064796448, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13269902765750885, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.122407928109169, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12051985412836075, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10627660155296326, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1008661538362503, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06778927147388458, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.058660369366407394, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05695952847599983, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05656351149082184, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03385477885603905, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029437659308314323, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029319314286112785, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026934141293168068, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026681631803512573, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.0178541149944067, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018009385094046593, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017281973734498024, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012165137566626072, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03385477885603905, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03385477885603905, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2515970468521118, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2257971614599228, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.21589826047420502, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.19004520773887634, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.11698701232671738, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.10678610950708389, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1382056325674057, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.12571334838867188, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12040290236473083, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10005395114421844, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.09370862692594528, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07028000056743622, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06063983216881752, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05670979246497154, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05576162412762642, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03550896793603897, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03025519661605358, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.029926180839538574, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.027014413848519325, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.026390086859464645, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.019719336181879044, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02013586461544037, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.018483931198716164, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014847105368971825, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03550896793603897, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03550896793603897, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10369611531496048, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09573040902614594, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09240089356899261, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08277066051959991, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04832262545824051, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04508519917726517, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05595738813281059, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05146252363920212, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.049239132553339005, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04266250506043434, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04034283757209778, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02847190760076046, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.024609403684735298, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023210305720567703, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02286842279136181, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014232064597308636, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012028424069285393, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011892342939972878, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010895694606006145, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01068020798265934, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007467102725058794, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007527426816523075, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0069931079633533955, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005070744082331657, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04266250506043434, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04266250506043434, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08777689188718796, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08108292520046234, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.07799377292394638, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06991248577833176, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04088285565376282, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03804956004023552, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.047706518322229385, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04393434897065163, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0416577011346817, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.036147382110357285, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.034244854003190994, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02425253391265869, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.021013736724853516, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.019647927954792976, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.019313251599669456, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0121358223259449, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010173352435231209, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010029671713709831, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009221336804330349, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009009500965476036, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006365620996803045, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00638926075771451, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005924845114350319, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004262920934706926, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04393434897065163, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04393434897065163, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2318304479122162, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21520951390266418, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20946061611175537, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.188112273812294, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10885079205036163, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10257744789123535, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12418785691261292, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11333763599395752, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11077068001031876, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0964425653219223, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09127992391586304, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06328535079956055, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05415377765893936, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0521090105175972, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05162351578474045, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03158356621861458, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02652743086218834, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026366882026195526, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023965975269675255, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023646360263228416, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016277462244033813, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01572575233876705, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015426225028932095, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009778434410691261, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03158356621861458, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03158356621861458, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2314879596233368, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.20091789960861206, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18995323777198792, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16072654724121094, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10818406939506531, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09636186808347702, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12697367370128632, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11615433543920517, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11175526678562164, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08687476813793182, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0786866620182991, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06578802317380905, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05663538724184036, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.053079359233379364, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.052233900874853134, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033270448446273804, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029079684987664223, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028815485537052155, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02514747716486454, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024604149162769318, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01881938986480236, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02024136669933796, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017780359834432602, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01587783358991146, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033270448446273804, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033270448446273804, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20058202743530273, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1873994618654251, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18309053778648376, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1655862033367157, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09474731236696243, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08991366624832153, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10613948106765747, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09792641550302505, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09621939063072205, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0847189873456955, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08023407310247421, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05432399362325668, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.047055311501026154, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.045610230416059494, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.045274894684553146, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02716882713139057, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02380426973104477, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023703034967184067, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021810805425047874, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021595511585474014, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01450083777308464, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014909529127180576, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014023861847817898, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010476471856236458, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.045610230416059494, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.045274894684553146, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2600115239620209, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2430143654346466, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23763319849967957, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21492896974086761, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12258347868919373, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11640802770853043, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13734544813632965, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1265084445476532, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12449903786182404, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1096700057387352, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1038644015789032, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06991986930370331, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06050854176282883, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05872839316725731, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.058297328650951385, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03494943305850029, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02998899109661579, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029860548675060272, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027327779680490494, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027052175253629684, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018247907981276512, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01780744642019272, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017628278583288193, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011303630657494068, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03494943305850029, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03494943305850029, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27074307203292847, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.24188564717769623, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23107293248176575, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.2031693458557129, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1260264813899994, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11467485874891281, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.14777901768684387, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1353924721479416, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12985052168369293, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10728605836629868, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10023079812526703, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.0756993442773819, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06531833112239838, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.061069536954164505, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.060045354068279266, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03809594362974167, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03245807811617851, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03210998326539993, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02884609065949917, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028159067034721375, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.020854458212852478, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021465180441737175, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019494004547595978, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015666062012314796, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03809594362974167, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03809594362974167, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11482495814561844, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10601693391799927, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10234712064266205, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09178683161735535, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05367222800850868, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.050100117921829224, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.062106918543577194, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0571017824113369, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05469592660665512, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04745471104979515, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.044913798570632935, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03172145411372185, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02745014801621437, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.025923481211066246, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025549400597810745, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015919161960482597, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013666660524904728, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0135190449655056, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01244641188532114, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01221828255802393, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008521875366568565, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008863285183906555, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008007150143384933, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006361360661685467, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.044913798570632935, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.044913798570632935, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10075456649065018, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09313247352838516, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08975953608751297, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0805126205086708, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04706370830535889, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04385872185230255, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05476916581392288, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.050356920808553696, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.047962114214897156, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.041631247848272324, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03947988897562027, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.027897926047444344, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.024103589355945587, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.022618183866143227, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.022260552272200584, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013959130272269249, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01172860711812973, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011577680706977844, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010637292638421059, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010409495793282986, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007332778535783291, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007360531482845545, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0068368674255907536, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004940945655107498, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.041631247848272324, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.041631247848272324, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23074829578399658, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2140057384967804, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20822811126708984, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18714262545108795, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10839910060167313, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10211613029241562, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12385618686676025, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11298201978206635, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11035701632499695, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09609342366456985, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09096714109182358, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06309351325035095, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05402425676584244, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.051920387893915176, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05140665918588638, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03150093927979469, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02649225853383541, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02632238157093525, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02395421266555786, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023624040186405182, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016294365748763084, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015792706981301308, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015430387109518051, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009932780638337135, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03150093927979469, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03150093927979469, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24043138325214386, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2042328268289566, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18963338434696198, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15439364314079285, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10985501110553741, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09561122208833694, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13639497756958008, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12395461648702621, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1155029758810997, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0883105993270874, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07813206315040588, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07033239305019379, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0608506016433239, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.054620444774627686, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.053065549582242966, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03582107275724411, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03096597082912922, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03045610897243023, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027189671993255615, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026251304894685745, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020656099542975426, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02300897240638733, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01880006678402424, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.018719982355833054, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03582107275724411, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03582107275724411, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19935329258441925, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18623758852481842, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18186841905117035, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16445647180080414, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09415768831968307, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08929870277643204, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10552059859037399, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09740261733531952, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0956631600856781, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08419371396303177, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07972045987844467, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05396536365151405, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04678715392947197, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04531329870223999, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.044971100986003876, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026982121169567108, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02362668141722679, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023520562797784805, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02164069004356861, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021420739591121674, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01433518249541521, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014784770086407661, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013839900493621826, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010350920259952545, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04531329870223999, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04531329870223999, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25827115774154663, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2413359433412552, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23591133952140808, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21331259608268738, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1217716634273529, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11560685932636261, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1365126371383667, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1257922202348709, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12370189279317856, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10891231149435043, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10312610864639282, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06947852671146393, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06016315147280693, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05834554508328438, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05790737271308899, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03469975292682648, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029775897040963173, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029643140733242035, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02711157500743866, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026833824813365936, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018069621175527573, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017660636454820633, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01743265427649021, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011179466731846333, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03469975292682648, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03469975292682648, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2728189527988434, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2448965162038803, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23437978327274323, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20617325603961945, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1274712234735489, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11630194634199142, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.14893494546413422, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13668011128902435, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1311267912387848, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10883990675210953, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10171136260032654, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07656236737966537, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06602233648300171, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06184357777237892, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.060825541615486145, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03856963291764259, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03301317244768143, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03265116736292839, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029435185715556145, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028759140521287918, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02127951942384243, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021983209997415543, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019926277920603752, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.016225505620241165, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03856963291764259, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03856963291764259, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11832495778799057, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10937527567148209, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10570964217185974, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09477504342794418, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05527966469526291, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.051669079810380936, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06379207223653793, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05865029990673065, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.056303009390830994, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04887635260820389, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0461837463080883, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03250046819448471, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02812815271317959, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026602765545248985, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02623044326901436, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016266891732811928, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013850049115717411, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013700202107429504, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012574299238622189, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012339376844465733, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008570823818445206, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008741246536374092, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008052418008446693, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006006124895066023, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03250046819448471, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03250046819448471, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1002717986702919, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09270406514406204, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08925183862447739, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08007241040468216, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04674166813492775, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04354751482605934, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05446474999189377, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05014651641249657, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04763338714838028, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04138678312301636, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03917384520173073, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.027681877836585045, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02398534305393696, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.022471914067864418, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.022098029032349586, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013860233128070831, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011644206941127777, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011488736607134342, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010564789175987244, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010331675410270691, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0072797806933522224, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007318869233131409, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006784508470445871, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004908010829240084, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04138678312301636, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04138678312301636, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2284030318260193, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2118704468011856, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20581631362438202, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18485917150974274, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10721405595541, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10083728283643723, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12242737412452698, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11231036484241486, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1091451346874237, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09500974416732788, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0897214338183403, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06234397366642952, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05370258539915085, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05135170370340347, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05078493431210518, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03110862337052822, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026229653507471085, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026019791141152382, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02371359057724476, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023342190310359, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016086317598819733, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01573438011109829, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01520710438489914, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009923754259943962, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03110862337052822, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03110862337052822, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24778415262699127, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2167220413684845, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20470206439495087, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.17519785463809967, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11567223817110062, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10320711135864258, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13715678453445435, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12585775554180145, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11942990869283676, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09542050957679749, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08631546050310135, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07111173123121262, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06167472526431084, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05702796205878258, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05591916665434837, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036104682832956314, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03169924020767212, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.031292907893657684, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02800498902797699, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02731560543179512, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020602956414222717, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.022653240710496902, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019270500168204308, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.018092898651957512, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036104682832956314, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036104682832956314, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1920599639415741, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.17933635413646698, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17510007321834564, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1581493616104126, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09073381125926971, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08604716509580612, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1018465980887413, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09391605854034424, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0921892523765564, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08107004314661026, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07686526328325272, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05217289179563522, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045211486518383026, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.043770868331193924, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04343196004629135, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02613113634288311, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.022997459396719933, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.022892026230692863, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021095626056194305, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020885087549686432, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014067025855183601, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.0146308159455657, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013598688878118992, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010516759008169174, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045211486518383026, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045211486518383026, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25226321816444397, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23567397892475128, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2303275763988495, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2081655114889145, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11880990117788315, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11274490505456924, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13335682451725006, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12274311482906342, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12068751454353333, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10617882758378983, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1006149873137474, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06780941784381866, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05870605632662773, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.056921783834695816, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.056500911712646484, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0339045487344265, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02909567952156067, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.028967147693037987, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0264913197606802, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026224592700600624, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017756078392267227, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017325712367892265, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01715136133134365, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011055948212742805, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0339045487344265, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0339045487344265, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27459484338760376, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.24643351137638092, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23597204685211182, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20759472250938416, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1281907856464386, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1171615868806839, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1490473598241806, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13711504638195038, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.131874680519104, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10945253819227219, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10220873355865479, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07652569562196732, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06621236354112625, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.062132224440574646, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06113894283771515, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03849806636571884, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03306775912642479, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03273302689194679, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029457401484251022, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028802508488297462, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.0210846196860075, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02189345471560955, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019826499745249748, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01608719862997532, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03849806636571884, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03849806636571884, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11961842328310013, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1107267439365387, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10703016817569733, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09597490727901459, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055907342582941055, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05227242410182953, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06457865983247757, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05937295779585838, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.056914590299129486, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0494857020676136, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04679582640528679, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032879557460546494, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02843494527041912, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026851220056414604, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026468966156244278, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01644437201321125, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013914564624428749, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013754871673882008, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012625698000192642, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012383894994854927, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008633725345134735, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00869670882821083, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008094281889498234, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005845505278557539, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032879557460546494, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032879557460546494, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10334303230047226, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09577998518943787, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0922524705529213, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08274383842945099, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04824664816260338, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04498319327831268, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05615890771150589, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.051764119416475296, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04910949990153313, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04277252033352852, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04050590470433235, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.028546657413244247, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.024740546941757202, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023172510787844658, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02278936840593815, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014275959692895412, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011973225511610508, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011804772540926933, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010868093930184841, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010623749345541, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007468098308891058, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007478430867195129, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006953330710530281, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004948832560330629, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04277252033352852, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04277252033352852, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22755979001522064, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21125631034374237, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2051563411951065, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18423786759376526, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10675116628408432, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10044249892234802, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12217427045106888, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11191225051879883, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10867118090391159, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09465666115283966, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08937503397464752, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.062164925038814545, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05350256338715553, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.051153652369976044, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05059317499399185, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.031043939292430878, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026151767000555992, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02594125084578991, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02366144396364689, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023291876539587975, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01606619358062744, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015714623034000397, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015184963122010231, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009952682070434093, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.031043939292430878, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.031043939292430878, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2390078902244568, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2088969647884369, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1959220916032791, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16803506016731262, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11102032661437988, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09859540313482285, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13525058329105377, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12357156723737717, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11555469036102295, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09153911471366882, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08503027260303497, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06984677910804749, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06022628769278526, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05448238551616669, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.053065646439790726, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035220060497522354, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029898613691329956, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02939874865114689, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.026298733428120613, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02542000077664852, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019559569656848907, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.021176494657993317, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017775634303689003, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.016408488154411316, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035220060497522354, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035220060497522354, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.18787886202335358, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.17543353140354156, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17118988931179047, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.15457534790039062, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08879992365837097, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08417021483182907, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09978684037923813, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09206059575080872, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09025876969099045, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07933007180690765, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07510299980640411, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.051111217588186264, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0443217009305954, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.042863424867391586, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04251554608345032, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.025592971593141556, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.022557178512215614, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.022448400035500526, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020680442452430725, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02047467604279518, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013735776767134666, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01439524907618761, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013253915123641491, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01039867103099823, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0443217009305954, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0443217009305954, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2473154515028, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23090507090091705, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22554846107959747, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20361341536045074, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1165388971567154, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11051178723573685, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1307946890592575, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1204962283372879, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11837656795978546, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10404131561517715, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09835748374462128, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06650847941637039, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05761805549263954, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.055831823498010635, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05541135370731354, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03319796547293663, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.028527352958917618, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.028398120775818825, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.025952065363526344, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.0256760623306036, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01722279004752636, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016980748623609543, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016594743356108665, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010808971710503101, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03319796547293663, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03319796547293663, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2714840769767761, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.24420498311519623, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23392127454280853, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20583079755306244, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12696103751659393, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11603502184152603, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.147410050034523, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1357862800359726, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13048817217350006, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.108528733253479, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10143455862998962, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07582192122936249, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06555668264627457, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.061418209224939346, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06044558063149452, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.038130734115839005, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03251330554485321, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.032167356461286545, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02893120050430298, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028265463188290596, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02086016908288002, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02130795642733574, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019599638879299164, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015379992313683033, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.038130734115839005, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.038130734115839005, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12412507086992264, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11490044742822647, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11114231497049332, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09961853921413422, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05805591121315956, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05430590733885765, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06683126837015152, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.061494696885347366, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05911809951066971, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05135932192206383, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04851604998111725, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034010857343673706, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02941947430372238, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.027865251526236534, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02748895436525345, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017007872462272644, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014367649331688881, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014218747615814209, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013012171722948551, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012771431356668472, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008890165947377682, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00887041725218296, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008359011262655258, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0058511653915047646, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034010857343673706, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034010857343673706, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1067131981253624, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09884549677371979, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0953369140625, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08554266393184662, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.049855928868055344, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.046520959585905075, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05790611356496811, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.053327664732933044, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.050769656896591187, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.044196613132953644, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04183930158615112, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.029430603608489037, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02551310881972313, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023927325382828712, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02354568801820278, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014703894965350628, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012337890453636646, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012172284536063671, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01118840929120779, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01094226073473692, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00768020935356617, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007655879948288202, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0071683600544929504, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005014003254473209, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.044196613132953644, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.044196613132953644, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23614585399627686, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2190166413784027, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21266262233257294, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19097447395324707, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11087939888238907, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10424335300922394, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12692780792713165, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11633018404245377, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1128903180360794, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09830792993307114, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09280107170343399, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06464593857526779, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05570399388670921, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05316006392240524, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05254821106791496, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03232071176171303, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027250219136476517, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027020005509257317, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024664677679538727, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024275025352835655, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01680048368871212, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016488268971443176, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015877092257142067, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010576008819043636, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03232071176171303, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03232071176171303, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25126323103904724, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2198578119277954, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2087208330631256, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1818307787179947, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11712852120399475, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10518012940883636, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13739898800849915, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12580369412899017, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12106150388717651, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09708819538354874, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08984112739562988, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07061974704265594, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06108774244785309, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05718855932354927, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05624726414680481, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0355960987508297, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.031108440831303596, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03081481158733368, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02737552672624588, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026778120547533035, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019785482436418533, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.021411923691630363, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018602045252919197, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.016535956412553787, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0355960987508297, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0355960987508297, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19275899231433868, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.179849311709404, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17551864683628082, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1585119068622589, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0911775678396225, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08636685460805893, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10230085998773575, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09448163956403732, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09265153855085373, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08139344304800034, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07705357670783997, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05246438458561897, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04552745819091797, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0440274141728878, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04367553070187569, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026243261992931366, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02321728691458702, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02310815453529358, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021301761269569397, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021080801263451576, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01407904177904129, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014887068420648575, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013586314395070076, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010810446925461292, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04552745819091797, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04552745819091797, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25015729665756226, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2334522157907486, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22800278663635254, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20590978860855103, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11788057535886765, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11178041249513626, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13229531049728394, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12190206348896027, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11977075785398483, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10522563010454178, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09954701364040375, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06731835752725601, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05827615037560463, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05645081028342247, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05601038783788681, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.033608049154281616, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02878337912261486, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02865052968263626, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026166070252656937, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.025885669514536858, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017407922074198723, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017042016610503197, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016758428886532784, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010714935138821602, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.033608049154281616, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.033608049154281616, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2780321538448334, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2499094009399414, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23971322178840637, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21093210577964783, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1301226019859314, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11908623576164246, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1508653461933136, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13859808444976807, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13377049565315247, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11103161424398422, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10378414392471313, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07760332524776459, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06690609455108643, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06302021443843842, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06209125369787216, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039086807519197464, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.033443838357925415, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.033139005303382874, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029743218794465065, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.029125401750206947, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02150120586156845, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021974576637148857, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02029324695467949, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.016018036752939224, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039086807519197464, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039086807519197464, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13113559782505035, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12138572335243225, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11742791533470154, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10537087172269821, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06140151247382164, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0574834831058979, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07067259401082993, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06497567147016525, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06253236532211304, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.054375551640987396, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05140304192900658, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036023955792188644, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.031171884387731552, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029538556933403015, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029145896434783936, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01803092658519745, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015345904976129532, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015193148516118526, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013944071717560291, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013694124296307564, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009488307870924473, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009637314826250076, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008936390280723572, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006573929451406002, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036023955792188644, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036023955792188644, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10967464745044708, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10156024992465973, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09769334644079208, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08764217793941498, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.051218099892139435, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04769313335418701, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05985935777425766, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05516694486141205, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05216686800122261, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04540940001606941, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.043026119470596313, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03043121099472046, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026387883350253105, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024602390825748444, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024169525131583214, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015224800445139408, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012709623202681541, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012518607079982758, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011536316946148872, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011259270831942558, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007974592968821526, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007946358993649483, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007406032178550959, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005240539088845253, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04540940001606941, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04540940001606941, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2450898140668869, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22752606868743896, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2211723029613495, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1986207515001297, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11521892249584198, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10838443040847778, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1316971480846405, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12060663104057312, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11722657829523087, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10225662589073181, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09657502919435501, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06715323776006699, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05777803435921669, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05522901564836502, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05461432784795761, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033562611788511276, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028299396857619286, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028074145317077637, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02564343996345997, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025248100981116295, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017429670318961143, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01710451766848564, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01647895947098732, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010964632965624332, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033562611788511276, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033562611788511276, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2552471160888672, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22352276742458344, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2125452756881714, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18046170473098755, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.119160495698452, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10669536888599396, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1398310661315918, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12745535373687744, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12316528707742691, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09792082756757736, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0882132425904274, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07246826589107513, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06233877316117287, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05855654925107956, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05765283852815628, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03670142963528633, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03214259445667267, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03187273070216179, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02823736146092415, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02766951359808445, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020841877907514572, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.022428343072533607, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019747724756598473, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.017673172056674957, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03670142963528633, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03670142963528633, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19195245206356049, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1791805624961853, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17487166821956635, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.15797221660614014, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09083784371614456, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08607970923185349, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10198046267032623, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09410202503204346, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09230750799179077, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08112085610628128, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07680906355381012, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05227109417319298, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04535679146647453, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04387743026018143, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04352235049009323, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026190361008048058, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023131201043725014, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023017842322587967, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021219396963715553, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021002471446990967, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014123690314590931, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014825055375695229, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01364320982247591, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010763069614768028, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04535679146647453, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04535679146647453, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25185322761535645, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23509790003299713, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22969330847263336, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20741896331310272, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11874396353960037, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11262624710798264, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13309119641780853, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12272529304027557, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1206497773528099, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10598786920309067, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10023944079875946, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06778079271316528, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05867237597703934, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0568716824054718, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05643634870648384, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0338369719684124, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02901609241962433, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02889033779501915, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02638489194214344, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026104357093572617, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017581230029463768, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017206136137247086, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016962628811597824, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010867305099964142, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0338369719684124, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0338369719684124, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2756015956401825, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.24685999751091003, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23618605732917786, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20779898762702942, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1285080462694168, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11718788743019104, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.14996080100536346, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13766694068908691, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13236793875694275, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10946547985076904, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10244926810264587, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07684614509344101, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06626468151807785, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06214132905006409, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06113908812403679, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03855043649673462, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03278004750609398, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03244725242257118, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02904708869755268, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02837621048092842, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.020842377096414566, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02133289724588394, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019529003649950027, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015236050821840763, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03855043649673462, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03855043649673462, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13307558000087738, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12341082096099854, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11955305933952332, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1073569655418396, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06230218708515167, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05844829976558685, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07156039774417877, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06582488864660263, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06342251598834991, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05527867004275322, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05227303132414818, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03641306608915329, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.031492650508880615, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029903054237365723, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02952246181666851, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01820097304880619, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015426873229444027, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01527125108987093, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014003783464431763, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013760870322585106, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009507996961474419, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009515409357845783, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008953319862484932, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006289742887020111, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03641306608915329, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03641306608915329, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1111006811261177, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10309913009405136, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09939577430486679, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08926796168088913, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.051921866834163666, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0484701469540596, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06036118417978287, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.055669672787189484, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.052828218787908554, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04611543193459511, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04368293657898903, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030657432973384857, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026612652465701103, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024924248456954956, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024515777826309204, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015320413745939732, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012838952243328094, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012657574377954006, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01166123803704977, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011396647430956364, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007997463457286358, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00796564295887947, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007454344537109137, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005191242787986994, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04368293657898903, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04368293657898903, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24466833472251892, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2275039106607437, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22084200382232666, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19875146448612213, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11513343453407288, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10828675329685211, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13198637962341309, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12090422958135605, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11707442998886108, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10230621695518494, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09678825736045837, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06720864772796631, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.057934410870075226, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05520079657435417, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05453282594680786, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03358803316950798, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028254834935069084, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02799488790333271, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025628486648201942, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025205645710229874, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017471298575401306, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017065152525901794, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016486436128616333, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010868030600249767, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03358803316950798, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03358803316950798, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2626953125, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2371172159910202, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22856567800045013, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1992691606283188, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12329432368278503, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11340247839689255, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14180071651935577, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12964902818202972, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12623822689056396, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10461258143186569, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09664768725633621, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07349345833063126, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06330861896276474, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06044843792915344, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.059783533215522766, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037398286163806915, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.033077422529459, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.032896365970373154, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.029617195948958397, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.029215099290013313, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02148682065308094, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.022806953638792038, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020654108375310898, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.017987878993153572, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037398286163806915, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037398286163806915, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1993301957845688, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18600814044475555, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18155589699745178, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16399720311164856, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09423808008432388, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08928560465574265, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10578227043151855, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09765830636024475, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09575429558753967, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08416790515184402, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07969941943883896, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.054184842854738235, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04698435217142105, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04544079676270485, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04507853835821152, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027104981243610382, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023835526779294014, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02371927909553051, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02183830365538597, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021612990647554398, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014494331553578377, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015107979997992516, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01398254930973053, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01077970676124096, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04544079676270485, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04544079676270485, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2546945810317993, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23785150051116943, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23234938085079193, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20978523790836334, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12012632936239243, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11392276734113693, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13471317291259766, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12416063994169235, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12202334403991699, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10725029557943344, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10140564292669296, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06858764588832855, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05937677249312401, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.057529184967279434, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05709048733115196, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03421599417924881, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029357390478253365, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029220273718237877, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02670006826519966, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026416637003421783, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017760736867785454, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.0174194797873497, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017122648656368256, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01099492609500885, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03421599417924881, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03421599417924881, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27956679463386536, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2501390874385834, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23921217024326324, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21031597256660461, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13038542866706848, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11868518590927124, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1523456871509552, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13976293802261353, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1343250870704651, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11084669083356857, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10364722460508347, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07809268683195114, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06721844524145126, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06290487200021744, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06186483055353165, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039256300777196884, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03296936675906181, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03261630982160568, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02911284752190113, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028408991172909737, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021295785903930664, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02120257541537285, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019923057407140732, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014808028005063534, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039256300777196884, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039256300777196884, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12843340635299683, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11917959153652191, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11546015739440918, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10384927690029144, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06022244691848755, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05650690943002701, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06920331716537476, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0635870024561882, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06129183992743492, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.053466737270355225, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05064254626631737, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0352448970079422, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030463669449090958, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02890416979789734, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02853798307478428, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01762612909078598, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014904259704053402, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014758260920643806, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013541048392653465, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013299938291311264, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009210237301886082, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009185424074530602, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008688626810908318, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006054494064301252, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0352448970079422, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0352448970079422, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11068973690271378, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10269667208194733, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09902574121952057, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08898889273405075, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.051739685237407684, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.048335134983062744, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06006072834134102, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.055424924939870834, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05268822982907295, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04596482962369919, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.043516162782907486, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030508344992995262, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026502197608351707, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024839727208018303, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024435587227344513, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01525338739156723, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012805655598640442, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012630272656679153, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011635543778538704, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01137551385909319, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00796907115727663, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007947385311126709, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007441626396030188, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005205153021961451, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.043516162782907486, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.043516162782907486, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2548447847366333, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.23709999024868011, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.23054973781108856, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20755460858345032, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12002455443143845, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1131257563829422, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13697274029254913, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12557171285152435, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12208276242017746, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10680623352527618, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10102242976427078, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06981717050075531, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06016368791460991, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05754467844963074, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.056911565363407135, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.034881412982940674, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029448049142956734, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029202904552221298, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02672516368329525, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026313209906220436, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018121760338544846, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017735349014401436, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017154112458229065, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01128379162400961, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.034881412982940674, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.034881412982940674, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2513597905635834, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22287443280220032, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21116606891155243, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1838531494140625, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11740005016326904, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10560692101716995, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14048418402671814, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12833912670612335, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12126366794109344, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0985763818025589, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09126520901918411, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07225612550973892, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06203632056713104, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05701574683189392, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05577552691102028, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03624007850885391, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0303468257188797, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02988564781844616, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0267015490680933, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02591130882501602, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019599203020334244, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020318910479545593, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017985405400395393, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014782826416194439, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03624007850885391, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03624007850885391, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20366409420967102, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.19000940024852753, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1854407638311386, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16751211881637573, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09637261927127838, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09129348397254944, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10812465101480484, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09980494529008865, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09793360531330109, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0860225260257721, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0814356580376625, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.055409207940101624, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04806720092892647, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04650380089879036, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0461285226047039, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027742352336645126, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024442942813038826, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02432742714881897, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.022403813898563385, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022176234051585197, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014886736869812012, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015565343201160431, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014374267309904099, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011200832203030586, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027742352336645126, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027742352336645126, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.26063212752342224, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2432582527399063, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23768816888332367, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21467217803001404, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12298332154750824, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11664582788944244, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1379450261592865, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12714476883411407, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12497365474700928, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10980355739593506, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10388655215501785, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07021139562129974, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06081300228834152, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05892297253012657, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05846935510635376, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03507223725318909, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03009559027850628, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0299609936773777, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027377841994166374, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02708708867430687, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018254917114973068, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017895758152008057, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017610227689146996, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011358733288943768, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03507223725318909, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03507223725318909, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.28186970949172974, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2514601945877075, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23964442312717438, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21059124171733856, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13139057159423828, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11901888251304626, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15471935272216797, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14178593456745148, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13552142679691315, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11142498254776001, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10428144037723541, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07939645648002625, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06822960823774338, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06345245987176895, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06230441480875015, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039981674402952194, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03333534672856331, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.0329241082072258, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029399868100881577, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028608931228518486, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021774381399154663, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021599948406219482, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02022562175989151, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015139728784561157, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039981674402952194, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.039981674402952194, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.14533348381519318, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1343228816986084, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12961506843566895, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11611577868461609, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06828834116458893, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06367374211549759, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07905276864767075, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07272844761610031, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06956104189157486, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06023985892534256, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05694977939128876, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040275149047374725, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03488029167056084, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03285742551088333, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03236568346619606, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020153328776359558, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01705748960375786, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016855595633387566, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015434696339070797, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015125436708331108, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010610049590468407, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010708097368478775, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009935016743838787, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.007239410653710365, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040275149047374725, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040275149047374725, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1278400421142578, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11821479350328445, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11363649368286133, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10183320194482803, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.059994760900735855, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0557343065738678, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07004297524690628, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06455785036087036, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06111753731966019, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05299121513962746, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05016229301691055, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03562203422188759, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030920589342713356, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028863517567515373, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02835802733898163, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017827488481998444, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014957279898226261, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014737799763679504, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01354618277400732, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013223782181739807, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009357649832963943, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009410140104591846, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008697720244526863, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006275546737015247, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03562203422188759, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03562203422188759, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2714567184448242, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2518540024757385, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24480678141117096, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21985487639904022, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12800893187522888, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1203499510884285, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1456676572561264, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13379047811031342, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13024677336215973, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11335336416959763, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10695517808198929, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07427380979061127, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06405803561210632, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06131162866950035, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.060676008462905884, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03708110377192497, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03123103827238083, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03098575957119465, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.028182655572891235, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02776009775698185, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019125737249851227, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01858372613787651, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018113182857632637, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011526692658662796, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03708110377192497, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03708110377192497, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12571276724338531, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11800248175859451, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11537685245275497, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10496222972869873, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05894079431891441, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.056079719215631485, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06656705588102341, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06071636080741882, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0597560852766037, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05295152962207794, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05044987052679062, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03443087637424469, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02981046587228775, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02899746038019657, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028807707130908966, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017648890614509583, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016091644763946533, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016020726412534714, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014979152008891106, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01487061195075512, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010317395441234112, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.011234392412006855, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010090801864862442, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00912693701684475, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03443087637424469, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03443087637424469, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19410252571105957, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18097899854183197, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17663811147212982, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.159383162856102, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09196099638938904, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08709396421909332, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10313409566879272, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09510718286037445, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09347444772720337, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08193907141685486, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07751712948083878, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05288771539926529, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04576466232538223, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044336799532175064, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04400285333395004, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026470692828297615, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02321613021194935, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023121049627661705, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021228261291980743, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021023372188210487, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014205782674252987, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014651885256171227, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013742796145379543, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01041580643504858, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044336799532175064, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044336799532175064, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2710660398006439, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.25279659032821655, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24701082706451416, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2228633314371109, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12811186909675598, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12138935923576355, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1436414122581482, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1322534680366516, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.13016590476036072, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1141522079706192, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1079784631729126, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07334481924772263, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06331906467676163, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06142275035381317, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.060972440987825394, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03663383424282074, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03142871335148811, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.031295131891965866, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028553439304232597, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.028267035260796547, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01922472193837166, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01875261962413788, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.018575364723801613, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012024130672216415, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03663383424282074, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03663383424282074, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2632233500480652, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.23959185183048248, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23124021291732788, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20459717512130737, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12340283393859863, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11441867053508759, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.14150720834732056, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1302279382944107, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12635575234889984, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10698554664850235, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10003917664289474, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07256612926721573, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06273968517780304, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05961241200566292, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05885963886976242, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03648371249437332, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.031389132142066956, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.031136753037571907, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.028166912496089935, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.027662822976708412, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.019871145486831665, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.020206626504659653, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.018883492797613144, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01445062831044197, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03648371249437332, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03648371249437332, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12564100325107574, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11625276505947113, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11219556629657745, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10061006993055344, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.059087369590997696, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05515851825475693, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06835831701755524, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06292751431465149, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.060192741453647614, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.052266404032707214, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04943983629345894, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034935664385557175, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030331190675497055, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02857956290245056, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02816050685942173, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017537884414196014, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015090477652847767, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014920688234269619, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013757435604929924, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013492312282323837, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009390517137944698, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009832443669438362, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008807256817817688, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.007080805487930775, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034935664385557175, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034935664385557175, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11507536470890045, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10653482377529144, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10263293981552124, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09203033149242401, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.054054275155067444, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.050349682569503784, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06298959255218506, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05796952545642853, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05504179373383522, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.047829125076532364, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04533321037888527, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03206973522901535, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027781421318650246, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.025998517870903015, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025567255914211273, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016056276857852936, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0134825324639678, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013296309858560562, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012235304340720177, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011960221454501152, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008458963595330715, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008484957739710808, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007880052551627159, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005695955362170935, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04533321037888527, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04533321037888527, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2613126337528229, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24272553622722626, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.23605075478553772, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2121739685535431, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12321816384792328, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1159740686416626, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1402893215417862, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12877918779850006, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12533025443553925, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10927171260118484, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10319846868515015, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07152355462312698, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.061632975935935974, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05903760343790054, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05842055380344391, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03570040315389633, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03013165108859539, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02990197390317917, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027257997542619705, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026860974729061127, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0184483639895916, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01799762435257435, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01747964322566986, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011300033889710903, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03570040315389633, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03570040315389633, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2162742167711258, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19238942861557007, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18392571806907654, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.154449462890625, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10116704553365707, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0923425480723381, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1188218966126442, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10807011276483536, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10410932451486588, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0841667652130127, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07456140220165253, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06120048463344574, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05285615101456642, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0497111938893795, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.048979997634887695, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03097034990787506, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02754759043455124, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02731131948530674, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024467505514621735, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024002695456147194, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017647333443164825, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019485045224428177, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016704702749848366, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015589170157909393, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03097034990787506, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03097034990787506, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20251396298408508, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18909387290477753, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18466037511825562, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16677743196487427, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09599891304969788, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09102929383516312, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1076684221625328, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09928906708955765, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09755905717611313, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08567503094673157, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08114601671695709, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05522064119577408, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04777572304010391, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.046304233372211456, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04595186933875084, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027656298130750656, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02424597181379795, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024139592424035072, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.022182388231158257, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02196681499481201, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014867434278130531, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015293882228434086, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014388371258974075, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010866407305002213, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027656298130750656, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027656298130750656, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2720324695110321, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.25406694412231445, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24828335642814636, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2243126928806305, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12863117456436157, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12201424688100815, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.14402459561824799, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1327609121799469, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.13063901662826538, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11477908492088318, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10860442370176315, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07338472455739975, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06351848691701889, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06163792684674263, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06119312345981598, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03665274381637573, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03150258585810661, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03136980161070824, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028657106682658195, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02836001291871071, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.019128013402223587, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01873255893588066, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.018480023369193077, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011928358115255833, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03665274381637573, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03665274381637573, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27903997898101807, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25289759039878845, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.24331629276275635, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21550831198692322, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13097330927848816, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12068003416061401, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15134094655513763, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13907386362552643, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1343049854040146, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11313629150390625, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10591895878314972, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07774701714515686, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0672082006931305, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06348303705453873, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.0625811219215393, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03915166109800339, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03377028554677963, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03347291052341461, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03032776527106762, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.029744185507297516, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021557558327913284, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022266460582613945, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02038595639169216, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.016387872397899628, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03915166109800339, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03915166109800339, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11992932111024857, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1110892966389656, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1072903648018837, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09624949842691422, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05631092190742493, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05261937156319618, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0651027113199234, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05996239185333252, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.057321835309267044, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04984856769442558, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04714148864150047, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.033210646361112595, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02877887897193432, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02712412364780903, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026728352531790733, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016621530055999756, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014142132364213467, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013972142711281776, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012857768684625626, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01260237954556942, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008780953474342823, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008966522291302681, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008224454708397388, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006179229822009802, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.033210646361112595, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.033210646361112595, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1075560674071312, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09964851289987564, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09574156254529953, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08594232052564621, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05034970864653587, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.046832576394081116, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05890461057424545, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05430217087268829, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05125805363059044, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.044590651988983154, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04223445802927017, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02994614653289318, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026002658531069756, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024223804473876953, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02378162555396557, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014995409175753593, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012569215148687363, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012374654412269592, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011421030387282372, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011141639202833176, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007916907779872417, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007951061241328716, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007349556777626276, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005338067654520273, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.044590651988983154, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.044590651988983154, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25283533334732056, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.23498445749282837, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22819223999977112, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20510272681713104, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11905410885810852, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11199037730693817, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13605065643787384, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12488826364278793, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12107446789741516, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1056567132472992, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09976650774478912, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06930623203516006, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.059750452637672424, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.057064708322286606, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05642355978488922, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03459170460700989, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02919309027493, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028939584270119667, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02643411047756672, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026014160364866257, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01791275478899479, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017560428008437157, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016914743930101395, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011126461438834667, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03459170460700989, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03459170460700989, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24217478930950165, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21076232194900513, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19974574446678162, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16923904418945312, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11300687491893768, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10060185939073563, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1324206441640854, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12131691724061966, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11678646504878998, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09200457483530045, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08282230794429779, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06864666193723679, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05953757092356682, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05575591325759888, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05485903099179268, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03491703048348427, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.031034842133522034, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030758364126086235, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027236543595790863, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02667067013680935, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02007140778005123, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.022142071276903152, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018976880237460136, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.017819426953792572, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03491703048348427, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03491703048348427, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20160402357578278, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18830056488513947, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18386784195899963, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16623085737228394, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09552539885044098, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0905715748667717, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10707687586545944, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09885016083717346, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0970267802476883, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08532997965812683, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08081849664449692, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05485118180513382, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.047590386122465134, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04609142243862152, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04573408141732216, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027455147355794907, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02423008531332016, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02412196807563305, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.022223256528377533, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022010043263435364, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014754894189536572, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01541927084326744, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01425925362855196, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011104199104011059, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04573408141732216, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04573408141732216, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.26656535267829895, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24920634925365448, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24354130029678345, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2201126217842102, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1259000301361084, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11947520077228546, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1409200131893158, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1299753338098526, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12785615026950836, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11249642819166183, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10642518103122711, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07178270071744919, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.062155831605196, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.060310181230306625, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0598691888153553, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03581732511520386, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030791737139225006, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.030657708644866943, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028026169165968895, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027740729972720146, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018636487424373627, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018271224573254585, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.018000440672039986, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011572378687560558, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03581732511520386, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03581732511520386, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2879636287689209, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25982895493507385, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.24948622286319733, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.22026985883712769, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13479194045066833, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12380015850067139, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1560364067554474, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14362874627113342, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.138502299785614, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11584702879190445, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10841323435306549, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08013841509819031, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06934399902820587, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06533166021108627, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06435631960630417, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04039807990193367, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.034714456647634506, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03438776731491089, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.031051358208060265, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.030405351892113686, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022244617342948914, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022871920838952065, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.021028034389019012, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.016748299822211266, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04039807990193367, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04039807990193367, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11624385416507721, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10777978599071503, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10393315553665161, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09333434700965881, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05450771003961563, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.050877779722213745, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06330156326293945, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05827045813202858, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05545128136873245, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.048292454332113266, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04572565481066704, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032259803265333176, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027923058718442917, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026204753667116165, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025786204263567924, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016132580116391182, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013592643663287163, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013414395973086357, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012348872609436512, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012087027542293072, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008477600291371346, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008547400124371052, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007888398133218288, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00575184216722846, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04572565481066704, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04572565481066704, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10441260784864426, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09684613347053528, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09297812730073929, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0835244357585907, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04886612668633461, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04542787745594978, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0572795495390892, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.052933309227228165, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04971843957901001, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04336503520607948, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0411132387816906, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.029108285903930664, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025310220196843147, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02349023148417473, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.023044319823384285, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0145573066547513, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012167333625257015, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011965068988502026, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011062497273087502, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010778377763926983, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007647363469004631, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007673900108784437, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007067074533551931, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0051011983305215836, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04336503520607948, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04336503520607948, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2457139641046524, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22851349413394928, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22179453074932098, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19948109984397888, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11569095402956009, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10882875323295593, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13269145786762238, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12146782130002975, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11760702729225159, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10274878889322281, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09712395071983337, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06757323443889618, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0581907257437706, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05548830330371857, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054832153022289276, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03376023843884468, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028421476483345032, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028159040957689285, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025763627141714096, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02533891052007675, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017535893246531487, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017188508063554764, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016517436131834984, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010977455414831638, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03376023843884468, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03376023843884468, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.227357879281044, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19827136397361755, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18306434154510498, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15558651089668274, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10569026321172714, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09203947335481644, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1344061642885208, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12159781157970428, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10981729626655579, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08693225681781769, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08051024377346039, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06901825219392776, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05932378023862839, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.052029795944690704, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.050185080617666245, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03479699417948723, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02885105274617672, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028103655204176903, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0254585649818182, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02431931160390377, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019385335966944695, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.021022606641054153, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01709415204823017, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01636570505797863, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03479699417948723, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03479699417948723, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20082426071166992, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18762223422527313, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1831779181957245, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16559942066669464, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09513044357299805, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09015662223100662, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10676935315132141, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09855253994464874, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0965980514883995, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08502758294343948, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0805182009935379, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05475037544965744, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04748194292187691, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04593826085329056, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04557296633720398, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02740289643406868, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02419687807559967, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024084242060780525, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02221781387925148, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02200394496321678, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01476195640861988, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015483666211366653, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014259782619774342, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011230280622839928, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04557296633720398, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04557296633720398, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2593783438205719, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.242426335811615, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2368714064359665, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21415308117866516, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12239422649145126, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11615996807813644, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13715921342372894, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12640716135501862, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12426738440990448, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10942452400922775, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10346219688653946, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06975246220827103, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06045258417725563, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05863570049405098, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05821072310209274, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03479638323187828, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02996395342051983, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0298314206302166, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027290193364024162, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.0270095057785511, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018037281930446625, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017825135961174965, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01739426702260971, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011345918290317059, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03479638323187828, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03479638323187828, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.28728991746902466, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2587205469608307, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.24794931709766388, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21866637468338013, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13465017080307007, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12325919419527054, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15660643577575684, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1439397931098938, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1384606808423996, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11537331342697144, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10784105956554413, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.0805281475186348, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06949608027935028, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06521575152873993, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06419497728347778, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04057661071419716, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03459539636969566, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03424885869026184, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030848857015371323, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.030172906816005707, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022349199280142784, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02274683490395546, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.021037133410573006, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.016553055495023727, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04057661071419716, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04057661071419716, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11940331757068634, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11072421073913574, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10691055655479431, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09604570269584656, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055917177349328995, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05229373276233673, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06462042033672333, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05954250320792198, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05689910054206848, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04957889765501022, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04689543694257736, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03287162259221077, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028491605073213577, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026861365884542465, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02646011859178543, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016437815502285957, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013866855762898922, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013699731789529324, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01258777268230915, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012334833852946758, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008604379370808601, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008610802702605724, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00805344246327877, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005696502979844809, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03287162259221077, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03287162259221077, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10549803823232651, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09792914986610413, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09414847195148468, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08461437374353409, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04933221638202667, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04593370854854584, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05761384963989258, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05321972817182541, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05018562823534012, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04383279010653496, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04154160991311073, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02922944538295269, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025452548637986183, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02368209883570671, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0232540350407362, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01461759302765131, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012230867519974709, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01204163022339344, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01111659687012434, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010848753154277802, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007655893452465534, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007648889906704426, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007104986347258091, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005025732330977917, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04383279010653496, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04383279010653496, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24247939884662628, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2253735065460205, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21879953145980835, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19679778814315796, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11410051584243774, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10726908594369888, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13070522248744965, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11984957009553909, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11603137105703354, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10132849216461182, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09583926945924759, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06660942733287811, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.057458870112895966, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05473043769598007, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05406159907579422, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03330804035067558, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028080858290195465, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027813630178570747, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025467509403824806, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02503826655447483, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01735059916973114, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017060833051800728, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01634465903043747, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010982324369251728, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03330804035067558, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03330804035067558, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24272945523262024, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2101280838251114, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19901137053966522, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.17225557565689087, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11264068633317947, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10013517737388611, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13227783143520355, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12102645635604858, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11670856922864914, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09223277866840363, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08543265610933304, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06808227300643921, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05902678519487381, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.055195532739162445, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05427679792046547, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03435872495174408, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03033813275396824, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03008090890944004, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02662719413638115, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026047972962260246, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019329316914081573, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02126365713775158, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018191371113061905, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.016766637563705444, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03435872495174408, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03435872495174408, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2014009952545166, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18823248147964478, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18381519615650177, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16615986824035645, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09529731422662735, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09039944410324097, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10683084279298782, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09866514801979065, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09677290916442871, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0852154791355133, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08071567118167877, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.054741255939006805, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04752539098262787, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.046005729585886, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04563979059457779, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02738960087299347, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024245429784059525, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024133890867233276, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02226944826543331, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022048281505703926, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01472028624266386, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01552134845405817, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.0142162861302495, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011259326711297035, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04563979059457779, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04563979059457779, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25538066029548645, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23871363699436188, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23334482312202454, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21098671853542328, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12045331299304962, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11433680355548859, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13492842018604279, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12434353679418564, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12229573726654053, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1077185720205307, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10192757844924927, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0686621144413948, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05946260690689087, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05766617879271507, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.057250794023275375, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03426649421453476, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02942620776593685, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02929190918803215, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02679641917347908, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026522526517510414, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017773255705833435, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017434265464544296, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.0171396192163229, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011003021150827408, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03426649421453476, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03426649421453476, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.29069751501083374, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2609620988368988, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.25001025199890137, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.22013503313064575, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13621319830417633, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12442479282617569, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15846851468086243, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14543618261814117, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14012779295444489, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.1162349060177803, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10868681222200394, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08160294592380524, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07026839256286621, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06598694622516632, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06497698277235031, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.041133299469947815, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03504582494497299, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03470863774418831, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03117598406970501, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.030497826635837555, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022773567587137222, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.0230825487524271, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.021468834951519966, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01684817485511303, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.041133299469947815, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.041133299469947815, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1281142383813858, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11888407170772552, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11487141996622086, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10329525917768478, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.060033880174160004, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05620015412569046, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06914647668600082, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06375636160373688, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06107795983552933, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.053285013884305954, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05044570565223694, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03525140509009361, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03058778867125511, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02888304367661476, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028469212353229523, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01764695905148983, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015034287236630917, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014862360432744026, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013691655360162258, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013434400781989098, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009309149347245693, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00950504932552576, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008741249330341816, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006526019424200058, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03525140509009361, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03525140509009361, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10803258419036865, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1002718135714531, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09622379392385483, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08649671822786331, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05044887214899063, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04689221829175949, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05915489047765732, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05471904203295708, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05133642256259918, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04484393075108528, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.042531177401542664, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030029311776161194, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026153089478611946, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024242080748081207, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02377246879041195, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01503120269626379, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012549269013106823, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012336988002061844, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011421004310250282, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011122297495603561, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007898437790572643, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007915941998362541, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007305772975087166, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005244558211416006, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04484393075108528, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04484393075108528, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24623247981071472, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2290034145116806, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22265343368053436, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20041610300540924, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1158493384718895, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10903974622488022, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13229316473007202, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1212715283036232, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11775025725364685, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10303819924592972, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09739157557487488, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06745084375143051, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05813205987215042, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05552483722567558, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05488615110516548, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033715639263391495, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028462039306759834, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028212351724505424, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02583152987062931, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02542026899755001, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01750990003347397, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017202608287334442, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016531798988580704, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011016507633030415, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033715639263391495, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.033715639263391495, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24501214921474457, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2121783345937729, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2009260654449463, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1702803522348404, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11420971155166626, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10078588128089905, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13405117392539978, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12250036001205444, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11820793896913528, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09261596947908401, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08274225145578384, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06925676017999649, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05948575958609581, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0556887723505497, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054784998297691345, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03495435416698456, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029993493109941483, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029717974364757538, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02590596303343773, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025315802544355392, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019534118473529816, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020290380343794823, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018416592851281166, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015325440093874931, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03495435416698456, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03495435416698456, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1961119920015335, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18336035311222076, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17909817397594452, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16187360882759094, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09274517744779587, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08800048381090164, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10394128412008286, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09601099789142609, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09421082586050034, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08302269130945206, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07863100618124008, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0532580204308033, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.046239130198955536, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044776950031518936, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04443245381116867, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02666235715150833, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023571517318487167, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02346300706267357, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021661072969436646, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021446293219923973, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014347318559885025, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015055492520332336, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013869469985365868, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010891085490584373, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044776950031518936, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044776950031518936, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2531270384788513, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23675142228603363, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23149816691875458, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2093423753976822, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11928395926952362, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11333838850259781, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1334531158208847, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12310107797384262, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12113149464130402, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10678408294916153, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1010790765285492, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06792822480201721, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05883752927184105, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.057096973061561584, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05668499693274498, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03388955444097519, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029115179553627968, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.028990967199206352, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026519764214754105, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02625659853219986, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017591549083590508, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017207292839884758, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016985634341835976, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010815268382430077, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03388955444097519, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03388955444097519, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.28574028611183167, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25541117787361145, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2438230961561203, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21444867551326752, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13336120545864105, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12122897803783417, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15662449598312378, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14355172216892242, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13750477135181427, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11343075335025787, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10610082745552063, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.0802820548415184, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0692075863480568, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06459879875183105, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.0634777843952179, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04031484201550484, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.034293729811906815, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03391505777835846, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030429290607571602, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02968200296163559, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021987032145261765, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022636180743575096, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020506130531430244, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.016440946608781815, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04031484201550484, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04031484201550484, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1314574033021927, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12219972163438797, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1183691918849945, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10651171952486038, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06153882294893265, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05776650458574295, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07065117359161377, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06509996205568314, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06259077042341232, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05474171042442322, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05182483792304993, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03594972938299179, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03115537203848362, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029542194679379463, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029147667810320854, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017973927780985832, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015258894301950932, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015099097974598408, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013887185603380203, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013641619123518467, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009399078786373138, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009449725970625877, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008846882730722427, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006286164280027151, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03594972938299179, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03594972938299179, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11062965542078018, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10284196585416794, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09903042018413544, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08913137763738632, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05164533108472824, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.048196300864219666, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0600861981511116, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05556894466280937, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05251019448041916, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04598560929298401, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0435759536921978, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030472641810774803, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026528585702180862, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02476426213979721, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02433098293840885, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015228736214339733, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01275451947003603, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01255982369184494, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011610280722379684, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011334072798490524, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007944995537400246, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007928953506052494, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007395274005830288, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00515591399744153, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0435759536921978, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0435759536921978, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24210722744464874, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22560276091098785, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21914975345134735, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19752107560634613, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11389520019292831, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10723526775836945, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13037919998168945, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1196175366640091, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11577939242124557, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10156668722629547, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09607820212841034, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06640066206455231, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05733732134103775, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05460892245173454, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.053946517407894135, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03319840878248215, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027971770614385605, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02770307846367359, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025428397580981255, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0250081904232502, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017292002215981483, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01693149283528328, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016315188258886337, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01081354171037674, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03319840878248215, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03319840878248215, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25129637122154236, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22680233418941498, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21883007884025574, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18929502367973328, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11760571599006653, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1080017238855362, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13550114631652832, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12369493395090103, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12054061144590378, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09968854486942291, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09117398411035538, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06986184418201447, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06037338078022003, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05758905038237572, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05688386410474777, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03550742194056511, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03146112337708473, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03128881752490997, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.028141716495156288, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.027728354558348656, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020269671455025673, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.021661462262272835, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019462479278445244, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01700720377266407, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03550742194056511, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03550742194056511, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19589659571647644, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1831466257572174, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17888504266738892, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16185899078845978, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09258083999156952, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08784319460391998, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10370936989784241, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09577970206737518, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09400533139705658, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08285161852836609, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07853298634290695, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05309023708105087, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04604944586753845, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04460623860359192, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04426427558064461, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026541568338871002, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023375922814011574, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023274200037121773, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021458452567458153, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02124498039484024, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014204904437065125, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01479230634868145, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013731662184000015, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010540099814534187, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04460623860359192, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04460623860359192, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25128310918807983, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23505327105522156, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.229848712682724, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20799435675144196, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11842001974582672, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11252644658088684, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13242948055267334, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12220022827386856, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1202695444226265, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10605521500110626, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10042385011911392, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06738241761922836, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05841720104217529, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05670848861336708, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05630064755678177, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03362385928630829, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.028927989304065704, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.028803881257772446, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026368334889411926, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026107843965291977, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017437707632780075, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.0171278715133667, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016840411350131035, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010810569860041142, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03362385928630829, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03362385928630829, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2838687300682068, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25335493683815, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.24155206978321075, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.2124609351158142, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1323622167110443, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12006174027919769, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15553976595401764, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1427755057811737, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1365479677915573, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11238313466310501, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.1051798090338707, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07976295799016953, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06874857097864151, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.0639696791768074, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06280426681041718, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04016425833106041, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.0337057001888752, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.033307865262031555, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02978678047657013, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02902352437376976, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021925417706370354, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02196761779487133, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020437952131032944, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015588085167109966, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04016425833106041, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04016425833106041, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12714320421218872, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11834868788719177, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11462344974279404, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1032552644610405, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.059566471725702286, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05598602443933487, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06837410479784012, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06298966705799103, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06056484207510948, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.053060129284858704, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.050265416502952576, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034805022180080414, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030129481106996536, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028589891269803047, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028222715482115746, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017398936673998833, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014734335243701935, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014580207876861095, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013412488624453545, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013175162486732006, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009106677025556564, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009069102816283703, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00859221164137125, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005974681582301855, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034805022180080414, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034805022180080414, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11058849841356277, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10284357517957687, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09913131594657898, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08919651061296463, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05164095014333725, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.048272665590047836, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05989100784063339, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.055400509387254715, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.052518535405397415, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04599488526582718, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04359038919210434, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030380478128790855, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026476476341485977, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024774709716439247, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024365579709410667, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015180779621005058, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012772507965564728, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012587811797857285, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011628132313489914, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011364859528839588, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007940822280943394, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00793866440653801, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0074074179865419865, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005190965719521046, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04359038919210434, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04359038919210434, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2512478530406952, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.23441985249519348, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2281036674976349, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20571257174015045, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11830709129571915, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11170552670955658, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1347680389881134, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12361644208431244, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12020684033632278, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1055581197142601, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09992408752441406, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0686824694275856, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05922858417034149, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.056697700172662735, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.056099433451890945, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0343199223279953, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029038282111287117, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02879558689892292, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02641412615776062, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026020783931016922, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01786734350025654, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017494607716798782, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01693780906498432, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011182929389178753, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0343199223279953, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0343199223279953, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24752722680568695, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21896252036094666, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20865368843078613, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18133887648582458, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1154739037156105, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1040392816066742, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1356416642665863, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12390052527189255, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11911843717098236, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09637945145368576, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08946117758750916, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06969958543777466, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05991367995738983, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05602612346410751, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05508362874388695, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03500938042998314, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029800212010741234, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02949083223938942, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02609260194003582, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025493519380688667, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01902737095952034, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01974017731845379, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01779276505112648, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014455171301960945, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03500938042998314, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03500938042998314, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1979290097951889, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18502715229988098, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1807561218738556, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16351789236068726, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09355141967535019, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08878379315137863, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10473431646823883, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09672877192497253, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09501206129789352, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08372966200113297, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07937167584896088, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.053627293556928635, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.046572163701057434, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04513274133205414, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.044804323464632034, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026843925938010216, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023729171603918076, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02362683042883873, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02179555967450142, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021588603034615517, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01442762278020382, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015113101340830326, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013962440192699432, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010898089967668056, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04513274133205414, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04513274133205414, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25585469603538513, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2392423450946808, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23400145769119263, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21177569031715393, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12059545516967773, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11457440257072449, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1349530965089798, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12440021336078644, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12246719747781754, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1079951748251915, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10225006192922592, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06861565262079239, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05947644263505936, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.057756826281547546, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05734287202358246, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03425747528672218, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02948489412665367, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02936280146241188, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026875264942646027, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02660943754017353, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017806634306907654, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017480669543147087, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01721670851111412, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011067274026572704, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03425747528672218, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03425747528672218, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2862006723880768, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25400078296661377, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2412426918745041, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21175719797611237, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13307754695415497, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11997006088495255, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15783478319644928, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14471235871315002, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1377364993095398, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11252611875534058, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10545964539051056, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08107535541057587, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06970348209142685, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06436510384082794, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06308414041996002, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.040910907089710236, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03399917483329773, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03355031833052635, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029988471418619156, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.029115306213498116, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.0225079283118248, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02234382927417755, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020844124257564545, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015892626717686653, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.040910907089710236, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.040910907089710236, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1362999677658081, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12683063745498657, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12297242879867554, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1108672246336937, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06396526843309402, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06017861142754555, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07309941947460175, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06736793369054794, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0650177076458931, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05696937069296837, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05403262749314308, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03725530952215195, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03227533400058746, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.030734194442629814, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.030356571078300476, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018644267693161964, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01590600237250328, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015760427340865135, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014494816772639751, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014268193393945694, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00979585200548172, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009872027672827244, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009285474196076393, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006644714158028364, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03725530952215195, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03725530952215195, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11678650975227356, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10856026411056519, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10482065379619598, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09434739500284195, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05458203703165054, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05108468979597092, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06303731352090836, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05830824747681618, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.055492546409368515, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.048585500568151474, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.046042993664741516, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03198539465665817, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027852090075612068, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02617718279361725, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025772346183657646, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015992052853107452, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013497438281774521, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013318674638867378, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01228378713130951, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012025929056107998, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008361385203897953, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008358973078429699, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007836878299713135, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0054826512932777405, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03198539465665817, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03198539465665817, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.26086944341659546, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24334768950939178, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2370685487985611, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21395421028137207, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12290746718645096, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11617685854434967, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13951615989208221, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12806709110736847, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12488434463739395, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10970897972583771, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10382930934429169, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07102523744106293, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.061307765543460846, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05887141823768616, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05829991400241852, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035469189286231995, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030058644711971283, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029837673529982567, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027324765920639038, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026948293671011925, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01835406944155693, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017947088927030563, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017449263483285904, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011304644867777824, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035469189286231995, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035469189286231995, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24863797426223755, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22310872375965118, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21471421420574188, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18408246338367462, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11520346999168396, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10540953278541565, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1335853934288025, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12145397812128067, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11831511557102203, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09592495113611221, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08723677694797516, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06881972402334213, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05920456722378731, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.056340545415878296, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05564199388027191, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035008952021598816, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030851421877741814, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030642827972769737, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027291525155305862, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02686934918165207, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019976260140538216, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.021316513419151306, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01912212185561657, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.016782065853476524, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035008952021598816, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035008952021598816, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1971520185470581, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1842348277568817, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17999328672885895, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1629384160041809, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09318534284830093, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08841146528720856, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10422070324420929, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09626993536949158, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09462583810091019, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0833488255739212, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07905339449644089, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05333961918950081, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04630684480071068, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04489690065383911, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.044565655291080475, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026682943105697632, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023524416610598564, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023427100852131844, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021580412983894348, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02137584052979946, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01427196804434061, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01486373320221901, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013817260041832924, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010592235252261162, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04489690065383911, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04489690065383911, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2551206350326538, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2386353313922882, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23339858651161194, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21121639013290405, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12029960006475449, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11427819728851318, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13440221548080444, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12400266528129578, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12215297669172287, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10767650604248047, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10198386013507843, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06842122972011566, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.059268854558467865, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.057578615844249725, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05717767775058746, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034170664846897125, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02938678115606308, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02926797978579998, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026778629049658775, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02651781588792801, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017797671258449554, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017398521304130554, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017228655517101288, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011003449559211731, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034170664846897125, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034170664846897125, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27686405181884766, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2475033849477768, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23635245859622955, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20774772763252258, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12878799438476562, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11710707098245621, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15070155262947083, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13850462436676025, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.132819265127182, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.1094503253698349, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.1023917868733406, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07722201943397522, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0665535032749176, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06210185959935188, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06101614609360695, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03882453963160515, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03248848393559456, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.032131049782037735, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.028668522834777832, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.027940016239881516, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021022768691182137, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02085806243121624, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01960643008351326, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01447877287864685, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03882453963160515, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03882453963160515, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1319197118282318, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12279899418354034, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11909370869398117, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10735701024532318, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.061843495815992355, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.058175839483737946, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07086659967899323, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06533065438270569, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06284675002098083, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05515052005648613, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.052350837737321854, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036077260971069336, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03124086745083332, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029676763340830803, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02929489128291607, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018040623515844345, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015292450785636902, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015136901289224625, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013937126845121384, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013694165274500847, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009422749280929565, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009406834840774536, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008902981877326965, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006190252956002951, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036077260971069336, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036077260971069336, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11352496594190598, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10556568950414658, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10181008279323578, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09168066829442978, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05304039269685745, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04958663135766983, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06148865073919296, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05687044560909271, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.053922899067401886, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.047276612371206284, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04481399804353714, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03122181072831154, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027182428166270256, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.025459975004196167, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025044217705726624, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015607958659529686, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013147828169167042, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012965511530637741, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011986851692199707, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01172404270619154, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00818733498454094, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008204641751945019, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007653748616576195, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00542064243927598, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04481399804353714, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04481399804353714, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.258899450302124, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2416844666004181, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.23555432260036469, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2127947062253952, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12198787927627563, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11541268974542618, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1382010579109192, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12689416110515594, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12387344241142273, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10892093926668167, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10317709296941757, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07032063603401184, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06070500984787941, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05842795595526695, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.057878948748111725, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03513399139046669, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02981668710708618, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029623134061694145, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027120888233184814, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02677251026034355, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018156511709094048, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01777030900120735, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01729213260114193, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011202700436115265, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03513399139046669, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03513399139046669, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24979853630065918, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2178414911031723, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2070997655391693, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1692061871290207, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11653687059879303, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10438288748264313, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13552875816822052, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1240386962890625, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12021464109420776, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09283405542373657, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08250880241394043, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06973898410797119, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06013908609747887, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05675980821251869, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05591845139861107, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03512093797326088, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03052782453596592, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03028864599764347, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.026054715737700462, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025511790066957474, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019454604014754295, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020562492311000824, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018427900969982147, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015565234236419201, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03512093797326088, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03512093797326088, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.206704780459404, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.19324788451194763, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18893088400363922, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.17098499834537506, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09762226045131683, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09266859292984009, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10914140939712524, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10082046687602997, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09915325790643692, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08736690878868103, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08280870318412781, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.055735811591148376, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04840289056301117, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.046963248401880264, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.046616747975349426, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027884196490049362, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024449462071061134, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024348950013518333, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.022396039217710495, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.0221848264336586, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014821081422269344, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015222674235701561, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01434625219553709, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010596385225653648, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027884196490049362, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027884196490049362, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2631421387195587, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24618007242679596, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24077904224395752, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.218018040060997, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1240932047367096, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11793544143438339, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13871046900749207, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1279136687517166, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12598834931850433, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11113759875297546, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10523717850446701, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07057178020477295, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06113668158650398, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05940913036465645, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.059004396200180054, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035267990082502365, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03033454902470112, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.030213601887226105, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027657782658934593, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02739512175321579, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018431682139635086, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017979806289076805, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017855260521173477, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011404032818973064, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035267990082502365, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035267990082502365, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27975407242774963, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.24924588203430176, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2372492104768753, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20838899910449982, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13011083006858826, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11767798662185669, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1530168056488037, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1407560110092163, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13431207835674286, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11029936373233795, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10324792563915253, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07830125838518143, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06780925393104553, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06284265220165253, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.061625055968761444, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03929942101240158, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03302978724241257, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03258303552865982, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029132556170225143, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028308779001235962, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.020979490131139755, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021463189274072647, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01931951940059662, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015034941025078297, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03929942101240158, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03929942101240158, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13426189124584198, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12514814734458923, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12140435725450516, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10958357900381088, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06301514059305191, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05932983011007309, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07205335050821304, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06640882790088654, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06404747813940048, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05627582594752312, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05345053970813751, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03669784590601921, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03180776536464691, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.030241038650274277, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029864437878131866, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01835785247385502, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015588882379233837, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015433870255947113, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014224587939679623, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013984951190650463, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009611209854483604, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009599708020687103, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009102345444262028, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006332010962069035, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03669784590601921, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03669784590601921, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11788023263216019, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10979058593511581, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10610659420490265, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09571173042058945, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05518076941370964, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05175139755010605, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06362941861152649, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0587843656539917, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0561014786362648, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04925267770886421, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04672044888138771, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032314151525497437, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028102323412895203, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026459448039531708, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02606380544602871, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016154278069734573, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013615199364721775, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013443851843476295, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012411137111485004, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012161120772361755, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008437083102762699, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008389033377170563, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007917662151157856, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005453115329146385, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032314151525497437, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032314151525497437, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2639636695384979, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24655018746852875, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24062398076057434, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21744120121002197, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12436996400356293, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.117833212018013, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14041192829608917, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12903159856796265, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12631887197494507, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11116326600313187, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10535363107919693, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07149085402488708, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06169149652123451, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.059531696140766144, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05900181084871292, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035667285323143005, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030292412266135216, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03010905347764492, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02755095437169075, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.027211707085371017, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01837899722158909, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017886390909552574, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017550664022564888, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011086022481322289, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035667285323143005, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.035667285323143005, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2577858567237854, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2286318689584732, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21813759207725525, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18298938870429993, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12102710455656052, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1098305881023407, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14092324674129486, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12908343970775604, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12422436475753784, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09897547960281372, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08899339288473129, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0729733482003212, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0629495233297348, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.059331219643354416, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05846976116299629, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037046585232019424, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03250275179743767, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03221310302615166, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02843446098268032, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02787843346595764, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.021170534193515778, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0225857961922884, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020071357488632202, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0177413709461689, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037046585232019424, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037046585232019424, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21187132596969604, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.19812031090259552, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.19372259080410004, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.17531011998653412, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10010495781898499, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.095082588493824, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1117357537150383, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10328927636146545, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10165578871965408, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08963074535131454, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08491124212741852, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05709297955036163, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.049587104469537735, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.048128001391887665, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04778771474957466, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028515147045254707, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02502571977674961, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02492380142211914, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02292904071509838, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022712115198373795, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015075625851750374, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015530254691839218, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014591960236430168, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010760035365819931, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028515147045254707, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028515147045254707, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.26782625913619995, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2505885362625122, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24520055949687958, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.22202230989933014, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1263064444065094, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1200738400220871, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1407870501279831, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13010460138320923, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12823475897312164, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11313624680042267, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10715683549642563, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0716402605175972, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06218406930565834, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06044226139783859, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06002912297844887, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03573308140039444, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030801616609096527, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.030681481584906578, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02807501144707203, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02780822478234768, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01848999410867691, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01815728470683098, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01790151745080948, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01138268318027258, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03573308140039444, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03573308140039444, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.28190767765045166, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25095924735069275, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23840630054473877, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20962116122245789, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13098832964897156, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11820541322231293, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15493319928646088, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14246684312820435, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1353107988834381, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11107229441404343, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10412832349538803, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07913152128458023, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0684589371085167, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06323467195034027, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.061953868716955185, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03962114825844765, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.033158112317323685, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03269793093204498, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029231403023004532, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028372367843985558, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021125422790646553, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021497303619980812, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019417723640799522, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014928392134606838, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03962114825844765, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03962114825844765, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13029593229293823, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1213420107960701, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11756525188684464, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10608077049255371, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06105375289916992, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05741795524954796, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07009094208478928, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0645708292722702, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06208840385079384, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.054481085389852524, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05174975469708443, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035707026720047, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030934173613786697, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029326181858778, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028952866792678833, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017856426537036896, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0151745630428195, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015014329925179482, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013843989931046963, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013601738028228283, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009376691654324532, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009424513205885887, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008861493319272995, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006308678071945906, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035707026720047, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035707026720047, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11651644855737686, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10837037861347198, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10454829782247543, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09425000101327896, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.054453521966934204, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.050957679748535156, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0631614699959755, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05836666747927666, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.055372703820466995, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04856710135936737, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04609472304582596, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03206012770533562, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027898207306861877, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026135286316275597, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02570357918739319, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016030726954340935, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013474451377987862, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013284722343087196, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01228178571909666, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012012162245810032, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008400451391935349, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008370805531740189, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007854507304728031, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005478616803884506, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03206012770533562, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03206012770533562, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2643027603626251, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24669934809207916, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24062974750995636, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21742384135723114, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12451958656311035, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1178700402379036, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1408436894416809, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12937510013580322, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12647190690040588, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11122571676969528, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10547605901956558, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07168892025947571, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06183846667408943, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.059604890644550323, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.059081725776195526, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03576931729912758, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030376333743333817, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030190087854862213, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027636639773845673, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.027294650673866272, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018426956608891487, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018014386296272278, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01756836473941803, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011262952350080013, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03576931729912758, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03576931729912758, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2528531551361084, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2180694043636322, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20629288256168365, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16914257407188416, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11763690412044525, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10414820164442062, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13823015987873077, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12626494467258453, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1218237429857254, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09376881271600723, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08294126391410828, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07144978642463684, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.061406612396240234, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.057456616312265396, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05651732534170151, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03624419495463371, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03127112239599228, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.031006455421447754, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.026853231713175774, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02624710276722908, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02051515132188797, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.021555479615926743, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01932605169713497, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01668948493897915, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03624419495463371, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03624419495463371, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22010204195976257, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20587779581546783, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20138803124427795, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18232154846191406, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10398519039154053, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09880496561527252, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11626532673835754, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10725107789039612, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10561239719390869, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09311516582965851, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08829467743635178, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05936376750469208, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.051528993993997574, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.050052184611558914, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04969946667551994, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02974017895758152, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0261205043643713, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.026014236733317375, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023953882977366447, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023731496185064316, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01591537334024906, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016342513263225555, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015437117777764797, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011484519578516483, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02974017895758152, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02974017895758152, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.27063417434692383, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2532981336116791, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2478492110967636, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.22438398003578186, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12760670483112335, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12133985012769699, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.14235174655914307, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13142764568328857, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1295671910047531, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11431946605443954, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10826463997364044, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07244370132684708, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06286408752202988, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.061122529208660126, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.060703396797180176, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.036219581961631775, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03120405413210392, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.031088659539818764, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028456242755055428, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.028194665908813477, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018922042101621628, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018497321754693985, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01833997666835785, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011740385554730892, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.036219581961631775, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.036219581961631775, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2822859585285187, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2504170536994934, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2374359369277954, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20884332060813904, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1309930980205536, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11790108680725098, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15530435740947723, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14303746819496155, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13553740084171295, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11090662330389023, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10400957614183426, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07948941737413406, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06868314743041992, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06320623308420181, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06185702979564667, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03978516533970833, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03304201364517212, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03255368769168854, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029057195410132408, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02815273404121399, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021214602515101433, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02133331634104252, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01943817175924778, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014635425060987473, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03978516533970833, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03978516533970833, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13335910439491272, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12419306486845016, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12034080177545547, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1085379347205162, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06255550682544708, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05878470838069916, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07189413160085678, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06624490022659302, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06359412521123886, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05581917613744736, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05303462967276573, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036604732275009155, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.031686216592788696, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.030012769624590874, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02960900403559208, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018312348052859306, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015473895706236362, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015307656489312649, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014109332114458084, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013854019343852997, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009592177346348763, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00954244751483202, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009053213521838188, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006283304654061794, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036604732275009155, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036604732275009155, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11876879632472992, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11049824208021164, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10654829442501068, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09608815610408783, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055593281984329224, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05201190337538719, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06460068374872208, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.059630777686834335, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05654758960008621, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.049609649926424026, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04714705049991608, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03281334415078163, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028498126193881035, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026685237884521484, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026247885078191757, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016415903344750404, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013759839348495007, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013565057888627052, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012543289922177792, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012265696190297604, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008596432395279408, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008543499745428562, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008030456490814686, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005587844643741846, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03281334415078163, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03281334415078163, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.26219624280929565, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24473059177398682, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.23846733570098877, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21542495489120483, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12353324145078659, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11680691689252853, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14026448130607605, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1286403387784958, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12549260258674622, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1102912425994873, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10464032739400864, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07142400741577148, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0614977590739727, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05915270745754242, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05858778953552246, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03565042465925217, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03016469068825245, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029964828863739967, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027444329112768173, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02707831934094429, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018416432663798332, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01795133762061596, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017528634518384933, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011253123171627522, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03565042465925217, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03565042465925217, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23850302398204803, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.20857979357242584, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19850577414035797, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16934283077716827, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11106687784194946, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09928172826766968, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1296282708644867, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11889662593603134, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11501993983983994, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09126124531030655, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08393870294094086, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0670071467757225, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05770023167133331, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.054171208292245865, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05330676957964897, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03389802947640419, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029302485287189484, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029058212414383888, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02557334303855896, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025023579597473145, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019003579393029213, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019969919696450233, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017914295196533203, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015262065455317497, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03389802947640419, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03389802947640419, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20519126951694489, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.19193707406520844, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1876777559518814, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16999688744544983, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09698058664798737, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09212884306907654, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1084018126130104, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10009387880563736, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09850172698497772, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08686108887195587, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08239234983921051, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05538720637559891, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04802761971950531, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04662300646305084, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04628894478082657, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027678286656737328, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02421003207564354, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024117877706885338, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02218121849000454, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021973801776766777, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014595734886825085, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014980918727815151, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01413149107247591, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010321686044335365, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027678286656737328, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027678286656737328, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2671338617801666, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2500186562538147, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24460725486278534, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.22163766622543335, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12607528269290924, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11985227465629578, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.14087235927581787, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12990699708461761, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12801478803157806, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11299958825111389, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10710113495588303, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07170141488313675, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06210670992732048, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.060365255922079086, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.059950172901153564, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03579472005367279, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030814118683338165, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.030689015984535217, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028107183054089546, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027847547084093094, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018535800278186798, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018245253711938858, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017931219190359116, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011553844437003136, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03579472005367279, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03579472005367279, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2813095152378082, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25130918622016907, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23911409080028534, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21155555546283722, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1310134381055832, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11858301609754562, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15552455186843872, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14236140251159668, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1352984607219696, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11194373667240143, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10535866022109985, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07961084693670273, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06856837123632431, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06337488442659378, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06211524084210396, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04002734273672104, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03343495726585388, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.0329752117395401, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02970246970653534, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02885567769408226, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02165757305920124, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02191356010735035, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019970014691352844, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01551918126642704, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04002734273672104, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04002734273672104, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12618137896060944, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1175113394856453, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11389200389385223, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10273676365613937, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05921502411365509, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05566095560789108, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06803777068853378, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06267929822206497, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06020774692296982, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.052857961505651474, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05025433003902435, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034673482179641724, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030027644708752632, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02844698540866375, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02806909754872322, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01736852154135704, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014695849269628525, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0145365996286273, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01341029442846775, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013166983611881733, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009134712629020214, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009102296084165573, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008633827790617943, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006053337827324867, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034673482179641724, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034673482179641724, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11216223984956741, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10438269376754761, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10048942267894745, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09067189693450928, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05249142646789551, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04898989200592041, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06128208711743355, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05657286196947098, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05335453152656555, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04682810604572296, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04453745484352112, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.031119564548134804, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02704356424510479, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0252035241574049, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024753589183092117, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015585862100124359, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013026122003793716, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012826886028051376, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01189242023974657, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011610946618020535, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008196542970836163, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008160502649843693, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007634428329765797, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005396494176238775, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04453745484352112, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04453745484352112, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2702806890010834, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2522907853126526, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24620182812213898, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.22255530953407288, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12740178406238556, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12061470001935959, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14409510791301727, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13217148184776306, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12938158214092255, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11382678896188736, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10795621573925018, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07340840995311737, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06319117546081543, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.060975123196840286, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.060451850295066833, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03662123531103134, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.031035467982292175, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030856825411319733, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.028229625895619392, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02788158506155014, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018852008506655693, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018339896574616432, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018004192039370537, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011382268741726875, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03662123531103134, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03662123531103134, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2269008457660675, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19373255968093872, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18192176520824432, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15428206324577332, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10327466577291489, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09083875268697739, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12398399412631989, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1137099415063858, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1085531935095787, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08436188101768494, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07716928422451019, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06408258527517319, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.055532172322273254, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05086778849363327, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04969990998506546, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0324445404112339, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028198622167110443, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02789231576025486, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024766722694039345, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024058334529399872, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01841474138200283, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020190363749861717, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017036525532603264, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.016024550423026085, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0324445404112339, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0324445404112339, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20275282859802246, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18979710340499878, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18560265004634857, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16810013353824615, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09579307585954666, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09103205054998398, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10699309408664703, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09877502918243408, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09726617485284805, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08581206202507019, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08137675374746323, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.054636985063552856, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04737219959497452, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04600363224744797, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04568614810705185, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0273091159760952, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02380799688398838, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023715436458587646, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0218012984842062, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021598270162940025, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014369657263159752, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014605466276407242, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013924508355557919, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009926814585924149, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04568614810705185, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04568614810705185, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2655651867389679, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24866849184036255, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2433503419160843, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2205217331647873, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12532411515712738, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11917754262685776, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13985438644886017, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.129047691822052, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1271757185459137, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11232485622167587, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1064835712313652, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07116374373435974, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0616728775203228, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05996788665652275, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05956230312585831, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03551274538040161, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030549684539437294, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.030431557446718216, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02786550112068653, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02760835736989975, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01836777664721012, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01800544746220112, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017792988568544388, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011281660757958889, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03551274538040161, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03551274538040161, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27657806873321533, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.24721580743789673, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23501068353652954, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20842793583869934, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12856541574001312, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11642781645059586, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15281230211257935, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14019204676151276, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13284754753112793, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11023604869842529, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10394564270973206, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07807482779026031, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06738197803497314, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06207658722996712, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06077679619193077, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03919987380504608, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.032564856112003326, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03208492696285248, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.028935367241501808, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02807462587952614, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021053776144981384, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02113846130669117, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019322386011481285, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014666052535176277, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03919987380504608, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03919987380504608, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12933427095413208, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.120733343064785, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11722864955663681, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10599777102470398, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06081020087003708, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.057298462837934494, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06945572793483734, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06406410783529282, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06178777292370796, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.054406460374593735, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.051657844334840775, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035338882356882095, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030657554045319557, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029162539169192314, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028806468471884727, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017681032419204712, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014980277977883816, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014832451939582825, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013671248219907284, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013441859744489193, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009227101691067219, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009129716083407402, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008746170438826084, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005919819697737694, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035338882356882095, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035338882356882095, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12024489790201187, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11217585951089859, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10858089476823807, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09819061309099197, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.056435540318489075, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05303884297609329, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06501323729753494, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.059934101998806, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05735092982649803, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05047477036714554, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04801254719495773, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03304790332913399, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028642496094107628, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02706655114889145, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02668326534330845, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016525113955140114, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013890686444938183, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013725369237363338, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012672694399952888, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012426131404936314, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00860759150236845, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008483389392495155, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008106552995741367, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005454600788652897, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03304790332913399, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03304790332913399, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.27390313148498535, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2560062110424042, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.25006359815597534, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.22643324732780457, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1292484998703003, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12255951762199402, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1456911563873291, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1337856501340866, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1312239170074463, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11565674096345901, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10978846251964569, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07416612654924393, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06395638734102249, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.061848659068346024, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06134548410773277, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03701332211494446, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03147614002227783, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.031313493847846985, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.028669552877545357, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.028337758034467697, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01906733401119709, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018572483211755753, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01827814429998398, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011545530520379543, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03701332211494446, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03701332211494446, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23502205312252045, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2021302580833435, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18869459629058838, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15075068175792694, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10858101397752762, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09525317698717117, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13213838636875153, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12072587013244629, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11313489824533463, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0858675017952919, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07591484487056732, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06822891533374786, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.058573924005031586, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05303417891263962, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.051663804799318314, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03449294716119766, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02878204733133316, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028283316642045975, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024565426632761955, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023667437955737114, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019302617758512497, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02005588635802269, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017585400491952896, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0151862483471632, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03449294716119766, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03449294716119766, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21794897317886353, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20400036871433258, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.19954077899456024, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18078428506851196, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10297644138336182, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09790337830781937, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11509781330823898, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1061672642827034, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10455812513828278, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0922965258359909, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08754372596740723, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.058757275342941284, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05089656263589859, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.049446284770965576, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04910435155034065, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029365016147494316, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.025565000250935555, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025463206693530083, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023404281586408615, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023188265040516853, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01545973215252161, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015642158687114716, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014980350621044636, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01057568658143282, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029365016147494316, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029365016147494316, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.27329307794570923, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2560591995716095, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2505354583263397, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.22710232436656952, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12892629206180573, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12263822555541992, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.14419026672840118, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13278788328170776, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.13084813952445984, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11560488492250443, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10979823023080826, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07344166189432144, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06349828094244003, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06175285950303078, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.061332348734140396, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03671461343765259, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.031531646847724915, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0314132459461689, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02878335677087307, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.028516802936792374, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01919039897620678, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018693827092647552, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.018612265586853027, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011879732832312584, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03671461343765259, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03671461343765259, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2849362790584564, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2546873688697815, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.24171362817287445, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21411561965942383, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1327090561389923, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11985348165035248, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15851134061813354, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14530253410339355, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13708071410655975, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11363423615694046, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10704943537712097, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08087418228387833, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06994251161813736, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06415467709302902, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06273327022790909, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04050442576408386, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.033731140196323395, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.033195946365594864, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029964283108711243, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.029032183811068535, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02155204676091671, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02203311212360859, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.0196602251380682, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015386142767965794, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04050442576408386, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04050442576408386, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12957578897476196, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12070087343454361, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11692099273204803, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10553713142871857, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.060814227908849716, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05712777376174927, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06980423629283905, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06442465633153915, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06183239445090294, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05425170809030533, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.051487427204847336, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03551458194851875, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03082386404275894, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029185544699430466, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028788181021809578, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017774848267436028, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01502605713903904, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014855049550533295, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013690797612071037, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013436836190521717, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009296182543039322, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00923464447259903, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008763845078647137, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00602460652589798, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03551458194851875, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03551458194851875, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11695989966392517, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10893625766038895, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10511457920074463, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0948648601770401, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05487750843167305, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05135118216276169, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06376303732395172, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05879269540309906, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.055798470973968506, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04899336397647858, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04658401757478714, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03239667788147926, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028105478733778, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02633143961429596, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02589610405266285, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01620972715318203, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013535693287849426, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01334270928055048, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012337330728769302, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012065471149981022, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00846641045063734, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008349533192813396, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007907463237643242, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005390997976064682, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03239667788147926, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03239667788147926, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.27556610107421875, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.257266640663147, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.25096240639686584, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.22689920663833618, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12999245524406433, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12314783781766891, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1471528708934784, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.134917214512825, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13210687041282654, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11614273488521576, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11026030778884888, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07494945079088211, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06449425965547562, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06224683299660683, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.061716120690107346, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037399109452962875, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.031699810177087784, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03152088820934296, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.028823228552937508, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.028475699946284294, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01926272176206112, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018741577863693237, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01840030588209629, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011674451641738415, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037399109452962875, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.037399109452962875, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2339085191488266, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1968599259853363, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18339388072490692, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15135063230991364, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10825008898973465, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09357606619596481, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12880241870880127, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11816918849945068, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11261395364999771, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08295056223869324, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07624324411153793, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06647574156522751, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.057258378714323044, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.052768953144550323, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.051661502569913864, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03352407366037369, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028459947556257248, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028108596801757812, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023734193295240402, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02299450896680355, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018537040799856186, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01947031356394291, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01714697852730751, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014657719060778618, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03352407366037369, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03352407366037369, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22807377576828003, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.21337902545928955, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20871205627918243, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18896116316318512, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10771683603525162, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10237841308116913, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1202535554766655, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11104124784469604, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10936751961708069, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09649187326431274, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09145557880401611, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06140049174427986, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05321275442838669, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05170561745762825, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05136193707585335, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0306696780025959, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02669619396328926, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.026596933603286743, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.024422641843557358, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02419907972216606, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016098786145448685, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016290687024593353, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01560395210981369, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010959915816783905, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0306696780025959, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0306696780025959, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2779473066329956, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.26017436385154724, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.25456926226615906, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.23071862757205963, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1311655044555664, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12469644844532013, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.14636605978012085, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13507618010044098, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1331501603126526, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11753785610198975, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11144975572824478, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07452137768268585, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06456508487462997, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0627775564789772, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06236286088824272, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.037205472588539124, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.031988393515348434, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03186822310090065, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02916591241955757, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02889280579984188, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.019290024414658546, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01884864643216133, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.018692251294851303, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011818296276032925, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.037205472588539124, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.037205472588539124, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2900710701942444, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2596666216850281, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2465079426765442, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21857039630413055, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1351984441280365, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12226084619760513, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16147685050964355, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14815238118171692, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13968037068843842, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11603131145238876, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10936572402715683, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08251727372407913, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07129783928394318, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06536202877759933, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.0639096200466156, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04150935262441635, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.0343482680618763, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.0337970107793808, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030560918152332306, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02960055135190487, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022430509328842163, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02242312952876091, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02057054080069065, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015620035119354725, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04150935262441635, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04150935262441635, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.14506322145462036, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13519009947776794, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1311548799276352, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11844000965356827, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06834855675697327, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0642687976360321, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07866378128528595, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07219474017620087, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06947910785675049, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.061038024723529816, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05813032388687134, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040155693888664246, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.034611135721206665, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03283076360821724, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03240286186337471, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020090457051992416, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016925061121582985, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01674829237163067, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01544337347149849, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015177277848124504, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01049683429300785, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010426228865981102, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009918730705976486, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006874780636280775, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040155693888664246, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040155693888664246, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12573778629302979, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11711239069700241, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11295673996210098, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10199549049139023, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05908891186118126, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05525388568639755, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06884552538394928, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06340014189481735, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06010192632675171, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05275138095021248, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05020653456449509, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035026028752326965, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030355073511600494, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028386330232024193, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02791450545191765, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01753423735499382, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01467385794967413, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01446343120187521, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013391859829425812, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013092724606394768, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009196766652166843, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009163815528154373, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00859031267464161, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006063447333872318, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035026028752326965, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035026028752326965, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.26565316319465637, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24815289676189423, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24184681475162506, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21881848573684692, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1254405379295349, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11867640167474747, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14261853694915771, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13057878613471985, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1274363398551941, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11218240857124329, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.1065300703048706, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07275071740150452, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06250329315662384, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06011321395635605, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05954423174262047, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03633161261677742, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030720889568328857, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030513491481542587, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027991922572255135, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.027618566527962685, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018791789188981056, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018373865634202957, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017870498821139336, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011642576195299625, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03633161261677742, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03633161261677742, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.17355714738368988, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15466755628585815, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14822135865688324, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12925103306770325, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08117491006851196, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07378113269805908, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09444781392812729, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.08612124621868134, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08371537923812866, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06845668703317642, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06361936032772064, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04876724258065224, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04190043359994888, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03963329270482063, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.039086077362298965, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.024602191522717476, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02148447372019291, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.021335195749998093, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01907879114151001, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0187472365796566, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01373131200671196, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014651736244559288, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013047545216977596, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011306890286505222, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04190043359994888, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04190043359994888, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.18586774170398712, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.17354337871074677, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16953879594802856, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1531369537115097, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08785330504179001, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0832982063293457, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09861279278993607, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09082432091236115, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08925380557775497, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07842309027910233, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07440613210201263, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.050468239933252335, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04363527148962021, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04229722544550896, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.041976310312747955, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.025288019329309464, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02205890789628029, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.021968241780996323, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020176364108920097, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019980676472187042, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01351618580520153, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01379945408552885, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013081556186079979, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009666417725384235, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04363527148962021, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04363527148962021, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2606738805770874, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24352088570594788, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.238035187125206, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2151610553264618, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1229923665523529, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11673301458358765, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13803407549858093, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12687326967716217, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12487641721963882, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10980843752622604, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1041874960064888, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07026110589504242, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06070271506905556, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05892963707447052, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05851053446531296, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035179879516363144, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030153684318065643, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03002774529159069, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02743876352906227, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027173159644007683, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018471864983439445, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01796908490359783, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017881890758872032, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011517507024109364, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035179879516363144, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035179879516363144, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2651684880256653, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.23544460535049438, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.22194761037826538, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.19652676582336426, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12326005846261978, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11005191504955292, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.14900951087474823, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13665986061096191, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12761807441711426, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10512853413820267, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.09941031038761139, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07642831653356552, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0663815587759018, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05997660383582115, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.058413486927747726, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03851167857646942, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03229154646396637, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.031693339347839355, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.028842682018876076, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02784954570233822, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.020818479359149933, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022162524983286858, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.018784180283546448, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01645749993622303, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03851167857646942, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03851167857646942, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13356387615203857, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12463250756263733, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12092489004135132, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1092996671795845, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06289515644311905, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.059253983199596405, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07218409329652786, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06632515788078308, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06393618136644363, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.056234005838632584, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05353759229183197, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03684215992689133, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0317760594189167, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.030213922262191772, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029836345463991165, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018441079184412956, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015594555996358395, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01544156763702631, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014239337295293808, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014003993012011051, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00967041589319706, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009617050178349018, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0091685950756073, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006390103604644537, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03684215992689133, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03684215992689133, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11850717663764954, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11044300347566605, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1068149209022522, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09648679196834564, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055666010826826096, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.052220311015844345, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06445721536874771, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05929616093635559, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05659553036093712, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.049751438200473785, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04736825078725815, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03279494494199753, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02837064303457737, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02672547847032547, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026330014690756798, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016407324001193047, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01377056259661913, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01359541155397892, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012566768564283848, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012309996411204338, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008589210920035839, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008506923913955688, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008070752955973148, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005560769699513912, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03279494494199753, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03279494494199753, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2647794187068939, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24747131764888763, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24147456884384155, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2184244841337204, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12490049004554749, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11834186315536499, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1415216624736786, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12954027950763702, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1268651932477951, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11170466244220734, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10609032213687897, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0720381885766983, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0619489960372448, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05978765711188316, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.059273432940244675, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03596380352973938, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030434073880314827, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030252398923039436, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027697568759322166, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.027363725006580353, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018502073362469673, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01797809824347496, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017663223668932915, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011163507588207722, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03596380352973938, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03596380352973938, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2214280217885971, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19679804146289825, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18779362738132477, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16255587339401245, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10324893891811371, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09369201213121414, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12145248055458069, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11101031303405762, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10671643167734146, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08669351786375046, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08021243661642075, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06282883882522583, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05405227467417717, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05051093176007271, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04966459050774574, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03184664621949196, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027540158480405807, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027272436767816544, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024442901834845543, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02391490712761879, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0180071908980608, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019039331004023552, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016950398683547974, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014787014573812485, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03184664621949196, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03184664621949196, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1841546893119812, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.17198756337165833, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16807706654071808, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.15194790065288544, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08700332790613174, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08254634588956833, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09738936275243759, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08983275294303894, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08838587999343872, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07771328836679459, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0737365260720253, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.049785275012254715, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04312828555703163, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.041839033365249634, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.041539594531059265, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.024892229586839676, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.021762754768133163, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02167314849793911, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019904140383005142, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019715765491127968, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013159830123186111, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013522855937480927, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01273573748767376, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009374096058309078, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04312828555703163, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04312828555703163, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.26180875301361084, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24459394812583923, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23918762803077698, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2164379358291626, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12344963848590851, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11726579070091248, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13812798261642456, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1272767335176468, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1254246085882187, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1103745847940445, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10470212996006012, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07032696157693863, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06086156144738197, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.059122826904058456, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05871296674013138, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03512180969119072, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030165070667862892, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03004629909992218, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027458328753709793, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02719922363758087, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018192848190665245, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017852485179901123, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01760505512356758, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011284672655165195, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03512180969119072, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03512180969119072, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.26901131868362427, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2387673556804657, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.22523106634616852, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1996394395828247, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12463486939668655, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11144115030765533, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1508181393146515, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1380539983510971, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12916457653045654, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10625357180833817, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10049114376306534, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07711564004421234, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06642335653305054, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.060279641300439835, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05875406414270401, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03879200294613838, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03177150711417198, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.031185084953904152, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02816889062523842, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.0271588247269392, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.020987659692764282, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.020993893966078758, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01903020776808262, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014687035232782364, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03879200294613838, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03879200294613838, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12588170170783997, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1174258291721344, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11385495960712433, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10289715230464935, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05918687582015991, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.055716633796691895, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06814295798540115, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0625355914235115, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.060168515890836716, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.052944380789995193, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05043281614780426, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03476572409272194, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.029982708394527435, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028455372899770737, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028090525418519974, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017413457855582237, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01472480222582817, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014573758468031883, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013460548594594002, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013233772478997707, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0091618113219738, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009147065691649914, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008674939163029194, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00613581296056509, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03476572409272194, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03476572409272194, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11359619349241257, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10587172210216522, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10223482549190521, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09229223430156708, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.053341712802648544, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.049983084201812744, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06192013621330261, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05702607333660126, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.054225239902734756, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.047673434019088745, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04536443576216698, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03151297941803932, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02729986235499382, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02562391757965088, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02521008439362049, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015769055113196373, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01321202889084816, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01303374208509922, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012062037363648415, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011803233996033669, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008257564157247543, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008201858028769493, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007741925306618214, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005389322526752949, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04536443576216698, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04536443576216698, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2652043402194977, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2478417605161667, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2418188899755478, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21875424683094025, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12506039440631866, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11850486695766449, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14176678657531738, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12988334894180298, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12703491747379303, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11190644651651382, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10629523545503616, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07223587483167648, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06210284307599068, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05989040434360504, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05937262624502182, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036043476313352585, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03054431453347206, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030357366427779198, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02781379409134388, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.027478935196995735, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018588852137327194, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01812911033630371, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017746344208717346, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011367743834853172, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036043476313352585, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036043476313352585, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22340551018714905, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1926776021718979, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.17988139390945435, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15121133625507355, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10394575446844101, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09074912965297699, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12673188745975494, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11496822535991669, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10771103203296661, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08293530344963074, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07661374658346176, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06552550196647644, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.055929649621248245, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.050906501710414886, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04968683794140816, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0332927368581295, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027813972905278206, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027360860258340836, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02393484115600586, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0231509730219841, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01898016221821308, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019523710012435913, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01747111603617668, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015061350539326668, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0332927368581295, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0332927368581295, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1907535046339035, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.17820438742637634, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17420418560504913, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.15768535435199738, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09010813385248184, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08553805947303772, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1009330227971077, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09297238290309906, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09151286631822586, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08058696985244751, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0765019953250885, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05165231227874756, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04466308653354645, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04337083548307419, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.043066419661045074, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02586095966398716, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.022615104913711548, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.022528883069753647, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020712297409772873, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020526167005300522, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013791568577289581, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014133257791399956, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013366061262786388, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009904365986585617, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04466308653354645, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04466308653354645, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.26321521401405334, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24616551399230957, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24081826210021973, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2178184539079666, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12412569671869278, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11793328821659088, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1389753371477127, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1278885453939438, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12601299583911896, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11102081090211868, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10536462813615799, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07082778960466385, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06116485223174095, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05945782735943794, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.059051189571619034, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03541019186377525, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03037029691040516, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03025364875793457, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027674326673150063, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027412164956331253, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018487222492694855, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018026726320385933, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017910026013851166, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011471392586827278, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03541019186377525, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03541019186377525, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.26677820086479187, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.23579996824264526, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.22141356766223907, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.19601202011108398, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12345495074987411, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.10980847477912903, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15032678842544556, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13797545433044434, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.12814779579639435, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10483582317829132, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.09936924278736115, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07700347155332565, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06635445356369019, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.059733666479587555, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05809122696518898, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03863472491502762, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03149852156639099, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.030854687094688416, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.027852753177285194, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02676207199692726, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.020733166486024857, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02091694064438343, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.018611738458275795, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014599024318158627, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03863472491502762, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03863472491502762, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12083377689123154, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11254087835550308, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10899629443883896, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09842647612094879, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05670803040266037, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05328379571437836, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06538258492946625, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.060101382434368134, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05764709785580635, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05063343420624733, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.048188716173172, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.033296771347522736, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028772685676813126, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02723120152950287, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026861362159252167, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016665417701005936, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014046493917703629, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013891237787902355, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012810342013835907, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012574248015880585, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008731191046535969, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008674118667840958, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008239318616688251, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0057243420742452145, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.033296771347522736, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.033296771347522736, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10907589644193649, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10154806077480316, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0979190394282341, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08834147453308105, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05111979693174362, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.047808364033699036, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.059572380036115646, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.054831452667713165, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0519757941365242, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.045634541660547256, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04346347972750664, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030297566205263138, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026241881772875786, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024547232314944267, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02413436956703663, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0151521610096097, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012658651918172836, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012476060539484024, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011552088893949986, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011296125128865242, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00793243758380413, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00788239948451519, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007407107390463352, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0051788995042443275, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.045634541660547256, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.045634541660547256, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25862422585487366, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24126377701759338, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.23502331972122192, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21236149966716766, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12169305235147476, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11503395438194275, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13837696611881256, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1267482489347458, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12361938506364822, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10865578800439835, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10316868126392365, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07046853750944138, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.060586389154195786, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05826244875788689, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05770399048924446, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03516526520252228, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02970413863658905, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029502686113119125, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02702389284968376, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02665904350578785, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018148718401789665, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017684675753116608, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01727432943880558, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011078630574047565, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03516526520252228, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03516526520252228, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24206280708312988, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2079205960035324, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1958845853805542, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16132943332195282, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11257138103246689, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0993448793888092, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1326512098312378, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12170911580324173, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11708889156579971, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08919300138950348, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07909435033798218, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06848634779453278, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.059114325791597366, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.055014923214912415, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054020486772060394, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03454836457967758, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029888805001974106, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02961917221546173, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025548120960593224, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024909446015954018, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019199583679437637, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020596813410520554, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017942916601896286, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015837719663977623, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03454836457967758, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03454836457967758, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19676895439624786, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18393270671367645, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1797741800546646, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1627371907234192, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09291519224643707, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08823757618665695, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10403881967067719, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09589312970638275, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09439106285572052, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08314234763383865, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0788855105638504, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.053193993866443634, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.046023156493902206, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04466330260038376, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04434511065483093, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026607606559991837, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023184169083833694, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023092329502105713, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021229438483715057, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02102728933095932, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014114377088844776, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014344967901706696, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01366984099149704, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009881622157990932, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04466330260038376, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04466330260038376, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.26291710138320923, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2458154559135437, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2404458224773407, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21763835847377777, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12395956367254257, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11778688430786133, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13880378007888794, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12776537239551544, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12589818239212036, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11096174269914627, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10526292771100998, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07063295692205429, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06109810993075371, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.059371642768383026, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05895717814564705, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035280514508485794, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0303179994225502, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.030193498358130455, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027630342170596123, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027364935725927353, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01830403506755829, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017975274473428726, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017717646434903145, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011406205594539642, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035280514508485794, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035280514508485794, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27312687039375305, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2417336106300354, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.22779035568237305, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.201761394739151, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1266632378101349, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11315594613552094, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15382738411426544, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14037629961967468, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13134485483169556, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10774930566549301, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10192099958658218, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07867330312728882, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06766951084136963, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.0614234134554863, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.059878334403038025, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03958108276128769, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.032621849328279495, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03203137218952179, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02895112708210945, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.027934564277529716, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02146906591951847, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021842410787940025, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019450953230261803, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01565236784517765, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03958108276128769, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03958108276128769, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12000641226768494, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1117115467786789, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10825790464878082, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09765124320983887, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05630697309970856, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05291702598333359, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06490056961774826, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0595751591026783, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.057234082370996475, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05024169385433197, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.047852929681539536, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0330849327147007, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02851184271275997, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.027024848386645317, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026672467589378357, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01655738614499569, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013940734788775444, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013791847042739391, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012711127288639545, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012482017278671265, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008686354383826256, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008601948618888855, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008206331171095371, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005684418138116598, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0330849327147007, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0330849327147007, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10757103562355042, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10015101730823517, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09640942513942719, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08697009086608887, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05039229616522789, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04705167934298515, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05895796790719032, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.054269734770059586, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05123428255319595, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04497290402650833, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0428616963326931, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.029976394027471542, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025961020961403847, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02420422062277794, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.023781003430485725, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015010877512395382, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01251088734716177, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012316145934164524, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011413960717618465, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011145791038870811, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007885473780333996, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007832683622837067, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007343502715229988, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005171789322048426, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04497290402650833, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04497290402650833, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.266327828168869, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24855895340442657, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24251848459243774, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2190905660390854, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12542524933815002, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11872442066669464, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1421104222536087, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13020800054073334, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12747298181056976, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11199131608009338, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.1063271313905716, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0724206417798996, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.062245577573776245, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0600271001458168, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05949359014630318, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0361325740814209, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030576828867197037, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030392512679100037, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.027783580124378204, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02744601108133793, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01860879361629486, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018101802095770836, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01776227355003357, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01127160806208849, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0361325740814209, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0361325740814209, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2328794151544571, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19379302859306335, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.17835113406181335, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15161946415901184, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10702122002840042, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09083742648363113, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1308150291442871, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11968470364809036, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11186076700687408, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08307883143424988, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07783400267362595, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06749121844768524, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0582725889980793, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05257193371653557, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05117126926779747, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03423846885561943, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028996901586651802, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02849436178803444, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02467556670308113, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023754313588142395, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01944407820701599, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02072429284453392, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017770197242498398, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.016178885474801064, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03423846885561943, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03423846885561943, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19699804484844208, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18418020009994507, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18001483380794525, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16295762360095978, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09299711138010025, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08830784261226654, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1040259599685669, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09592698514461517, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09443512558937073, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08320511877536774, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0789325162768364, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.053144097328186035, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.046002257615327835, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044661544263362885, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.044352173805236816, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02654626965522766, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02312335930764675, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023030247539281845, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021155979484319687, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020953301340341568, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013982956297695637, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014209453947842121, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013542060740292072, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009670211002230644, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044661544263362885, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.044661544263362885, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2623717188835144, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24550148844718933, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24011015892028809, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21738046407699585, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1236715093255043, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11753461509943008, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13837295770645142, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12745092809200287, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12557464838027954, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1107589453458786, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10509802401065826, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07044381648302078, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06093937158584595, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05921071395277977, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.058804020285606384, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035157106816768646, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030190473422408104, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03006870485842228, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02751557156443596, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027256028726696968, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01821829192340374, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017834803089499474, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017641546204686165, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011231167241930962, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035157106816768646, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035157106816768646, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.27141669392585754, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2404702603816986, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.22650781273841858, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20098333060741425, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12579862773418427, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11240202188491821, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1529068648815155, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13968917727470398, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13045845925807953, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10720743983983994, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10152779519557953, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07825905084609985, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06723374128341675, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06092577800154686, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05935932695865631, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03936675563454628, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03222762793302536, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03162695840001106, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.028598159551620483, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.027569113299250603, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.021382981911301613, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02144523151218891, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019342100247740746, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015173125080764294, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03936675563454628, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03936675563454628, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12631461024284363, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11795705556869507, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11457138508558273, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10367728024721146, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05940763279795647, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05601444095373154, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0679982602596283, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06258690357208252, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06037582457065582, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05322974920272827, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0507027767598629, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034620095044374466, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02993391454219818, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02848721295595169, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028141112998127937, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017331654205918312, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014613918028771877, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014473095536231995, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01334743294864893, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013125349767506123, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009055525064468384, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00888116005808115, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008585008792579174, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005720209795981646, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034620095044374466, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034620095044374466, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1164470985531807, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10866929590702057, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1052735447883606, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09527548402547836, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05470993369817734, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.051445122808218, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0631810650229454, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05803539231419563, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.055592991411685944, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04899713769555092, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04672825708985329, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032119520008563995, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027751201763749123, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02623092010617256, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025862915441393852, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016075892373919487, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013464808464050293, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013308254070580006, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012300612404942513, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012065764516592026, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00838630273938179, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008227542974054813, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007902982644736767, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005292639601975679, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032119520008563995, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032119520008563995, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2721402943134308, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.25448378920555115, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24856296181678772, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.22529612481594086, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1282907873392105, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12173381447792053, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1448879987001419, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13289347290992737, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13033795356750488, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11501174420118332, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10938936471939087, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07376181334257126, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06351443380117416, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.061412882059812546, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06091006100177765, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03681914135813713, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.031265053898096085, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.031104398891329765, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02851283550262451, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.028186501935124397, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018950197845697403, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018449408933520317, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018162215128540993, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011471579782664776, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03681914135813713, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03681914135813713, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2465144544839859, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.20750409364700317, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19177667796611786, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15050926804542542, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1145152673125267, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09870291501283646, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1389010101556778, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12686920166015625, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11888085305690765, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08426342159509659, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07757635414600372, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07175908237695694, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.061348024755716324, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05573625490069389, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054352618753910065, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036093033850193024, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02991131693124771, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029401876032352448, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024228744208812714, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02326568216085434, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01986568048596382, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020409032702445984, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018090171739459038, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01508421916514635, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036093033850193024, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036093033850193024, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21259857714176178, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.198878675699234, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1944257616996765, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.17605292797088623, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10038001835346222, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09535809606313705, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11220710724592209, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10346969962120056, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1018916666507721, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08986987918615341, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08523046970367432, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05730767548084259, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04960912466049194, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04819696024060249, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0478646457195282, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028637094423174858, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024926835671067238, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02483285591006279, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.022805120795965195, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022595813497900963, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015072766691446304, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015277245081961155, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01460679154843092, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010357496328651905, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028637094423174858, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028637094423174858, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.27061915397644043, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.25336095690727234, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.24787811934947968, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.22457940876483917, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12763795256614685, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12132973968982697, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.14295873045921326, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13145963847637177, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.129587322473526, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1143372431397438, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1086433008313179, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0727975144982338, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0628553181886673, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06111491098999977, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06070069968700409, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03639113903045654, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.031214077025651932, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.031096110120415688, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028472745791077614, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02820904739201069, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018992042168974876, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018510989844799042, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.018414989113807678, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01176486536860466, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03639113903045654, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03639113903045654, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2779248058795929, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.24619047343730927, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23178631067276, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20535199344158173, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12888933718204498, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11506427079439163, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15698382258415222, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14337649941444397, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1337089240550995, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10983414202928543, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10391518473625183, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08009029179811478, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.06906238198280334, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.062433354556560516, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06079242005944252, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04018981754779816, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.032984402030706406, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03234957531094551, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.029248714447021484, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02816626988351345, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02152714505791664, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021920524537563324, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01937440037727356, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015422266907989979, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04018981754779816, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04018981754779816, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12678740918636322, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11807338893413544, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1144818663597107, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1032768189907074, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05954241007566452, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0559668093919754, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06835561245679855, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0628870502114296, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.060546454042196274, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05308949947357178, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.050448186695575714, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03477853909134865, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030074283480644226, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02856311947107315, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0281947273761034, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01741175539791584, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01467273198068142, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014522725716233253, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013356351293623447, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013124777935445309, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009091834537684917, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008965211920440197, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008591226302087307, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005815819371491671, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03477853909134865, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03477853909134865, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11448004841804504, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10660655051469803, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10299612581729889, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09298398345708847, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05377183109521866, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05037781596183777, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06244891136884689, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05738304927945137, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05467874929308891, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.047990843653678894, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04570499807596207, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03176584094762802, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027440447360277176, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02580302394926548, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025400929152965546, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015892626717686653, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013271099887788296, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013102786615490913, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012096949853003025, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011848513036966324, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00829867273569107, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008181579411029816, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007775142788887024, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005318989045917988, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04570499807596207, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04570499807596207, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2732711732387543, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2550671100616455, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24880771338939667, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.22487445175647736, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12890790402889252, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12212705612182617, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14599783718585968, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1337086409330368, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13100571930408478, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11509186774492264, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.1092950701713562, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07437418401241302, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06390939652919769, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.061718907207250595, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0612114816904068, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03712237998843193, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03144264221191406, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03126964718103409, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02858482301235199, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.028247108682990074, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01911819539964199, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018612369894981384, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018293943256139755, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01163435261696577, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03712237998843193, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03712237998843193, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24476251006126404, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2070678472518921, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19472870230674744, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15747077763080597, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11303592473268509, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09937994182109833, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13297949731349945, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1218051165342331, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11800547689199448, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0870945155620575, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07654278725385666, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06853553652763367, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05882476270198822, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.054840087890625, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05388713628053665, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03446245193481445, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029103538021445274, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02885972149670124, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02410963922739029, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02345256134867668, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018816189840435982, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019235141575336456, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017555266618728638, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013981414027512074, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03446245193481445, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03446245193481445, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22284364700317383, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20840618014335632, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20381689071655273, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18459360301494598, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10529649257659912, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10002768039703369, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11768338829278946, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10852574557065964, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10690385103225708, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0942152738571167, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08937117457389832, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06011262908577919, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05202382430434227, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05054257810115814, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.050191815942525864, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03004353865981102, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02611757069826126, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02601846307516098, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023876477032899857, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023661978542804718, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015812361612915993, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01596337929368019, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015327230095863342, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010778514668345451, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03004353865981102, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03004353865981102, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.27584463357925415, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2580218017101288, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.25240829586982727, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.22856184840202332, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13014662265777588, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12371261417865753, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1455031782388687, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13408790528774261, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.13214370608329773, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11650058627128601, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11053458601236343, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07408128678798676, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06409649550914764, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06230194494128227, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06188437342643738, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03698313608765602, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03176131844520569, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.031638018786907196, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028938980773091316, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02866521291434765, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01914997585117817, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018742870539426804, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01854833774268627, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01178143359720707, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03698313608765602, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03698313608765602, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2819303274154663, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2507208287715912, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.23639002442359924, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.20954664051532745, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1309930384159088, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.11730634421110153, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15874244272708893, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.14557480812072754, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13577748835086823, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11201359331607819, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10580473393201828, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08115877956151962, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0701216608285904, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.0634700134396553, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06182923540472984, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04087013378739357, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.0335761234164238, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03293830156326294, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02986607886850834, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.028785889968276024, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022182991728186607, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022349586710333824, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020096808671951294, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015794765204191208, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04087013378739357, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04087013378739357, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13002778589725494, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12109719216823578, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1174008920788765, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10589279234409332, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06111690029501915, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05745919048786163, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07056036591529846, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06461144983768463, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06217561662197113, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05453142523765564, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05190025269985199, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035949040204286575, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030938180163502693, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029333777725696564, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028942333534359932, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01801709644496441, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015075808390974998, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014918558299541473, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013731727376580238, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013482914306223392, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00949951633810997, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009225658141076565, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00896633043885231, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005990749225020409, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035949040204286575, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035949040204286575, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11951989680528641, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11125768721103668, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10764062404632568, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09719820320606232, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.056145962327718735, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.052684444934129715, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06500758975744247, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.059706296771764755, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05710800737142563, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05010317638516426, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04766564071178436, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0330718569457531, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028555601835250854, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026941021904349327, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026549935340881348, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01653684675693512, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013823054730892181, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013660810887813568, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012585802935063839, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012336328625679016, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008619318716228008, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008452046662569046, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0080947270616889, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005429121665656567, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0330718569457531, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0330718569457531, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.27372896671295166, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2553507685661316, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24910429120063782, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.22510679066181183, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12912686169147491, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12225356698036194, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1461586207151413, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.133978009223938, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13118989765644073, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11527436226606369, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10942371934652328, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07444193959236145, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06405723094940186, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06181982159614563, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.061297062784433365, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03715327009558678, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.031466372311115265, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.031292062252759933, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.028589868918061256, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02824871614575386, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019132772460579872, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018590139225125313, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018291836604475975, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011535778641700745, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03715327009558678, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03715327009558678, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18518756330013275, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15872575342655182, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14977402985095978, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12792165577411652, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08634960651397705, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07614946365356445, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10144758969545364, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09247579425573349, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08934271335601807, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06889703869819641, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06320183724164963, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.052351657301187515, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.045134302228689194, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04237672686576843, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.041703078895807266, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026580750942230225, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.023308465257287025, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.023123133927583694, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.020131470635533333, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01970854587852955, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01521175168454647, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01632191799581051, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01439201831817627, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012896602973341942, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.045134302228689194, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.045134302228689194, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2348610758781433, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.21952322125434875, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2146318256855011, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1941692978143692, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11099138855934143, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10538322478532791, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12426169961690903, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11449848860502243, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1127486377954483, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09925538301467896, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0941179171204567, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06348752975463867, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05486742779612541, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.053274355828762054, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.052902158349752426, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03174622356891632, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.027472972869873047, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02736489474773407, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02508445829153061, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.024848520755767822, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016720199957489967, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016711704432964325, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016194799914956093, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011174446903169155, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03174622356891632, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03174622356891632, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2817893326282501, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.26363325119018555, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2578136920928955, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.23333074152469635, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1330614984035492, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12639033794403076, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.14901024103164673, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13713422417640686, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.13512523472309113, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11906830966472626, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11301884055137634, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.076017826795578, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06562382727861404, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06376029551029205, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06331395357847214, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038020357489585876, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03257301077246666, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03243822604417801, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.029681596904993057, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02939457818865776, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.019878434017300606, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.019335772842168808, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.019265225157141685, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012306290678679943, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038020357489585876, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038020357489585876, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2930505573749542, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2626646161079407, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.249739408493042, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.22164249420166016, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13648559153079987, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12373426556587219, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16231077909469604, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1491585671901703, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14098584651947021, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.117427296936512, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11067976802587509, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08301696181297302, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07160814851522446, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06581199169158936, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06438709795475006, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04164668545126915, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03420327231287956, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03367462009191513, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030369676649570465, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.029418036341667175, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022172076627612114, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02181784063577652, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020294521003961563, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01459931768476963, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04164668545126915, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04164668545126915, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.40.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13296982645988464, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12383131682872772, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12008413672447205, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10835842788219452, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06255815178155899, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05883357301354408, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07185758650302887, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06604062020778656, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06359568983316422, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05575636774301529, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.053017739206552505, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036598317325115204, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03159632533788681, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.030026230961084366, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02964198961853981, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018309079110622406, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015421314164996147, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015266899019479752, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014040063135325909, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013797745108604431, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009577582590281963, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009419593960046768, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00907235685735941, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0061150058172643185, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036598317325115204, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036598317325115204, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.40.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1214359775185585, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11309175193309784, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10933728516101837, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09868618845939636, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05710570141673088, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.053533006459474564, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06626086682081223, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06080778315663338, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05807044357061386, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05093209445476532, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.048488374799489975, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03370184823870659, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.029080867767333984, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.027413222938776016, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.027006970718503, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016865434125065804, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014074769802391529, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013901306316256523, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012814217247068882, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012554231099784374, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0088056530803442, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008628477342426777, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008273969404399395, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00555709283798933, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03370184823870659, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03370184823870659, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.40.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.27971309423446655, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2610063850879669, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.25472205877304077, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.23020796477794647, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1320914626121521, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12509965896606445, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14949259161949158, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13692666590213776, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1342061460018158, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11786752194166183, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.1119748204946518, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07615790516138077, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06548139452934265, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06324777752161026, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06272196024656296, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03801741451025009, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03219134733080864, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03202017396688461, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02925233356654644, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.028906119987368584, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01956160180270672, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01902402937412262, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018714459612965584, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011818052269518375, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03801741451025009, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03801741451025009, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.40.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24310895800590515, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19719287753105164, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.17718558013439178, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.14926055073738098, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11073873937129974, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09001564979553223, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14010962843894958, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12794826924800873, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.117324598133564, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08412650972604752, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07959149032831192, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07251203060150146, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06184318661689758, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.054028093814849854, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.052049051970243454, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03657934069633484, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029077567160129547, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028388094156980515, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024065004661679268, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.022720661014318466, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02028033137321472, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02031421847641468, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01787908934056759, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014841403812170029, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03657934069633484, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03657934069633484, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.40.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2405988574028015, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22481882572174072, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21981264650821686, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19882290065288544, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11377127468585968, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10800641030073166, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12736651301383972, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11729444563388824, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1155368760228157, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10167285054922104, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09639215469360352, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06499253958463669, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.056214459240436554, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05459735170006752, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.054218839854002, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03252328559756279, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.028179891407489777, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.028065118938684464, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.025724872946739197, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.025483423843979836, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017122384160757065, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01715896837413311, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016593564301729202, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011503208428621292, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03252328559756279, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03252328559756279, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.40.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.28512364625930786, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.26662448048591614, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2608236074447632, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.23598143458366394, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13465432822704315, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1279274821281433, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1507396101951599, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13877254724502563, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.13673542439937592, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12044163048267365, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11426471918821335, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0767940804362297, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06638550013303757, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0645078644156456, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06406089663505554, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038376182317733765, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.032943353056907654, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.032811667770147324, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.03001253679394722, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.029723919928073883, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.020001225173473358, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01954541727900505, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.019379347562789917, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012414192780852318, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038376182317733765, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038376182317733765, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.40.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2951524257659912, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.26509296894073486, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.25197818875312805, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.22357584536075592, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1375976800918579, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12484095245599747, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16407889127731323, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1506763994693756, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14205332100391388, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11854139715433121, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11172246187925339, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08370646089315414, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07226096838712692, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06631267070770264, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.0648488774895668, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.041890308260917664, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03440488874912262, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03384504094719887, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030561214312911034, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02958083525300026, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022066766396164894, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021894104778766632, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02010556124150753, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014532510191202164, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.041890308260917664, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.041890308260917664, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.41.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13404452800750732, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12469899654388428, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12074510753154755, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1088278740644455, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06308401376008987, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.059203118085861206, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0728602409362793, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06679542362689972, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06416425108909607, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05611814558506012, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.053383298218250275, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03709022328257561, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.031950004398822784, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.030281784012913704, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02987646870315075, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018575359135866165, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015568696893751621, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015403025783598423, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014144772663712502, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013886705972254276, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009723363444209099, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009530676528811455, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00918339192867279, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006180086173117161, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03709022328257561, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03709022328257561, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.41.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12401151657104492, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11533975601196289, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11145638674497604, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10054238140583038, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.058338772505521774, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05464909225702286, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06795316934585571, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06212526932358742, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05931674689054489, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.051900070160627365, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.049476806074380875, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034519296139478683, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.029703008010983467, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.027996249496936798, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.027584213763475418, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017272252589464188, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014383681118488312, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01420622505247593, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013068783096969128, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012806653045117855, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009033652022480965, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00881364569067955, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008492994122207165, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005682372022420168, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034519296139478683, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034519296139478683, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.41.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2811432480812073, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2621254622936249, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.25563040375709534, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.23050054907798767, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.13284926116466522, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12566035985946655, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15052534639835358, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13784082233905792, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1350088119506836, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11832278966903687, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11223697662353516, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0767904594540596, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06594053655862808, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06366956233978271, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06311188638210297, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03832238167524338, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03249863162636757, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.032321635633707047, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.029483914375305176, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02913941629230976, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01980624347925186, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019335482269525528, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018925156444311142, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012225569225847721, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03832238167524338, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03832238167524338, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.41.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24154895544052124, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1984962671995163, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18255305290222168, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15148788690567017, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11023092269897461, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09235864877700806, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1352020800113678, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12278624624013901, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11667946726083755, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08441166579723358, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07843433320522308, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06979593634605408, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05956854298710823, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.053882893174886703, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05245538428425789, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03521793708205223, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029179587960243225, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028796497732400894, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02437419258058071, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023427298292517662, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019597552716732025, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020270371809601784, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017847329378128052, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015283660031855106, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03521793708205223, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03521793708205223, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.41.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2451978474855423, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22912269830703735, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2240019291639328, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20264720916748047, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11596639454364777, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11005070805549622, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1296846568584442, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11959372460842133, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1177777647972107, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1036132425069809, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09825992584228516, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06622827053070068, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05730428919196129, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05563758686184883, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05524464696645737, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03309212252497673, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.028654685243964195, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.028543706983327866, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02614540606737137, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02589605748653412, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017327893525362015, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017373796552419662, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016783686354756355, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011539251543581486, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03309212252497673, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03309212252497673, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.41.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.28726324439048767, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.268567830324173, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2626749575138092, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.23762983083724976, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13569369912147522, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12885673344135284, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15189331769943237, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13987429440021515, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1378292590379715, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12132856249809265, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11507273465394974, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07733533531427383, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06688366085290909, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06498393416404724, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06452275812625885, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038630496710538864, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03311021253466606, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03298032283782959, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.030122727155685425, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02983231469988823, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.02003922313451767, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.0195241067558527, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.0194042157381773, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01223886851221323, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038630496710538864, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038630496710538864, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.41.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.29581180214881897, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.26647624373435974, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.25341352820396423, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.22488102316856384, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13825884461402893, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12559214234352112, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16499638557434082, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15149685740470886, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1425354927778244, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.1193733736872673, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11262564361095428, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08446721732616425, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07285895198583603, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06678996980190277, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06528230756521225, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04254297539591789, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.034905534237623215, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03431811183691025, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03113858588039875, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.030144669115543365, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022743098437786102, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022551022469997406, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020739074796438217, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015395251102745533, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04254297539591789, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04254297539591789, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.42.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.14428673684597015, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13446052372455597, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13041985034942627, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11782292276620865, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06797594577074051, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06396983563899994, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0779288113117218, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07165691256523132, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06914081424474716, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06063190847635269, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05761782452464104, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03964922949671745, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.034270599484443665, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.032606806606054306, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.032206661999225616, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.019826781004667282, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016716813668608665, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016557451337575912, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015207961201667786, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014952996745705605, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010317129082977772, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01013981644064188, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009766358882188797, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006503083277493715, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03964922949671745, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03964922949671745, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.42.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13159509003162384, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12263189256191254, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11878972500562668, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10730260610580444, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.062005460262298584, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05826485902070999, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07158058136701584, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06569565832614899, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06304528564214706, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05536430701613426, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05273950472474098, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036398015916347504, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03141394630074501, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02974666655063629, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02934163063764572, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01820656843483448, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015233609825372696, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015068493783473969, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013870990835130215, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013617563992738724, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00946141593158245, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009248455055058002, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008920526131987572, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005894220434129238, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036398015916347504, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036398015916347504, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.42.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2857469618320465, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2667505741119385, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2604396939277649, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2353922426700592, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.13498909771442413, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1279219537973404, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1527763158082962, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1398507058620453, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1372486799955368, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12061674147844315, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11459186673164368, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07784205675125122, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06691119819879532, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06468871980905533, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0641530230641365, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03884482383728027, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.032928165048360825, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03275853022933006, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.029928890988230705, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.029587632045149803, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01999109797179699, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019433870911598206, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0191377941519022, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012092567048966885, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03884482383728027, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03884482383728027, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.42.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2283944934606552, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19281277060508728, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.17451469600200653, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1470104306936264, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10536083579063416, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08880111575126648, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13432276248931885, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12326548248529434, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10991702228784561, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08276651054620743, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0774327963590622, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06936608254909515, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05933850258588791, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05109556391835213, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04897407814860344, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.034912437200546265, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026954855769872665, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026034368202090263, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.022543061524629593, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021102044731378555, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018820883706212044, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018365228548645973, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01616596058011055, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012558150105178356, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.034912437200546265, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.034912437200546265, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.42.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25035497546195984, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2337692379951477, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2284734696149826, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2066546529531479, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11843841522932053, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11236713826656342, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13260358572006226, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12227146327495575, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12035764753818512, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10579638928174973, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10022161900997162, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06771602481603622, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.058591026812791824, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05685972049832344, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05643733590841293, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03384462743997574, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029269637539982796, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02914806269109249, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02668362855911255, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02642143703997135, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017738990485668182, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017745383083820343, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017164940014481544, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01176113449037075, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03384462743997574, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03384462743997574, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.42.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.290524423122406, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2713605761528015, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2653081715106964, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2400028109550476, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1372966468334198, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.13027261197566986, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15365450084209442, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14156843721866608, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1394662857055664, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12264779210090637, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11622318625450134, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0782923772931099, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06773098558187485, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06575362384319305, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06528430432081223, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0390653982758522, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03351059928536415, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.033370811492204666, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.03047272562980652, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03016781434416771, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.020208733156323433, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.019779430702328682, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.019543640315532684, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012394776567816734, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0390653982758522, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0390653982758522, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.42.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.30170679092407227, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2722701132297516, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.25964903831481934, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.2302255481481552, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.14108656346797943, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12861105799674988, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1670653074979782, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15365450084209442, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14533264935016632, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12181176245212555, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11465124040842056, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08530570566654205, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07365299016237259, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06790374964475632, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06649753451347351, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04269518703222275, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.035058412700891495, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03450830653309822, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.031145004555583, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.03018731251358986, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02242845669388771, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022011231631040573, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02053060196340084, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014311076141893864, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04269518703222275, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04269518703222275, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.43.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1429445743560791, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13289402425289154, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12869077920913696, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1160847470164299, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06732872873544693, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06317693740129471, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07762228697538376, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0713152140378952, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06850353628396988, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.059924960136413574, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0569109208881855, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03956298902630806, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03413006663322449, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03232989460229874, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03189237788319588, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.019819805398583412, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01663213036954403, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016451586037874222, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015117483213543892, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014837516471743584, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010390794835984707, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01020257268100977, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009795334190130234, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006638197228312492, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03956298902630806, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03956298902630806, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.43.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12945391237735748, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12034901231527328, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11634108424186707, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1049954742193222, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06094792112708092, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05710504949092865, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07080670446157455, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06492196768522263, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.062021493911743164, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.054304517805576324, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.051696084439754486, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036033496260643005, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.031063104048371315, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029269810765981674, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028832528740167618, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018027540296316147, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015025445260107517, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014839145354926586, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013658521696925163, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013384648598730564, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009401435032486916, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009204190224409103, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008817512542009354, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005915932822972536, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036033496260643005, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036033496260643005, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.43.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.28215745091438293, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.26268303394317627, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2561355531215668, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.23109427094459534, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1332363784313202, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12596510350704193, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15131261944770813, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13838523626327515, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13552750647068024, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11867503076791763, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11262907832860947, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07707144320011139, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0661877691745758, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06382153928279877, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06324166059494019, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03847391903400421, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03247253969311714, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.032285649329423904, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.029436934739351273, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.029071539640426636, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01978076808154583, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019169310107827187, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018878405913710594, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011856324970722198, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03847391903400421, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03847391903400421, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.43.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23890075087547302, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19710364937782288, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1835498809814453, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1464342176914215, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10594141483306885, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09049417823553085, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13053612411022186, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11619972437620163, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11139553040266037, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08119721710681915, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07545249164104462, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06604451686143875, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05774443596601486, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05313869193196297, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.052006397396326065, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03435841575264931, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030886821448802948, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030596889555454254, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.026713555678725243, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026029620319604874, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020688802003860474, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.023327460512518883, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01940629445016384, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.019641108810901642, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03435841575264931, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03435841575264931, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.43.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.24845996499061584, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2319677174091339, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22667056322097778, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20466084778308868, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11762688308954239, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11153171211481094, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13207939267158508, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12152186036109924, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11952003836631775, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10496778786182404, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09942463040351868, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06752185523509979, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0582653284072876, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05647420510649681, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.056049518287181854, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03381234034895897, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029095463454723358, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02896842733025551, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02651667594909668, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026244744658470154, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01785070262849331, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017671233043074608, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017260784283280373, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011743104085326195, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03381234034895897, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03381234034895897, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.43.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2926039397716522, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2732791006565094, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2671164572238922, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.24127423763275146, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13828697800636292, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.13120011985301971, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15529343485832214, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1428111344575882, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14053690433502197, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1234663873910904, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11695568263530731, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07908330112695694, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06836926937103271, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06629656255245209, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06580403447151184, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03958559408783913, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03388550132513046, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03374313563108444, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.030820541083812714, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.030512463301420212, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.020737435668706894, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02016270160675049, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.020057523623108864, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012848629616200924, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03958559408783913, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03958559408783913, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.43.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3010493814945221, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.27256473898887634, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2600207030773163, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.23103585839271545, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1411161869764328, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1288546472787857, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.167287677526474, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1537570059299469, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14520831406116486, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12233006209135056, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11537541449069977, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08560124039649963, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07392669469118118, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06810897588729858, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06667596101760864, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042987022548913956, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03545469418168068, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03489142656326294, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03168024867773056, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.030726686120033264, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022719118744134903, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02265891805291176, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020766103640198708, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01527821272611618, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042987022548913956, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042987022548913956, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.44.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1483457237482071, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1382015198469162, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13409645855426788, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1209990382194519, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07002311199903488, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06588498502969742, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08022178709506989, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07376206666231155, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07121358811855316, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06237669661641121, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05919420346617699, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04082803800702095, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03528718650341034, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.033591751009225845, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03318144008517265, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020410437136888504, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.017180273309350014, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.017018649727106094, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015607587993144989, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015340550802648067, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010582350194454193, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010369825176894665, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010015016421675682, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00657648965716362, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04082803800702095, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04082803800702095, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.44.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13524410128593445, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1259801685810089, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12205806374549866, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1101081445813179, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06380558758974075, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.059952881187200546, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07360927760601044, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06752437353134155, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06488147377967834, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.056877732276916504, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.054036710411310196, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03746042773127556, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03229435533285141, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.030622266232967377, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03021339513361454, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01872948743402958, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01568170264363289, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015514891594648361, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014252671040594578, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013992230407893658, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00974263995885849, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009515056386590004, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009184029884636402, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006056175101548433, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03746042773127556, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03746042773127556, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.44.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.28845131397247314, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2690696120262146, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2625516355037689, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2371044158935547, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.13657726347446442, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1292392462491989, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15439607203006744, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14159362018108368, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1387840062379837, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12175233662128448, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11550857871770859, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0787568911910057, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06770089268684387, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06541294604539871, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06485699117183685, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03930332511663437, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03330100327730179, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03311946988105774, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.03021715208888054, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.029859455302357674, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020225748419761658, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01965279132127762, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019365012645721436, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012208059430122375, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03930332511663437, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03930332511663437, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.44.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24699664115905762, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19494816660881042, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1730402112007141, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1547389179468155, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1106051579117775, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0864429846405983, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14153487980365753, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12931440770626068, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11929138004779816, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0857810527086258, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08293046057224274, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07310013473033905, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06249801814556122, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05395922437310219, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05176892504096031, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036776021122932434, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028946224600076675, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02827957458794117, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02423112653195858, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.022754771634936333, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020168963819742203, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020257556810975075, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017520904541015625, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014591910876333714, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036776021122932434, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036776021122932434, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.44.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2509554624557495, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23417504131793976, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22883416712284088, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20676998794078827, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11876363307237625, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11260132491588593, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13337986171245575, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12274666875600815, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12071185559034348, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10599590092897415, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1005258783698082, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0681501030921936, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.058872122317552567, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.057058267295360565, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05662688985466957, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0340878888964653, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029454000294208527, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0293276347219944, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026852626353502274, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.0265736635774374, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017960399389266968, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017969030886888504, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01736406981945038, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012034973129630089, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0340878888964653, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0340878888964653, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.44.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.29442957043647766, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.27480873465538025, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.26857590675354004, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2426101118326187, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13915732502937317, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1319883018732071, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1560361087322235, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14376753568649292, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1414681077003479, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12423880398273468, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11765214055776596, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07954694330692291, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06878475844860077, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06669686734676361, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0661999061703682, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.039719175547361374, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03402162343263626, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03387036547064781, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0309129748493433, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03059164620935917, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.020577283576130867, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.020137695595622063, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.019886396825313568, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012674558907747269, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.039719175547361374, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.039719175547361374, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.44.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3039171099662781, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.27470511198043823, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2615671455860138, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.2323983907699585, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.14232584834098816, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12969893217086792, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.17023557424545288, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1558900624513626, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14659100770950317, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12341156601905823, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11644356697797775, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.0871795266866684, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07507917284965515, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06881245970726013, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06729093194007874, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04404860734939575, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03604592755436897, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03543223813176155, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03227316588163376, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.031258877366781235, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.023929879069328308, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.023387055844068527, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02187875285744667, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.016093390062451363, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04404860734939575, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04404860734939575, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.45.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.15100796520709991, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1403561383485794, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1360754668712616, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12259174883365631, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07125432789325714, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06691933423280716, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0819317176938057, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0751403346657753, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07250270992517471, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06332548707723618, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.059997182339429855, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04170713573694229, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03597158193588257, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.034203268587589264, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.033774830400943756, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020875491201877594, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.017525410279631615, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.017356041818857193, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015892978757619858, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015622594393789768, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01088748686015606, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010624799877405167, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01030731201171875, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006800925359129906, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04170713573694229, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04170713573694229, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.45.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1370311826467514, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1273144781589508, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12323395907878876, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11113525182008743, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0646277591586113, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06057163327932358, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07477129250764847, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06852029263973236, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06574928760528564, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05747220292687416, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05456967279314995, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03808874264359474, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03279712051153183, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03102729842066765, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.030601462349295616, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01905464567244053, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015912296250462532, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015737004578113556, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014442279934883118, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014164823107421398, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009939146228134632, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009699747897684574, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009359410032629967, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006212438456714153, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03808874264359474, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03808874264359474, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.45.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.28980836272239685, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2697819471359253, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.26302993297576904, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2372969686985016, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.13711439073085785, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12956269085407257, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1553291380405426, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14224492013454437, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13943500816822052, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1219983696937561, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11555740982294083, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07919830828905106, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06803766638040543, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06567472964525223, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06511776149272919, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039534199982881546, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03344288095831871, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.033265795558691025, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.030281390994787216, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.029917506501078606, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020346425473690033, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0197654627263546, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019453519955277443, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012283611111342907, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039534199982881546, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039534199982881546, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.45.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25404253602027893, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1910669058561325, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.16521120071411133, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15023574233055115, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11324441432952881, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08202588558197021, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1448056697845459, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13236217200756073, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12277776747941971, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08474045246839523, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0816325768828392, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0746307447552681, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0637551099061966, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05499270185828209, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05274321883916855, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03736763820052147, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029092004522681236, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028437167406082153, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023590972647070885, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021989857777953148, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01994294300675392, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019903426989912987, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01708988845348358, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013742462731897831, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03736763820052147, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03736763820052147, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.45.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2452717423439026, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22880296409130096, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22343404591083527, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20187845826148987, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11621539294719696, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11011432111263275, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1305316537618637, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1201607957482338, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11812642216682434, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10363978147506714, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09819287061691284, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06670375168323517, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05763104557991028, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05581808462738991, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05539345741271973, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03335484117269516, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.028787296265363693, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02865958772599697, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026225654408335686, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02595466747879982, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017526375129818916, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.0175308920443058, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016927387565374374, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011700903996825218, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03335484117269516, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03335484117269516, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.45.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2943331003189087, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2747231721878052, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.26840683817863464, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.24254727363586426, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1392989456653595, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.13206833600997925, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15647375583648682, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14394071698188782, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14159558713436127, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12432822585105896, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11774668097496033, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07988683134317398, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06895686686038971, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06682636588811874, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06632585823535919, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03991975635290146, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.034237295389175415, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.034089647233486176, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.031153002753853798, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03082340955734253, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.02088802680373192, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02050136774778366, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.020183853805065155, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.013227074407041073, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03991975635290146, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03991975635290146, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.45.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.303180456161499, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2738777995109558, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2609502077102661, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.23167726397514343, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1419990211725235, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12944689393043518, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16861893236637115, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15525536239147186, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14626885950565338, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12299509346485138, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11582338809967041, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08615162968635559, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07466783374547958, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06857738643884659, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06708856672048569, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04319692403078079, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.035802897065877914, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.0352189764380455, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.032002195715904236, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.031014082953333855, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022773781791329384, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.023050250485539436, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020789803937077522, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01569046638906002, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04319692403078079, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04319692403078079, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.46.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1475263386964798, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13730525970458984, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1330450475215912, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12002938240766525, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06967909634113312, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0654468685388565, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08002954721450806, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07362658530473709, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0708547830581665, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.062004346400499344, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05881065875291824, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040759455412626266, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03522716835141182, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.033437687903642654, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03300692141056061, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02038576453924179, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01714465580880642, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01696486584842205, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015569537878036499, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015286584384739399, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01062049064785242, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010418577119708061, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010032779537141323, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006674290634691715, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040759455412626266, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040759455412626266, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.46.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13413287699222565, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12478426098823547, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1207742840051651, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1089741587638855, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06330551207065582, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05938924476504326, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07321155816316605, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0671842098236084, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06441797316074371, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05637393891811371, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.053594622761011124, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03726816922426224, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03215351700782776, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03041699342429638, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029998406767845154, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01866011507809162, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015626946464180946, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015452666208148003, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0142021169885993, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013930792920291424, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009750491939485073, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00956298690289259, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009190866723656654, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006183899473398924, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03726816922426224, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03726816922426224, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.46.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.29083383083343506, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2710813283920288, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2645125985145569, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2386547476053238, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.13776244223117828, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1303112506866455, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15592661499977112, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14288751780986786, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.14000236988067627, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12268444150686264, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11630181223154068, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07948122918605804, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06835299730300903, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06600330770015717, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06543178856372833, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03967437148094177, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03360198810696602, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0334206260740757, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.030472276732325554, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.030114153400063515, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020435437560081482, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019861683249473572, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01955151930451393, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012360217049717903, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03967437148094177, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03967437148094177, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.46.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24305768311023712, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19544366002082825, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.17499211430549622, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15414072573184967, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10980592668056488, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08992724120616913, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1393234133720398, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12766417860984802, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11722986400127411, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08577539771795273, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08144364506006241, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07171222567558289, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06165428087115288, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05354049429297447, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05149074271321297, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0359686017036438, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028820492327213287, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028131380677223206, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024295317009091377, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.022918684408068657, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019428756088018417, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020186856389045715, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01684550568461418, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014682827517390251, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0359686017036438, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0359686017036438, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.46.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.24115604162216187, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22488556802272797, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21956171095371246, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1981009840965271, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11429747194051743, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10828378051519394, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12836001813411713, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1182030737400055, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11621444672346115, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10183409601449966, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09639091044664383, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.065636046230793, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05669668689370155, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05492314323782921, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05449829250574112, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03282735496759415, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02834421955049038, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.028218206018209457, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02580367960035801, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.025537122040987015, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017273737117648125, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017292078584432602, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016683142632246017, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011575242504477501, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03282735496759415, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03282735496759415, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.46.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.29464593529701233, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.27490079402923584, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.26848769187927246, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2423994541168213, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13944587111473083, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1321532279253006, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1564687341451645, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14409086108207703, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14175879955291748, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12430236488580704, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1176011711359024, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07982544600963593, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06896918267011642, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06684793531894684, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06634339690208435, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03983631730079651, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.034114331007003784, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03396515175700188, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.030969971790909767, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03065013885498047, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.02067715860903263, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.020224470645189285, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.019968248903751373, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012770400382578373, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03983631730079651, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03983631730079651, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.46.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3078961670398712, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.27839624881744385, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2655662000179291, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.23573337495326996, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.14423979818820953, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.13174962997436523, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.171815425157547, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15732917189598083, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14868366718292236, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.1250050812959671, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11769313365221024, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08751179277896881, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07560219615697861, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06964346766471863, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06817746162414551, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04419386386871338, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03624527528882027, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.035682011395692825, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03237142041325569, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.03139370679855347, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02384931966662407, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.023157162591814995, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02197214961051941, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015601566061377525, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04419386386871338, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04419386386871338, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.47.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1468791663646698, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1366879940032959, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13262680172920227, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11952188611030579, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0692112073302269, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0650993287563324, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0790577381849289, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07290149480104446, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07040345668792725, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06159774586558342, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.058310430496931076, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040244363248348236, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03487971052527428, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03321439027786255, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.032814547419548035, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020120451226830482, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.017048699781298637, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016887711361050606, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015489090234041214, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01523008942604065, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010508034378290176, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01037534885108471, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009954715147614479, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006698877550661564, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040244363248348236, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040244363248348236, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.47.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13219718635082245, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12302568554878235, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11906386911869049, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10731131583452225, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.062269389629364014, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.058397457003593445, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07174071669578552, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06606535613536835, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06332412362098694, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0554267093539238, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.052564870566129684, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036498770117759705, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.031615156680345535, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029870249330997467, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029454492032527924, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018236003816127777, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015303052961826324, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015124436467885971, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013892218470573425, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013624508865177631, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009484080597758293, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009309266693890095, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008911626413464546, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005920290946960449, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036498770117759705, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.036498770117759705, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.47.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.289072185754776, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.26955682039260864, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2630902826786041, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.23746497929096222, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.13663046061992645, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1293298304080963, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15374858677387238, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14152652025222778, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13890960812568665, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12168239802122116, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.1150660589337349, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07826447486877441, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06767818331718445, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06542472541332245, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06488937884569168, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0390467494726181, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0332656055688858, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03309033438563347, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.030156945809721947, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.029812898486852646, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02006816491484642, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01957791857421398, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019222691655158997, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012087901122868061, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0390467494726181, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0390467494726181, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.47.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24887603521347046, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21320010721683502, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.199560284614563, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16817724704742432, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11493609100580215, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10067223012447357, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13775165379047394, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1264953315258026, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11977352201938629, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09146512299776077, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08493982255458832, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07112948596477509, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06112097576260567, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.055795084685087204, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05449533462524414, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03582329303026199, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029760967940092087, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029325157403945923, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025256985798478127, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024386519566178322, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019752753898501396, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02003304474055767, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018103070557117462, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014595183543860912, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03582329303026199, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03582329303026199, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.47.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.250090628862381, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23350436985492706, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22823122143745422, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20626647770404816, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1184292882680893, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11234220862388611, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1322590708732605, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12219741940498352, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1203545480966568, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10569587349891663, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1000036895275116, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06747657060623169, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05856960266828537, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05686204135417938, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.056463755667209625, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.033738136291503906, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029314007610082626, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029199285432696342, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026716480031609535, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02646181918680668, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017720751464366913, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017819423228502274, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01715836301445961, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011888468638062477, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.033738136291503906, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.033738136291503906, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.47.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.29414963722229004, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.27475327253341675, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.268664687871933, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.24293522536754608, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13916046917438507, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1320716291666031, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15533791482448578, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1434469372034073, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14138691127300262, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12422527372837067, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11749842762947083, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07913046330213547, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06863636523485184, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06666268408298492, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06619387120008469, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03953342139720917, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03400759771466255, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03387445583939552, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.030909590423107147, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.030609456822276115, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.020578643307089806, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02010173164308071, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.019928833469748497, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012664820067584515, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03953342139720917, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03953342139720917, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.47.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.29094287753105164, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25941774249076843, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.24461358785629272, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.216799795627594, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13501465320587158, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12115024775266647, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16333846747875214, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15050257742404938, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.13989770412445068, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.1158345490694046, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10927881300449371, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08340885490179062, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07214847207069397, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06512080878019333, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06338310241699219, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04171016439795494, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.033895496279001236, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03320107236504555, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030045410618185997, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02888224460184574, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022031215950846672, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.0219125896692276, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019731655716896057, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014583056792616844, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04171016439795494, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04171016439795494, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.48.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.14381468296051025, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13382086157798767, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12982937693595886, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1169845312833786, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06776458770036697, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06368470937013626, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07771801203489304, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07151956856250763, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06891200691461563, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.060329217463731766, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05724358558654785, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.039556413888931274, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03422314301133156, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.032513365149497986, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03210948407649994, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.019775403663516045, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01667756773531437, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016506923362612724, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015142261981964111, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014880512841045856, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010304497554898262, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010124399326741695, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00973520614206791, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006486683618277311, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.039556413888931274, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.039556413888931274, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.48.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1313004493713379, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12215100228786469, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11818819493055344, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.106622114777565, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06182402744889259, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.058011360466480255, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07145792990922928, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06563988327980042, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06287210434675217, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.055060241371393204, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05232670158147812, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03631213307380676, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.031394269317388535, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029679471626877785, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029267624020576477, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018162649124860764, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015229976736009121, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015058085322380066, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01384122483432293, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013576097786426544, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009484798647463322, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009292550384998322, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008927131071686745, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005962183233350515, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03631213307380676, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03631213307380676, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.48.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.29008594155311584, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2704737186431885, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2640467882156372, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.23827819526195526, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1372329443693161, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12991945445537567, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15450970828533173, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14212216436862946, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13948822021484375, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12229769676923752, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11572723835706711, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07879920303821564, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0680147185921669, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06576688587665558, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06522935628890991, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039342250674963, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03356165811419487, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03339659050107002, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.030466601252555847, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.030123528093099594, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02035583183169365, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01993943750858307, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019504887983202934, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012572361156344414, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039342250674963, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039342250674963, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.48.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23752006888389587, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.20493179559707642, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19343969225883484, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16894900798797607, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.109053835272789, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09571130573749542, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13003742694854736, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11859779804944992, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11423734575510025, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0900055542588234, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08420350402593613, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06728780269622803, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.058014512062072754, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05367039516568184, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05262681841850281, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03409603610634804, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029650947079062462, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029385268688201904, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02618192881345749, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02553265541791916, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019386611878871918, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0210562814027071, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018118254840373993, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.016666147857904434, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03409603610634804, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03409603610634804, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.48.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25338369607925415, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23661896586418152, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23126758635044098, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20900858938694, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12005030363798141, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11387915909290314, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13392651081085205, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12381459027528763, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12196217477321625, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10709449648857117, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10128933191299438, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06839060038328171, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.059370093047618866, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.057653363794088364, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05724282190203667, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034155331552028656, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029744653031229973, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029628189280629158, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027102366089820862, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026842841878533363, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017917055636644363, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01810246892273426, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017354493960738182, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012103916145861149, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034155331552028656, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034155331552028656, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.48.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.29519233107566833, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2757726013660431, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2697109580039978, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.24374842643737793, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1396973431110382, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1325952112674713, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15579821169376373, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14402979612350464, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14191143214702606, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12468378990888596, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.117924764752388, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07934314757585526, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06888996064662933, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06692434847354889, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0664551854133606, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.039605021476745605, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.034125808626413345, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.033992838114500046, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.03101327456533909, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03071095421910286, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.020537709817290306, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.020152375102043152, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.019886866211891174, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012670244090259075, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.039605021476745605, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.039605021476745605, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.48.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.29289788007736206, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.26252609491348267, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.24833078682422638, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.22006778419017792, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1364717334508896, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12303739041090012, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1649092584848404, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15138830244541168, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14110395312309265, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11748003214597702, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11082759499549866, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08447498083114624, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07289236783981323, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06602402776479721, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06431681662797928, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042778924107551575, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03468118980526924, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.034003909677267075, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030915847048163414, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.029801011085510254, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.023445378988981247, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022770535200834274, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.021295715123414993, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015743181109428406, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042778924107551575, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042778924107551575, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.49.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.15581253170967102, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.144999161362648, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1408640295267105, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12697875499725342, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0734788030385971, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0691697746515274, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08360107988119125, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07714170217514038, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07472508400678635, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06536012887954712, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06183174252510071, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04247179254889488, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03687448799610138, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.035222552716732025, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03482489660382271, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.021229982376098633, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01800079643726349, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.017846211791038513, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.016327083110809326, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01606728509068489, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011008105240762234, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010814910754561424, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01044820249080658, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006829868536442518, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04247179254889488, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04247179254889488, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.49.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.14051415026187897, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13074372708797455, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1268232762813568, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11432044953107834, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06626597046852112, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06227623671293259, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07596515864133835, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06986680626869202, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0674137994647026, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.058972377330064774, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05592215433716774, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.038662709295749664, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.033423684537410736, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03177819028496742, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03139231353998184, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.019325576722621918, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016266370192170143, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016108784824609756, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01476567517966032, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014512849971652031, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010050331242382526, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00982973724603653, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009500141255557537, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00623747706413269, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.038662709295749664, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.038662709295749664, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.49.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2950652837753296, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.27506935596466064, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2686532437801361, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.242420956492424, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.139557883143425, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.13212044537067413, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15676231682300568, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14435122907161713, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1418323814868927, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12432900071144104, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.1175747886300087, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0798802450299263, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06907190382480621, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06685805320739746, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06632740050554276, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039855167269706726, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.034013330936431885, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0338490791618824, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.030831025913357735, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.03049461916089058, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02054811269044876, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020026708021759987, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01973085105419159, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012410962954163551, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039855167269706726, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039855167269706726, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.49.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25191396474838257, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22056369483470917, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20807892084121704, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1790631264448166, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11636018007993698, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10374873876571655, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13915099203586578, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12836182117462158, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12122166156768799, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09650333225727081, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08948519080877304, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0716002807021141, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06182157248258591, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05628648027777672, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05492590367794037, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03600579872727394, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02958064153790474, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029112115502357483, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02565775252878666, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02476688288152218, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01935785636305809, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019410550594329834, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017574064433574677, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013509273529052734, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03600579872727394, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03600579872727394, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.49.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25654080510139465, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23957404494285583, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23416054248809814, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21148857474327087, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1216035932302475, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11531767249107361, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13590016961097717, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12545539438724518, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12355192005634308, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10845628380775452, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1025068387389183, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0694057047367096, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0601530522108078, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.058397114276885986, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05797310918569565, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034701474010944366, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030101552605628967, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029978737235069275, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027426306158304214, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027154404670000076, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018286630511283875, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01829879730939865, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01770707592368126, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012198301963508129, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034701474010944366, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.034701474010944366, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.49.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2980818748474121, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.27840885519981384, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.27223625779151917, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.24589453637599945, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.14107787609100342, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.13385912775993347, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15745072066783905, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1454342156648636, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14331838488578796, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1258816421031952, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11897104233503342, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.08021172881126404, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06958113610744476, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06758574396371841, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06710316240787506, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04003960266709328, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.034422531723976135, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03428557142615318, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.03126149997115135, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.030953966081142426, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.02076077274978161, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02028726041316986, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.020096300169825554, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012676624581217766, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04003960266709328, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04003960266709328, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.49.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3000624477863312, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2697696089744568, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.25631386041641235, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.2271556407213211, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13993944227695465, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12693041563034058, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16675055027008057, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15364450216293335, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14444510638713837, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12056977301836014, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11342146247625351, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08503259718418121, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07364825904369354, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06738816201686859, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06585107743740082, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04254178702831268, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03484806790947914, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03424268588423729, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030922427773475647, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.029879342764616013, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02237800322473049, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022056031972169876, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020320072770118713, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.014382450841367245, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04254178702831268, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04254178702831268, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.50.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.15287820994853973, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14211101830005646, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13785359263420105, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12414702028036118, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0720900446176529, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06770764291286469, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08242005109786987, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07600940763950348, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07336962968111038, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06407583504915237, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.060609638690948486, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.041984252631664276, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.036392051726579666, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03459513187408447, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.034171078354120255, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020998232066631317, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.017740663141012192, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.017570000141859055, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.016088031232357025, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015814118087291718, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010966545902192593, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010775165632367134, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010368634015321732, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006909847259521484, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.041984252631664276, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.041984252631664276, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.50.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13748739659786224, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12774237990379333, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12367531657218933, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11142527312040329, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06477333605289459, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06072978675365448, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07451978325843811, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06857389211654663, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06591231375932693, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05758824571967125, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.054542116820812225, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03791018947958946, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03281594067811966, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.031080253422260284, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03066730499267578, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018957655876874924, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01591763086616993, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01574719324707985, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01443486101925373, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014167928136885166, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009859084151685238, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009664786048233509, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009278502315282822, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006150311790406704, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03791018947958946, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03791018947958946, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.50.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.29185014963150024, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2717913091182709, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.26511824131011963, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.23917455971240997, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1379932165145874, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1304323822259903, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1553460657596588, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14298798143863678, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1403159499168396, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12274818867444992, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.116010382771492, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07908200472593307, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06836783140897751, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06606290489435196, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06551764905452728, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039442405104637146, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03357551991939545, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03340364992618561, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.030389854684472084, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.030031289905309677, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02028784155845642, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019736729562282562, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019423305988311768, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01212309766560793, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039442405104637146, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039442405104637146, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.50.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25750601291656494, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2058035135269165, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18972843885421753, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16032080352306366, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11320073902606964, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09620276093482971, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14139018952846527, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12599708139896393, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1199638620018959, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08803106844425201, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08323217183351517, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0707349106669426, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.062447551637887955, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.056670863181352615, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.055209141224622726, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036797571927309036, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03293727710843086, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.032562099397182465, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02871689386665821, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02783993072807789, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.022068044170737267, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02504296787083149, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020421374589204788, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.02099505439400673, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036797571927309036, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.036797571927309036, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.50.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.26004302501678467, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24253837764263153, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23702652752399445, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21393796801567078, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12332300841808319, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11689100414514542, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13783200085163116, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12727922201156616, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12533099949359894, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10988255590200424, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10380429029464722, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07046113908290863, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.061089351773262024, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0592661090195179, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05884481966495514, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035281501710414886, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.030632056295871735, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.030511239543557167, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.027903828769922256, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027627311646938324, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018690811470150948, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018726417794823647, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01809745468199253, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012612220831215382, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035281501710414886, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035281501710414886, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.50.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.3011763095855713, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2810702919960022, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2747121751308441, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.24812684953212738, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.14256900548934937, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1351672112941742, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15936093032360077, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14710167050361633, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14487482607364655, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12709678709506989, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.12008356302976608, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0812319815158844, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.07043326646089554, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06836920231580734, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0678822249174118, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04065246134996414, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03492461517453194, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03478151187300682, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.03172389045357704, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03140569478273392, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.021318241953849792, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.020733928307890892, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.02063494175672531, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.013169636018574238, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04065246134996414, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04065246134996414, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.50.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3061268627643585, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2762814164161682, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2631284296512604, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.23326225578784943, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1431930661201477, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.13039672374725342, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1695169061422348, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15653781592845917, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14754334092140198, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12372733652591705, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11631126701831818, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08664803206920624, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07520900666713715, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06909817457199097, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06761451065540314, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04343527555465698, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.036022983491420746, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.035435695201158524, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03212147578597069, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.031118927523493767, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022901376709342003, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02310967445373535, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020877810195088387, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015619534999132156, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04343527555465698, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04343527555465698, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.51.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1587999314069748, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14772604405879974, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14354060590267181, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1293315887451172, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07501672953367233, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07060974091291428, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08529678732156754, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07873418927192688, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07629171013832092, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06666968762874603, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06302959471940994, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0433746837079525, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.037656333297491074, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03596510365605354, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03555493429303169, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02165841870009899, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01836223155260086, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.018211524933576584, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01663641259074211, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016374284401535988, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011204035952687263, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.011007068678736687, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010631054639816284, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006909613963216543, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0433746837079525, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0433746837079525, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.51.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1437791883945465, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1337367594242096, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12975923717021942, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11699631810188293, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06789440661668777, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06384976208209991, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07761316001415253, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07149666547775269, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0690668523311615, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06037377566099167, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.057162947952747345, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.039506945759058, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03420522063970566, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03257656469941139, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.032177843153476715, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.019737180322408676, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016654489561915398, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016498491168022156, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015099050477147102, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01484854519367218, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010234730318188667, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010034698992967606, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009692052379250526, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006349050905555487, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.039506945759058, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.039506945759058, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.51.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.29870617389678955, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2783285975456238, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.271610289812088, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2450844645500183, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1414366215467453, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.13379044830799103, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15889392793178558, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1463679075241089, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.143763929605484, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12584653496742249, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11884176731109619, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.08091335743665695, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0700191855430603, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06772907078266144, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06718763709068298, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0403735376894474, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03442005440592766, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03424666449427605, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.03117358312010765, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.03082018531858921, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02075319178402424, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020210593938827515, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01991467922925949, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012428274378180504, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0403735376894474, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0403735376894474, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.51.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2658490240573883, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22392378747463226, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20704346895217896, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18329137563705444, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12075057625770569, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1029028445482254, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1488357037305832, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13678981363773346, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1282542198896408, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09931276738643646, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09401154518127441, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07667271792888641, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06611410528421402, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05883194878697395, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05699067562818527, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038573287427425385, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03146091848611832, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030913114547729492, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02735062874853611, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.026157483458518982, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.021136518567800522, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02160654589533806, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01888260990381241, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01568295620381832, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038573287427425385, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038573287427425385, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.51.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.26237574219703674, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.24472276866436005, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23903165757656097, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.21581387519836426, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12438720464706421, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11785071343183517, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13913103938102722, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1284247487783432, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12642499804496765, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.11078773438930511, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10477256774902344, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07113151252269745, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.061604779213666916, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.059766001999378204, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05932685360312462, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035584330558776855, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03083888627588749, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03071478195488453, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028069622814655304, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.027791989967226982, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018804460763931274, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018791837617754936, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.018203511834144592, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012575400993227959, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035584330558776855, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.035584330558776855, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.51.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.3033130168914795, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2830061614513397, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2765612006187439, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2496967911720276, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.14363397657871246, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1361628919839859, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1604372262954712, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1482352614402771, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1459730863571167, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12794101238250732, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.12083864957094193, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.08184610307216644, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.07094056904315948, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0688391849398613, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06834723800420761, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04083715006709099, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03509046882390976, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.034948695451021194, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.03183542191982269, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03151119127869606, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.021212883293628693, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02071998082101345, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.0205168928951025, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012989738956093788, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04083715006709099, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04083715006709099, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.51.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3072236180305481, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2774339020252228, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2641366124153137, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.23413783311843872, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1438186764717102, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1309545934200287, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.17128024995326996, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15747301280498505, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14818468689918518, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.1243719533085823, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11693413555622101, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08766528964042664, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07572270929813385, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06946925073862076, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06793348491191864, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04419993236660957, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03630917891860008, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.035702165216207504, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03242097049951553, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.03139859437942505, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.023857515305280685, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02343696728348732, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.0218157097697258, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01598331145942211, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04419993236660957, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04419993236660957, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.52.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.16421209275722504, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15260173380374908, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14822916686534882, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.13356010615825653, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07754522562026978, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07296732813119888, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0883944183588028, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.08138217777013779, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07894826680421829, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06889976561069489, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06515230238437653, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0450064092874527, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.038947027176618576, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03719479218125343, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.036777790635824203, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.022500289604067802, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.019021667540073395, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01886865682899952, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01724248006939888, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016971508041024208, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011706788092851639, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.011447629891335964, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.011105223558843136, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0072597782127559185, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0450064092874527, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0450064092874527, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.52.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.14888745546340942, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1384207308292389, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1343255192041397, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12102711200714111, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07037003338336945, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0661415383219719, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08059492707252502, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07409593462944031, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0715927705168724, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06254854053258896, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05924609676003456, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04107910022139549, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03547656536102295, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03375939652323723, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03335012495517731, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020532315596938133, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.017277870327234268, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.017115198075771332, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01566469296813011, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015397799201309681, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010689027607440948, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01043242309242487, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010107865557074547, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006621478125452995, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04107910022139549, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04107910022139549, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.52.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.3006208539009094, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.27998292446136475, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2731834650039673, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.24637332558631897, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1423199474811554, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.13460350036621094, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15991227328777313, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1472846120595932, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.14466285705566406, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12651346623897552, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11951043456792831, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.08142207562923431, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.07046255469322205, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06816603243350983, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06760263442993164, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.040625084191560745, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.034650519490242004, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.034482359886169434, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.03135533258318901, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0310008954256773, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02090710960328579, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020368194207549095, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020053580403327942, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012550665065646172, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.040625084191560745, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.040625084191560745, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.52.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.26810428500175476, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21900984644889832, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20045174658298492, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.17629316449165344, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12192479521036148, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10161140561103821, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1489678919315338, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1368829607963562, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1293160617351532, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09617514163255692, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09037574380636215, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07653575390577316, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06590169668197632, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05909355729818344, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05740626901388168, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038351256400346756, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.031200356781482697, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03070063889026642, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.026192136108875275, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0250365249812603, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020647389814257622, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020819993689656258, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018476171419024467, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01458478532731533, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038351256400346756, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038351256400346756, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.52.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.255057156085968, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23771414160728455, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23221835494041443, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20953789353370667, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1209842711687088, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11458005011081696, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13527631759643555, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1249576210975647, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12298546731472015, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10768545418977737, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1017487421631813, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06913390010595322, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05993405357003212, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05812574177980423, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.057697732001543045, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03457392379641533, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029982926324009895, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029856259003281593, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02726985141634941, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026998288929462433, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01821226067841053, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01826080121099949, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01761636510491371, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012196958996355534, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03457392379641533, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03457392379641533, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.52.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.30313923954963684, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2827364504337311, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.276233971118927, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2492886185646057, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.14370112121105194, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.13612018525600433, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1605585366487503, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14832043647766113, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14607509970664978, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12796153128147125, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1209074929356575, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.08199718594551086, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.07104457914829254, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06892707943916321, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06842207908630371, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040937479585409164, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.035277459770441055, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.035134464502334595, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.032023780047893524, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.031704582273960114, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.021384021267294884, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02104824222624302, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.020681610330939293, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.013503704220056534, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040937479585409164, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040937479585409164, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.52.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.30721819400787354, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.27750876545906067, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2645973861217499, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.23441645503044128, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1438743621110916, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1312343329191208, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16987566649913788, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15696382522583008, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14814604818820953, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12433250993490219, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11666601896286011, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.0867784395813942, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07538045942783356, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06943675130605698, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06797222048044205, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04353071376681328, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.036152079701423645, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03558065742254257, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03222037851810455, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.031246231868863106, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.023020094260573387, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.023105230182409286, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.021086549386382103, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015558718703687191, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04353071376681328, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04353071376681328, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.53.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.15972797572612762, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1484503149986267, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1440877914428711, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12971702218055725, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07548771053552628, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07095614075660706, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08594361692667007, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07936276495456696, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07680913805961609, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06699556112289429, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06327052414417267, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04370469972491264, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03796284273266792, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.036206237971782684, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.035781487822532654, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.021845383569598198, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.018510041758418083, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.018348636105656624, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01675553061068058, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016486305743455887, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011328422464430332, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.011142921634018421, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010740483179688454, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.007048164959996939, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04370469972491264, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04370469972491264, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.53.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1442079246044159, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13401809334754944, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12997666001319885, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11701645702123642, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06815198808908463, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06400109082460403, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07796996086835861, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0718553364276886, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06933055818080902, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06049758568406105, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.057223446667194366, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03966430202126503, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.034381456673145294, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03270013630390167, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.032296981662511826, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.019846517592668533, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016750603914260864, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01659177616238594, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015174970962107182, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01491475012153387, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010315543971955776, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01013929396867752, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009758231230080128, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006465004291385412, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03966430202126503, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03966430202126503, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.53.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.30168241262435913, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2808505594730377, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.274105042219162, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.24699319899082184, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1428416520357132, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.13504913449287415, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.16043348610401154, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1478734165430069, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.14522214233875275, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1268732100725174, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11982618272304535, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.08167338371276855, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.07072776556015015, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06843435019254684, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06786683201789856, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.040756333619356155, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03476620838046074, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03459341451525688, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.031431373208761215, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.031071525067090988, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020944654941558838, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02039235644042492, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020104479044675827, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01251080259680748, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.040756333619356155, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.040756333619356155, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.53.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25415360927581787, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21172651648521423, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19352182745933533, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.17261099815368652, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11572911590337753, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09692582488059998, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14418739080429077, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13279612362384796, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12244065850973129, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09429534524679184, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08939184248447418, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07415138930082321, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06400606781244278, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05622868984937668, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054248321801424026, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03720780834555626, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029921509325504303, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.029231786727905273, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025785306468605995, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024491455405950546, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020086120814085007, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020457260310649872, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017635157331824303, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014497305266559124, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03720780834555626, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03720780834555626, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.53.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.25100457668304443, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23392030596733093, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22848108410835266, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20594075322151184, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11913138628005981, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11279994249343872, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1333162486553192, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12305024266242981, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12110830843448639, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10593590885400772, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10003948211669922, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06821222603321075, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05906622111797333, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05727309361100197, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.056845102459192276, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03412916883826256, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029609795659780502, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02948514185845852, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026932954788208008, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02665787562727928, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018100056797266006, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.018123993650078773, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017506996169686317, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012223058380186558, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03412916883826256, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03412916883826256, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.53.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.30334386229515076, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.28280124068260193, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.276294082403183, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2492097020149231, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.14376671612262726, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1361895054578781, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.16059711575508118, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14838510751724243, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14612704515457153, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1279020458459854, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.12071092426776886, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.08188706636428833, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.07102110236883163, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06890273094177246, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0684049129486084, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040861934423446655, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.035115908831357956, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03497324138879776, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.031824950128793716, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03149953857064247, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.021189920604228973, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02073889598250389, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.02048460952937603, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012987877242267132, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040861934423446655, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040861934423446655, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.53.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3134709894657135, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2834426164627075, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2707427442073822, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.23987160623073578, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.14696329832077026, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1343233734369278, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.17393893003463745, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15959535539150238, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1512582153081894, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12700095772743225, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11920240521430969, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08851102739572525, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07663051784038544, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.07086105644702911, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06943873316049576, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.044621117413043976, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03674286976456642, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.036200299859046936, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.032710205763578415, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.03175880387425423, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.023975662887096405, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.023242155089974403, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.02211596630513668, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015440856106579304, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.044621117413043976, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.044621117413043976, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.54.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.16321329772472382, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15159830451011658, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14711841940879822, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1323433220386505, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07712184637784958, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0724719688296318, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08793847262859344, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.08109001070261002, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07847455888986588, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06840649992227554, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06455732882022858, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04473026469349861, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.038804493844509125, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.036993831396102905, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03656260296702385, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.022356081753969193, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.018923230469226837, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.018754223361611366, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.017118992283940315, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01683850586414337, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011605018749833107, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.011406018398702145, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01099187321960926, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.007224712520837784, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04473026469349861, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04473026469349861, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.54.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1460331380367279, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13565316796302795, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13142788410186768, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1182800829410553, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06900925934314728, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06474462896585464, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07923469692468643, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07289579510688782, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07021963596343994, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.061228495091199875, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05794435366988182, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040363747626543045, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.034900590777397156, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.033127229660749435, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03270291909575462, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02018076926469803, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016971023753285408, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016797572374343872, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015368700958788395, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015094874426722527, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010508875362575054, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010293669067323208, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009912857785820961, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006559452507644892, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040363747626543045, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.040363747626543045, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.54.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.3069276809692383, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2856762111186981, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2786672115325928, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.2510204613208771, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.14543381333351135, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.13744109869003296, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.16345389187335968, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1505611091852188, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.14785662293434143, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1290798783302307, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.12189280241727829, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.08329443633556366, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.07201679795980453, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06964018195867538, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06906406581401825, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04155779629945755, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03541196137666702, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.035237867385149, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.03201134875416756, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.031644754111766815, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.021391158923506737, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020831771194934845, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020516082644462585, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012843603268265724, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04155779629945755, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04155779629945755, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.54.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.26692408323287964, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22789892554283142, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21146994829177856, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18273252248764038, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12104344367980957, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10523378849029541, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1506531685590744, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13804765045642853, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12858153879642487, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09992380440235138, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09391950815916061, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07730347663164139, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06636791676282883, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05856085196137428, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05659354478120804, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038769666105508804, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.030736388638615608, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.030099790543317795, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.026567310094833374, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02524624951183796, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02078961953520775, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02048414945602417, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01826956868171692, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01394323818385601, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038769666105508804, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038769666105508804, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.54.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2464858740568161, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22962141036987305, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2242058366537094, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20215128362178802, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11703503131866455, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11077643930912018, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1309749335050583, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12091323733329773, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11899413168430328, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1040634885430336, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0982215404510498, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06697619706392288, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05805683881044388, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05628253519535065, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05585511401295662, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03350801765918732, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029106207191944122, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0289867352694273, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02647210843861103, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026201406493782997, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017713429406285286, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017836106941103935, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017122959718108177, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012053564190864563, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03350801765918732, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03350801765918732, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.54.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.3006840646266937, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.28028637170791626, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.27385586500167847, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.24685105681419373, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.14258411526679993, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.13504894077777863, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.15937748551368713, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14722849428653717, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14495070278644562, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12683473527431488, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11970566213130951, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.08135675638914108, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.07052750885486603, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06840477883815765, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06789761036634445, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04063144326210022, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.034935686737298965, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03479025512933731, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.03167330473661423, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.03134731575846672, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.021159404888749123, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02075682394206524, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.020454246550798416, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.013176629319787025, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04063144326210022, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.04063144326210022, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.54.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3025071918964386, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.27318286895751953, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2597930431365967, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.23012548685073853, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.14169490337371826, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1289517730474472, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16940495371818542, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15568006038665771, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14604727923870087, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.12249407917261124, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11519066989421844, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08668546378612518, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07483387738466263, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06843031942844391, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06685352325439453, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04376281797885895, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03567894548177719, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03504360467195511, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.031820714473724365, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.030767451971769333, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.023661186918616295, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022960945963859558, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.0215610284358263, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01549066137522459, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04376281797885895, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04376281797885895, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.55.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1596810221672058, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1482909470796585, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14375875890254974, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12924182415008545, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0754474475979805, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0708358883857727, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08626365661621094, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07954102754592896, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07679690420627594, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0668838694691658, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06321188062429428, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04390711337327957, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03809285908937454, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03621644154191017, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03576560690999031, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02195742353796959, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.018528098240494728, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.018356269225478172, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.016754936426877975, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016464633867144585, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011391385458409786, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.011186922900378704, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010761932469904423, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0070937746204435825, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04390711337327957, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04390711337327957, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.55.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1439903825521469, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13362669944763184, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12951262295246124, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.116480752825737, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0680139884352684, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06376932561397552, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07796911150217056, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0718550905585289, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06921891868114471, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.060293085873126984, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.057041604071855545, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03970532864332199, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03439844399690628, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03264952078461647, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03223327174782753, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01984807848930359, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016726097092032433, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01655743643641472, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01513279415667057, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014861272647976875, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01032465323805809, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010140928439795971, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00974609237164259, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0064671095460653305, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03970532864332199, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03970532864332199, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.55.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.30059728026390076, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2796422243118286, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2727760076522827, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.24557721614837646, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.14239150285720825, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.13452835381031036, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.16001258790493011, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14754696190357208, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.14478421211242676, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1263384371995926, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11915602535009384, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.08150585740804672, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.07058066874742508, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06820119172334671, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0676388144493103, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04067094624042511, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.034655869007110596, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.034472689032554626, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.031308021396398544, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.030936224386096, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020901765674352646, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020360326394438744, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020041702315211296, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012490559369325638, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04067094624042511, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04067094624042511, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.55.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2605281472206116, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22278887033462524, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20791122317314148, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.17816990613937378, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11917698383331299, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1038428395986557, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.146211177110672, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1324513852596283, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12489966303110123, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09661494195461273, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09084982424974442, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0750221461057663, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06429477781057358, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.058278851211071014, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05679210275411606, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038133405148983, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.031750958412885666, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.031292617321014404, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02756172977387905, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02662499062716961, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.021773304790258408, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.022222356870770454, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020024485886096954, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01697179488837719, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038133405148983, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.038133405148983, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.55.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2408609837293625, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22426427900791168, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21895970404148102, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19742757081985474, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1143493503332138, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10819678008556366, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12801624834537506, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11816105246543884, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11628836393356323, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10161623358726501, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09590707719326019, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06552375108003616, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05673103779554367, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05498422682285309, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05458131060004234, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0327497199177742, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.028439778834581375, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02831861563026905, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.025853265076875687, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.025590049102902412, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.017315339297056198, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017430465668439865, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01673855446279049, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01177979726344347, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0327497199177742, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0327497199177742, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.55.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2966822385787964, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2764979302883148, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.27002638578414917, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.24352392554283142, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1407785415649414, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.13326182961463928, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1575174182653427, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14535211026668549, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.14310775697231293, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12515656650066376, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1181444600224495, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.08046182245016098, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06966141611337662, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06755763292312622, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06705189496278763, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040194071829319, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03460700437426567, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.034461792558431625, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.03138760104775429, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.031063303351402283, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.021068286150693893, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.02069883421063423, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.02037693001329899, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.013329180888831615, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040194071829319, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.040194071829319, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.55.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.3069610893726349, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2761725187301636, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.26204922795295715, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.2321585714817047, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.14349964261054993, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.13003559410572052, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1725134551525116, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15829963982105255, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14824046194553375, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.1238727793097496, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11655371636152267, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08807121217250824, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07615193724632263, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06932682543992996, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06763986498117447, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.0445164293050766, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03619266301393509, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03553225100040436, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03226903825998306, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.031152494251728058, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.024193454533815384, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.02342928573489189, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.022054776549339294, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015870649367570877, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.0445164293050766, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.0445164293050766, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.56.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.15865139663219452, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14718346297740936, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14258569478988647, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12813067436218262, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07497816532850266, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07025416195392609, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08570916950702667, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07917802780866623, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07630567997694016, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06636650860309601, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06261464208364487, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.043584588915109634, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03787501901388168, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03596600890159607, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03551067039370537, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02178165316581726, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.018420038744807243, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01823648065328598, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01664864458143711, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016350630670785904, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011325147934257984, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.011163124814629555, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010702230967581272, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.007111820857971907, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.043584588915109634, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.043584588915109634, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.56.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.14096680283546448, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13075850903987885, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12646758556365967, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11363112181425095, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06659290194511414, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06229858100414276, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07671033591032028, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07065944373607635, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06777912378311157, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05896645784378052, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05574526637792587, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03906191885471344, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03385356441140175, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.031983450055122375, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03152843564748764, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01953783445060253, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016429755836725235, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016241900622844696, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014871419407427311, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014577889814972878, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01019560918211937, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010053505189716816, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009583884850144386, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006474113557487726, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03906191885471344, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03906191885471344, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.56.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.3022473156452179, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2810363173484802, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.27406415343284607, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.24659813940525055, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.14319299161434174, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1352531611919403, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.16105708479881287, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1484023928642273, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1456766128540039, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12698239088058472, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11981623619794846, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.08198323845863342, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.07097664475440979, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06858120858669281, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0680004134774208, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04092658311128616, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03485843166708946, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03467792272567749, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.03148492053151131, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.03111225552856922, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.021018709987401962, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020484289154410362, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020134562626481056, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012586662545800209, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04092658311128616, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04092658311128616, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.56.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24934643507003784, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21388238668441772, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19783438742160797, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1676248162984848, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11539661139249802, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09958209097385406, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14132770895957947, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13022610545158386, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11990442126989365, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09134595096111298, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08602997660636902, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07233481854200363, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.062435828149318695, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.055648528039455414, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.053946468979120255, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03617250174283981, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029011333361268044, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.028323940932750702, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024241402745246887, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02305694855749607, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019111815840005875, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01897764764726162, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01690702512860298, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012635125778615475, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03617250174283981, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03617250174283981, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.56.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2368694692850113, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2204909473657608, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21524076163768768, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19395872950553894, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11250464618206024, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10640905052423477, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1258169710636139, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1163177564740181, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11443468183279037, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09992852807044983, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09428395330905914, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06430917978286743, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.055833958089351654, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.054097313433885574, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05368705466389656, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.032157834619283676, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.027964327484369278, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.027843227609992027, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.025397324934601784, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.025136979296803474, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01691150665283203, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017118439078330994, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016335496678948402, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011532343924045563, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.032157834619283676, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.032157834619283676, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.56.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.28602391481399536, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2663522958755493, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.26010891795158386, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.23440368473529816, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13565361499786377, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12836109101772308, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.151848703622818, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.14010761678218842, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.13793151080608368, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.12053035199642181, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.11370378732681274, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.07751226425170898, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06715266406536102, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06510821729898453, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.06462264806032181, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038738347589969635, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.03331354632973671, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.033170148730278015, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.030181260779500008, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02986760437488556, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.02023448795080185, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01987137272953987, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01955404318869114, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.012726415880024433, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038738347589969635, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.038738347589969635, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.56.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.29510417580604553, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.26490288972854614, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.2511257231235504, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.2223368138074875, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13783212006092072, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12470443546772003, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16541585326194763, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15227088332176208, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14250291883945465, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11871206760406494, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11160704493522644, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08444508910179138, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07313629984855652, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06654999405145645, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06492714583873749, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04232660308480263, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03468811511993408, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03404569998383522, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03084651194512844, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02976001799106598, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02247718721628189, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022375017404556274, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020360149443149567, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015035643242299557, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04232660308480263, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.04232660308480263, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.57.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1551939696073532, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14398331940174103, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13950476050376892, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12537480890750885, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07330813258886337, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0687209963798523, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08381414413452148, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07744308561086655, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07462511956691742, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06492649018764496, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06126846745610237, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04267023131251335, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.037071142345666885, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.035183511674404144, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.034734562039375305, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02132773958146572, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.018029319122433662, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.017847836017608643, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.016297483816742897, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016003888100385666, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01109878160059452, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01093234308063984, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01048113964498043, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006974467542022467, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04267023131251335, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04267023131251335, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.57.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13699282705783844, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1271296590566635, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12297146767377853, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11047221720218658, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06469225883483887, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06056445837020874, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07443954795598984, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06862557679414749, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06585519760847092, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05730380862951279, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05414509028196335, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03789447620511055, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03284928575158119, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03107280470430851, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.030642511323094368, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018959302455186844, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015952061861753464, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01577945612370968, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014438296668231487, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014164287596940994, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009897148236632347, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009754977189004421, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00931429024785757, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006282915826886892, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03789447620511055, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03789447620511055, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.57.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2997848391532898, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2788127064704895, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2719499170780182, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.24479180574417114, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.14210154116153717, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1342056691646576, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.15972961485385895, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14722205698490143, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.14451950788497925, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12603791058063507, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11884936690330505, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.08137127757072449, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.07044374197721481, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06808202713727951, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0675075426697731, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04061950370669365, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03460202366113663, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03443191200494766, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.03125389665365219, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0308876633644104, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02088058739900589, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.020329829305410385, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.020017830654978752, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012505611404776573, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04061950370669365, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.04061950370669365, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.57.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.25350305438041687, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2108759582042694, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.196158766746521, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16953375935554504, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11649274826049805, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09943121671676636, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.13804125785827637, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12677429616451263, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1220257505774498, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08997012674808502, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0853305459022522, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0707755759358406, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06094253063201904, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.056248512119054794, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0550997368991375, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03541186451911926, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.029466331005096436, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02915993705391884, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024273760616779327, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023475375026464462, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01891113445162773, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019095463678240776, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017394084483385086, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013241022825241089, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03541186451911926, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03541186451911926, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.57.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2321362942457199, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.21604490280151367, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21084149181842804, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18999016284942627, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11023231595754623, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10423877090215683, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12318967282772064, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11396140605211258, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11212663352489471, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09786899387836456, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09224050492048264, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06289483606815338, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05467233061790466, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05297388881444931, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.052569430321455, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.031426265835762024, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.027317818254232407, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.027200698852539062, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02479606866836548, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.024534909054636955, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01644224114716053, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01664075441658497, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015882613137364388, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011094782501459122, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.031426265835762024, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.031426265835762024, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.57.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2754327952861786, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2563740909099579, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.25040963292121887, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2256627380847931, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.13059966266155243, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.12356191873550415, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.14597997069358826, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.13490016758441925, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.13283990323543549, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1159910336136818, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.10935482382774353, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.074428990483284, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06459072232246399, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.06262151896953583, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.062145479023456573, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03714621812105179, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.031939420849084854, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.03180883079767227, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.028905004262924194, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.028601624071598053, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.019232751801609993, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01890183798968792, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.018567539751529694, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011890939436852932, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03714621812105179, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03714621812105179, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.57.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.300253301858902, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2679959237575531, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.25373756885528564, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.2244376838207245, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13979314267635345, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1260393112897873, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16811074316501617, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15457512438297272, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14496402442455292, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11993991583585739, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.11277113854885101, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08577395975589752, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07418796420097351, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06747780740261078, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06582365930080414, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042962342500686646, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.0351182222366333, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03448503464460373, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.03112318180501461, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.030006401240825653, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022751450538635254, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022596783936023712, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020541569218039513, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.015083257108926773, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042962342500686646, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042962342500686646, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.58.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1491905152797699, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13827012479305267, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13372370600700378, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11998724192380905, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07037413120269775, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06581494212150574, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08088207244873047, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07473747432231903, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07164453715085983, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06226043030619621, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05871414393186569, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.041140731424093246, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03576929122209549, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03378629684448242, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.033311426639556885, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020584678277373314, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.017330780625343323, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.017132185399532318, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015658071264624596, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015347275882959366, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010730618610978127, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01056897733360529, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010092361830174923, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006753204390406609, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.041140731424093246, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.041140731424093246, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.58.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12897635996341705, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11951064318418503, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11530312895774841, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1034822016954422, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.060785550624132156, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.056726545095443726, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07035109400749207, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06494550406932831, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.061910852789878845, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05376800149679184, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05078766494989395, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03574822098016739, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.031088802963495255, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02920142188668251, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028747057542204857, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01789318397641182, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015009012073278427, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014812711626291275, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01357237808406353, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01327851228415966, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009357795119285583, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009230541996657848, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008754427544772625, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0059413001872599125, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03574822098016739, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03574822098016739, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.58.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2943893373012543, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.27379822731018066, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.26697105169296265, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.24018147587776184, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.139557883143425, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1318313330411911, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1571362018585205, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1446695476770401, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.14194166660308838, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12372554838657379, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.116689532995224, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.08007717877626419, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0692257210612297, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06687744706869125, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0663224384188652, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0399625301361084, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.034061066806316376, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.03388053551316261, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.03076006844639778, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.030391797423362732, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.02058243378996849, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.02011077105998993, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019715582951903343, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01249669585376978, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0399625301361084, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0399625301361084, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.58.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24258893728256226, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19896504282951355, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18337972462177277, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15467235445976257, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1093374714255333, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09396807849407196, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1334298998117447, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.12245509028434753, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11714804917573929, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08474753797054291, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08021320402622223, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06864731758832932, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.059030622243881226, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05306089296936989, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0515739805996418, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03444601595401764, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02822648547589779, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027876760810613632, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023504571989178658, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.022501695901155472, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.018598871305584908, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01906423084437847, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016702814027667046, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01368219405412674, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03444601595401764, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03444601595401764, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.58.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21767781674861908, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20252415537834167, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1976691484451294, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1780301183462143, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10336733609437943, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09771930426359177, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11566416919231415, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10689498484134674, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10515023022890091, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09174223989248276, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08647428452968597, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05909828841686249, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.051307156682014465, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04968898743391037, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04930786415934563, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029557839035987854, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02568238042294979, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025572901591658592, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023319300264120102, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023076700046658516, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01555005181580782, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015721825882792473, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01500997506082058, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010583269409835339, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029557839035987854, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029557839035987854, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.58.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2562206983566284, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.23850688338279724, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.23288355767726898, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20980460941791534, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.12153163552284241, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.11495769768953323, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.13598322868347168, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.12555855512619019, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.12360711395740509, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1078779324889183, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.1017058938741684, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06944139301776886, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.06014498323202133, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.058288685977458954, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0578492172062397, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03466860204935074, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.029786603525280952, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.029655834659934044, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.026956897228956223, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.026675565168261528, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.018061188980937004, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.017707057297229767, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.017438022419810295, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011245359666645527, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03466860204935074, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03466860204935074, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.58.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2914637625217438, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.25946953892707825, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.24534615874290466, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.21658997237682343, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.13579444587230682, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.12208529561758041, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.16410823166370392, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.15021315217018127, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.14082330465316772, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.11605856567621231, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.10908448696136475, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.08379272371530533, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.07230643182992935, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06577063351869583, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.06416071951389313, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042143888771533966, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03468148037791252, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.034075140953063965, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.030794033780694008, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02971954457461834, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.022649938240647316, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022872639819979668, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.020492445677518845, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.016014348715543747, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042143888771533966, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.042143888771533966, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.59.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1396876573562622, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12937280535697937, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12486881017684937, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11206722259521484, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06585089862346649, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06145825982093811, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07612083852291107, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0701974630355835, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06706941872835159, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05819419026374817, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.055008187890052795, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.038783762603998184, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.033654000610113144, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03165920823812485, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03117874078452587, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.019420571625232697, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016335029155015945, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016131527721881866, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014777574688196182, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014464294537901878, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010210497304797173, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010129899717867374, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009572160430252552, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006648460868746042, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.038783762603998184, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.038783762603998184, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.59.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12389867007732391, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11474849283695221, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11042498797178268, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09909588098526001, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.058368366211652756, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.054319191724061966, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06808532774448395, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06276216357946396, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.059455737471580505, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05162639915943146, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.048796992748975754, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03465098887681961, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030079105868935585, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028070051223039627, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.027587110176682472, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017361557111144066, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014490550383925438, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014277488924562931, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01312077697366476, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012808216735720634, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009147433564066887, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009032883681356907, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008502224460244179, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005922743119299412, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03465098887681961, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03465098887681961, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.59.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2798338830471039, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.25972479581832886, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2529636025428772, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.22728390991687775, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1324177384376526, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12484150379896164, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.14944183826446533, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13769207894802094, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13475243747234344, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1171761304140091, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11055552214384079, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07617580145597458, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06589212268590927, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06348949670791626, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06292036175727844, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03802487254142761, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03235681727528572, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.032168708741664886, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.029184240847826004, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.028814977034926414, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.019625844433903694, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019190365448594093, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018763843923807144, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011972426436841488, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03802487254142761, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03802487254142761, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.59.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1671823263168335, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13964539766311646, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12991207838058472, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10925563424825668, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07663594186306, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06591205298900604, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09206609427928925, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0839117243885994, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07975015044212341, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05956409126520157, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0553254596889019, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04730721563100815, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04148557782173157, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.038213733583688736, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.037416186183691025, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.024152349680662155, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.021980073302984238, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.021745998412370682, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.019078558310866356, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.018613070249557495, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014128792099654675, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01646333560347557, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013175551779568195, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013776126317679882, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04148557782173157, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04148557782173157, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.59.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19909651577472687, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18521174788475037, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1807449609041214, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16279159486293793, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09456256031990051, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08936188369989395, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10583329945802689, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09780149906873703, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0961783304810524, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08390093594789505, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07915376126766205, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05420805141329765, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.047005947679281235, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0455252081155777, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0451662503182888, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027142783626914024, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02364351600408554, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0235429760068655, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0214930921792984, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021266842260956764, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014410980977118015, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014641867950558662, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013920859433710575, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010061176493763924, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0455252081155777, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0455252081155777, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.59.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2331758737564087, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.21701331436634064, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21184010803699493, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19076552987098694, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1106247529387474, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.1046130508184433, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12385105341672897, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11432898789644241, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11251859366893768, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09815564751625061, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09256355464458466, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06328706443309784, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05482006072998047, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05311866104602814, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05271812155842781, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.031618695706129074, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.027264971286058426, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.027147434651851654, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.024701746180653572, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02444331906735897, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016564393416047096, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016391398385167122, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016000432893633842, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010665382258594036, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.031618695706129074, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.031618695706129074, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.59.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2711670994758606, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.23840586841106415, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.22458957135677338, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.19829043745994568, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.12581530213356018, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.1120072528719902, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.15260785818099976, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.13937686383724213, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1310756802558899, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.10654403269290924, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.1003802940249443, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.07833389192819595, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0673203244805336, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.06116098538041115, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05962710082530975, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03953511267900467, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03261282294988632, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.03207852318882942, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.028889399021863937, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.027888722717761993, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.02172129787504673, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.021979115903377533, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01975012943148613, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01593177206814289, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03953511267900467, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03953511267900467, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.60.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13490347564220428, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12510190904140472, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12071116268634796, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10832244157791138, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06361418217420578, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.059392042458057404, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07385436445474625, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06800643354654312, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0647689700126648, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05631434917449951, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05330478772521019, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.037622611969709396, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03257615864276886, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.030602624639868736, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.030128471553325653, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.018859844654798508, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015791255980730057, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015584181062877178, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014300210401415825, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01399487815797329, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009946922771632671, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009791849181056023, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009322176687419415, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00642356975004077, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.037622611969709396, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.037622611969709396, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.60.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11802186071872711, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10948383808135986, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10530480742454529, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09456649422645569, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05566531792283058, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05179368704557419, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06505714356899261, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.060001686215400696, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05665762349963188, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04929964244365692, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04659493267536163, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03311736136674881, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02876032330095768, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026771120727062225, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026293257251381874, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016596511006355286, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013826114125549793, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013610223308205605, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012527721002697945, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012220494449138641, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00876291748136282, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008638049475848675, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00813727080821991, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005664572585374117, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03311736136674881, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03311736136674881, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.60.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.28704833984375, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.26689815521240234, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2601926922798157, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.23395901918411255, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1360834538936615, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.12843932211399078, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1533782184123993, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.14118248224258423, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.13839301466941833, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.12058065831661224, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.11367099732160568, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0781722366809845, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06756872683763504, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.06521622836589813, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06465616077184677, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039009712636470795, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03324628248810768, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0330694206058979, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.030032113194465637, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.029663756489753723, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.020110851153731346, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019701074808835983, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.019241975620388985, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01230993214994669, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039009712636470795, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.039009712636470795, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.60.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.19646874070167542, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.16384157538414001, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14972926676273346, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1271185427904129, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08926163613796234, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07576076686382294, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11296119540929794, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10259371995925903, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09457577019929886, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07092221826314926, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06603467464447021, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05825813487172127, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05051838234066963, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.044615738093853, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04313825070858002, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029835451394319534, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.025739949196577072, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025274982675909996, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02261718362569809, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021734116598963737, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017512617632746696, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01962425373494625, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015834953635931015, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.016204746440052986, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.044615738093853, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.044615738093853, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.60.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16012896597385406, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.14884349703788757, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1452174335718155, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13086023926734924, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07636439055204391, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07219638675451279, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.0855456218123436, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07898730784654617, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07766357809305191, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06782225519418716, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06413467228412628, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04427846148610115, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03867581859230995, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.037491485476493835, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.037213169038295746, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.022322533652186394, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.020701056346297264, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.020624827593564987, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019135475158691406, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.01897118054330349, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012635257095098495, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014399691484868526, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012283433228731155, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.011611448600888252, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04427846148610115, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04427846148610115, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.60.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1794757843017578, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.16688406467437744, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1630699634552002, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1468961387872696, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08565982431173325, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.080990269780159, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09610879421234131, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08854936808347702, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08708484470844269, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07614339888095856, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0719904825091362, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0497095100581646, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04338367283344269, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04207517206668854, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04177199676632881, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02510767988860607, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023228853940963745, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023145996034145355, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021487636491656303, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021300040185451508, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01427700836211443, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016172317788004875, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013876864686608315, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.013033398427069187, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04338367283344269, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04338367283344269, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.60.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.23838472366333008, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2070900946855545, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1945173442363739, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17111483216285706, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.11022935807704926, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09755989909172058, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.13564370572566986, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.12224747240543365, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.11493707448244095, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09276704490184784, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08761861175298691, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06982477754354477, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05991191789507866, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.0544341579079628, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05306238308548927, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03603867068886757, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.03040344826877117, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.029929598793387413, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02725030854344368, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02639060840010643, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.0212808046489954, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.022013533860445023, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.019602404907345772, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.017535243183374405, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03603867068886757, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03603867068886757, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.61.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12636691331863403, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1172189712524414, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11286578327417374, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10125917196273804, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05948697030544281, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05538642778992653, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06978963315486908, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0639939159154892, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.060513317584991455, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05268605798482895, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04997701197862625, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03557461127638817, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030655406415462494, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028605574741959572, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028105422854423523, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017830006778240204, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014767820946872234, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014545298181474209, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013389536179602146, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013071560300886631, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009390980005264282, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009222186170518398, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008714685216546059, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006062811706215143, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03557461127638817, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03557461127638817, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.61.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10497461259365082, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09735573083162308, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09345956146717072, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08391207456588745, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04940728098154068, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04585224762558937, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.058362189680337906, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05354408174753189, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.050292693078517914, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0437520369887352, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.041475266218185425, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02972724661231041, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025684447959065437, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023778259754180908, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02330912835896015, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014923179522156715, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012326155789196491, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012113259173929691, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011184473522007465, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010890704579651356, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007909350097179413, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007795232813805342, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007292203605175018, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0051906113512814045, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0437520369887352, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0437520369887352, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.61.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.26696866750717163, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.24814850091934204, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.24153654277324677, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.21716105937957764, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.12663620710372925, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11934360861778259, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1447599232196808, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.13181950151920319, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.12878915667533875, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.11222828924655914, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10626388341188431, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.07395356893539429, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.06341561675071716, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.060929231345653534, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.06028832867741585, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03726784884929657, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.03131778538227081, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.031078778207302094, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02837199904024601, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02796054258942604, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01949658803641796, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01901577226817608, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.018511192873120308, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012375169433653355, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03726784884929657, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03726784884929657, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.61.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.17683713138103485, + "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": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13758213818073273, + "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": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12409134209156036, + "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": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10258252918720245, + "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": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07873118668794632, + "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": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06365850567817688, + "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": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10075479000806808, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.08811558783054352, + "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": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08456400781869888, + "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": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.056688640266656876, + "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": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05502108111977577, + "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": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05178561061620712, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04438893496990204, + "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": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.040290627628564835, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.039291463792324066, + "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": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.027374036610126495, + "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": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.024380557239055634, + "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": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02419065497815609, + "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": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02095865271985531, + "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": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.020390179008245468, + "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": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017047682777047157, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019338469952344894, + "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": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0159626305103302, + "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": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01683804579079151, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04438893496990204, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04438893496990204, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.61.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.14402827620506287, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1338133066892624, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.130474254488945, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.11754263192415237, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.06814476102590561, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.06434363126754761, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.07660897076129913, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07055262476205826, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.06931127607822418, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.060412779450416565, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.057064540684223175, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.039142560213804245, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.033866193145513535, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03276722878217697, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03250648081302643, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.019572075456380844, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.016990020871162415, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.016911746934056282, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.015433389693498611, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.0152671467512846, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01035616360604763, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01049682404845953, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.009988002479076385, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.007171108853071928, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.039142560213804245, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.039142560213804245, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.61.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.13902674615383148, + "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": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1291588842868805, + "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": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.12590764462947845, + "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": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.11335966736078262, + "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": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.06617257744073868, + "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": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.06249989941716194, + "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": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.07456442713737488, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.06855485588312149, + "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": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.06730829179286957, + "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": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.058704473078250885, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.055572424083948135, + "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": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03859229385852814, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0333503782749176, + "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": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.032280873507261276, + "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": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03203119710087776, + "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": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.019460812211036682, + "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": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.01749645173549652, + "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": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.017425958067178726, + "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": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.016087142750620842, + "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": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.01593630388379097, + "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": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01097217295318842, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011816460639238358, + "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": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.010641922242939472, + "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": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009201277047395706, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03859229385852814, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03859229385852814, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.61.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.12253644317388535, + "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": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.10733824968338013, + "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": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.09911785274744034, + "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": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.08696448802947998, + "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": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.057073742151260376, + "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": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.04976169392466545, + "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": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.07389768213033676, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.06558321416378021, + "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": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.058982111513614655, + "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": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.04807906225323677, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.045968856662511826, + "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": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.03808975592255592, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.032565776258707047, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.028860529884696007, + "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": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.027883924543857574, + "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": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.01972302421927452, + "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": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.017098333686590195, + "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": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.016729407012462616, + "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": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.015623463317751884, + "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": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.015100638382136822, + "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": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.011741380207240582, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.013410789892077446, + "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": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.010553393512964249, + "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": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011413374915719032, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.03808975592255592, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.03808975592255592, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + } + ], + "last_module_idx": 126, + "base_perplexity": 4.300232575958736 +} \ No newline at end of file