diff --git "a/measurement.json" "b/measurement.json" new file mode 100644--- /dev/null +++ "b/measurement.json" @@ -0,0 +1,149370 @@ +{ + "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": 0.022514499723911285, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.022311491891741753, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.01109473779797554, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.01104359794408083, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.01098217535763979, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.007633717264980078, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.03642715886235237, + "qparams": { + "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": 0.022093061357736588, + "qparams": { + "group_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": 0.010998030193150043, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.010957187041640282, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.011100989766418934, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.01363090518862009, + "qparams": { + "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": 0.010942257009446621, + "qparams": { + "group_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": 0.008279900066554546, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.007610967382788658, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.008421246893703938, + "qparams": { + "group_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": 0.007604658138006926, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.007387305144220591, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.007603795733302832, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.007386474404484034, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0076432665809988976, + "qparams": { + "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": 0.007603030651807785, + "qparams": { + "group_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": 0.0073928325437009335, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0073856171220541, + "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": 0.022514499723911285, + "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": 0.022514499723911285, + "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": 0.022553129121661186, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.022275609895586967, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.010420451872050762, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.010347791947424412, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.010274723172187805, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.005303594749420881, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.029961634427309036, + "qparams": { + "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": 0.022029053419828415, + "qparams": { + "group_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": 0.0103073101490736, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.010233061388134956, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.01036335714161396, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.011374511756002903, + "qparams": { + "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": 0.010211875662207603, + "qparams": { + "group_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": 0.006518459413200617, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.005261454731225967, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.006482572294771671, + "qparams": { + "group_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": 0.005246057640761137, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.00479462742805481, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.005243105813860893, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.004790725652128458, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.004984090104699135, + "qparams": { + "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": 0.005241864826530218, + "qparams": { + "group_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": 0.0044329240918159485, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0047884853556752205, + "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": 0.022553129121661186, + "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": 0.022553129121661186, + "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": 0.05967531353235245, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.05007101222872734, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.03956145420670509, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.03350541368126869, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.026725346222519875, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.01891622692346573, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05124795809388161, + "qparams": { + "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": 0.0385764054954052, + "qparams": { + "group_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": 0.02834981307387352, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.022216476500034332, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.02341546304523945, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.027479378506541252, + "qparams": { + "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": 0.019013872370123863, + "qparams": { + "group_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": 0.013233579695224762, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.01153398398309946, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013998888432979584, + "qparams": { + "group_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": 0.007495848461985588, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.006705666892230511, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.006705554202198982, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.005692281294614077, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007090963423252106, + "qparams": { + "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": 0.006186806131154299, + "qparams": { + "group_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": 0.003969585057348013, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004510048311203718, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.05007101222872734, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.05007101222872734, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 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": 0.0250783022493124, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.015700632706284523, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.009050632826983929, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.01029346976429224, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.009874910116195679, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.004316965118050575, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.016683734953403473, + "qparams": { + "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": 0.014849072322249413, + "qparams": { + "group_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": 0.011822397820651531, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.007578269578516483, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.008005894720554352, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.008452496491372585, + "qparams": { + "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": 0.0074170855805277824, + "qparams": { + "group_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": 0.005338963586837053, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.004737357143312693, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.004459663759917021, + "qparams": { + "group_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": 0.003646872704848647, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0034842698369175196, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.003361210459843278, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0030804648995399475, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.002778861206024885, + "qparams": { + "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": 0.003344156313687563, + "qparams": { + "group_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": 0.0022483416832983494, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.002953809918835759, + "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": 0.0250783022493124, + "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": 0.0250783022493124, + "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": 0.055448487401008606, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.05190567672252655, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.0507415272295475, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.04627082124352455, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.024908199906349182, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.02377404272556305, + "qparams": { + "group_size": { + "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": 0.02785174362361431, + "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": 0.025774342939257622, + "qparams": { + "group_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": 0.02525158040225506, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.02267349883913994, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.021718045696616173, + "qparams": { + "group_size": { + "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": 0.014174580574035645, + "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": 0.01245639193803072, + "qparams": { + "group_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": 0.012086856178939342, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.01199154183268547, + "qparams": { + "group_size": { + "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": 0.007166438270360231, + "qparams": { + "group_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": 0.006656275596469641, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.006623952649533749, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0062595633789896965, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.006205284968018532, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.00406090309843421, + "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": 0.004617923870682716, + "qparams": { + "group_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": 0.003933374769985676, + "qparams": { + "group_size": { + "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": 0.0037076075095683336, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.05190567672252655, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.05190567672252655, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "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": 0.08258835226297379, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.07744863629341125, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.07578907161951065, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.06911179423332214, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.037094324827194214, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03544862940907478, + "qparams": { + "group_size": { + "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": 0.04138623923063278, + "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": 0.03823968395590782, + "qparams": { + "group_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": 0.037598393857479095, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03375519812107086, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03230643644928932, + "qparams": { + "group_size": { + "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": 0.020909518003463745, + "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": 0.01818467676639557, + "qparams": { + "group_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": 0.017685802653431892, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.01756555214524269, + "qparams": { + "group_size": { + "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": 0.010454349219799042, + "qparams": { + "group_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": 0.009233977645635605, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009195729158818722, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.008592475205659866, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.00851826835423708, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.0055837482213974, + "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": 0.005793000105768442, + "qparams": { + "group_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": 0.00540353124961257, + "qparams": { + "group_size": { + "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": 0.004099564626812935, + "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": 0.04138623923063278, + "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": 0.04138623923063278, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.070232093334198, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.06133633852005005, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.03714856877923012, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.03431615233421326, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0294857956469059, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.017727429047226906, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.04408891871571541, + "qparams": { + "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": 0.055112797766923904, + "qparams": { + "group_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": 0.030995942652225494, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.026532471179962158, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.026602106168866158, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.019547002390027046, + "qparams": { + "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": 0.024729294702410698, + "qparams": { + "group_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": 0.01632016897201538, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.013201575726270676, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.011108847334980965, + "qparams": { + "group_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": 0.010929936543107033, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009361669421195984, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010548346675932407, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.008839443325996399, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0066325305961072445, + "qparams": { + "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": 0.010327868163585663, + "qparams": { + "group_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": 0.004896271042525768, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.008349204435944557, + "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": 0.03714856877923012, + "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": 0.03714856877923012, + "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": 0.013448807410895824, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.010726813226938248, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.008224677294492722, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.007226092740893364, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.005756696220487356, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0038488961290568113, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.010644244961440563, + "qparams": { + "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": 0.008388723246753216, + "qparams": { + "group_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": 0.006261657923460007, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0047687808983027935, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.00507514039054513, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0056372941471636295, + "qparams": { + "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": 0.004148303065448999, + "qparams": { + "group_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": 0.002879250096157193, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0024631561245769262, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.002900462131947279, + "qparams": { + "group_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": 0.0016322610899806023, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.001442278502508998, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.001466527464799583, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0012130021350458264, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0015451281797140837, + "qparams": { + "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": 0.0013689588522538543, + "qparams": { + "group_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": 0.00093931017909199, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0009793255012482405, + "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": 0.013448807410895824, + "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": 0.013448807410895824, + "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": 0.013078085146844387, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.01023164577782154, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.007879136130213737, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.006905152462422848, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0054551842622458935, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.003735624486580491, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.008988307788968086, + "qparams": { + "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": 0.007903055287897587, + "qparams": { + "group_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": 0.006006260402500629, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0044229100458323956, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.004529986530542374, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.004573552403599024, + "qparams": { + "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": 0.0037896970752626657, + "qparams": { + "group_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": 0.002705508377403021, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.00238304422236979, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.002311906311661005, + "qparams": { + "group_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": 0.0015392731875181198, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0014023127732798457, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0013642276171594858, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0011670945677906275, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0012592836283147335, + "qparams": { + "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": 0.0012657658662647009, + "qparams": { + "group_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": 0.0008786758407950401, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.00093374791322276, + "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": 0.013078085146844387, + "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": 0.013078085146844387, + "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": 0.07606016099452972, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.06434828042984009, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.059170957654714584, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.05096278339624405, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.03392864018678665, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.028964273631572723, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.042422857135534286, + "qparams": { + "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": 0.038796838372945786, + "qparams": { + "group_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": 0.03564542531967163, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.027543507516384125, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.025804299861192703, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.021671656519174576, + "qparams": { + "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": 0.018625382333993912, + "qparams": { + "group_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": 0.016400974243879318, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.015838241204619408, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.010865036398172379, + "qparams": { + "group_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": 0.008686397224664688, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.008475309237837791, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0074607860296964645, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.007088290527462959, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.005747911985963583, + "qparams": { + "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": 0.005890041124075651, + "qparams": { + "group_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": 0.004961119499057531, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004129583947360516, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05096278339624405, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05096278339624405, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "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": 0.1476062834262848, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.088862344622612, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.05484520271420479, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.06438905745744705, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06319908052682877, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.033117376267910004, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.09167767316102982, + "qparams": { + "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": 0.08186469972133636, + "qparams": { + "group_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": 0.0708274096250534, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04135609418153763, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.044189341366291046, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.04760013893246651, + "qparams": { + "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": 0.04054274410009384, + "qparams": { + "group_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": 0.03213987872004509, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.029834188520908356, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.024366013705730438, + "qparams": { + "group_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": 0.019260894507169724, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018550436943769455, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.015934718772768974, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014528617262840271, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.014095642603933811, + "qparams": { + "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": 0.015829868614673615, + "qparams": { + "group_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": 0.011527984403073788, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01309987809509039, + "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": 0.04135609418153763, + "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": 0.04135609418153763, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.11057674884796143, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.10531095415353775, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1037542074918747, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09547944366931915, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04908597469329834, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.047618329524993896, + "qparams": { + "group_size": { + "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": 0.0540364570915699, + "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": 0.049965981394052505, + "qparams": { + "group_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": 0.04950924217700958, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04564759507775307, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.044091831892728806, + "qparams": { + "group_size": { + "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": 0.027322163805365562, + "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": 0.02384348399937153, + "qparams": { + "group_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": 0.023477258160710335, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02338865026831627, + "qparams": { + "group_size": { + "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": 0.0136990612372756, + "qparams": { + "group_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": 0.012563196010887623, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012542112730443478, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01193744782358408, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011883488856256008, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.007562591694295406, + "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": 0.008249158971011639, + "qparams": { + "group_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": 0.007444051560014486, + "qparams": { + "group_size": { + "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": 0.006317139603197575, + "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": 0.0540364570915699, + "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": 0.0540364570915699, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "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": 0.1343027502298355, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.12816107273101807, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.12642446160316467, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1167796328663826, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05977672338485718, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.058056287467479706, + "qparams": { + "group_size": { + "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": 0.06567893922328949, + "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": 0.06074938178062439, + "qparams": { + "group_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": 0.06028980016708374, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.055700741708278656, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05376997962594032, + "qparams": { + "group_size": { + "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": 0.032941121608018875, + "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": 0.028658060356974602, + "qparams": { + "group_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": 0.028239475563168526, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.028136039152741432, + "qparams": { + "group_size": { + "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": 0.01639948971569538, + "qparams": { + "group_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": 0.014364774338901043, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014336987398564816, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013559088110923767, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013500386849045753, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.00850652065128088, + "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": 0.008407455869019032, + "qparams": { + "group_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": 0.008361181244254112, + "qparams": { + "group_size": { + "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": 0.0053485301323235035, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.05376997962594032, + "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": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.05376997962594032, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "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": 0.12605570256710052, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11310424655675888, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09906839579343796, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09035175293684006, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.060767099261283875, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.044814810156822205, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07764542102813721, + "qparams": { + "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": 0.07677826285362244, + "qparams": { + "group_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": 0.06259114295244217, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05542277917265892, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0546625591814518, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.043512534350156784, + "qparams": { + "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": 0.04408203437924385, + "qparams": { + "group_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": 0.02842303365468979, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.025597481057047844, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02033960074186325, + "qparams": { + "group_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": 0.01710892654955387, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015099398791790009, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.016239896416664124, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013997415080666542, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.011881383135914803, + "qparams": { + "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": 0.014436553232371807, + "qparams": { + "group_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": 0.009355098009109497, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010768724605441093, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.0546625591814518, + "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": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.0546625591814518, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "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": 0.029319044202566147, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.02693275175988674, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.025831587612628937, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.023115774616599083, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.013257232494652271, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.012265993282198906, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.016135776415467262, + "qparams": { + "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": 0.014371685683727264, + "qparams": { + "group_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": 0.013526289723813534, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.011749854311347008, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.01123678870499134, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.008262555114924908, + "qparams": { + "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": 0.006903032772243023, + "qparams": { + "group_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": 0.006373292300850153, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.00623778672888875, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.004179009702056646, + "qparams": { + "group_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": 0.003350410144776106, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0032880050130188465, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.003057661233469844, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0029724957421422005, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.002239903900772333, + "qparams": { + "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": 0.0021897596307098866, + "qparams": { + "group_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": 0.00198867404833436, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0015344080748036504, + "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": 0.029319044202566147, + "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": 0.029319044202566147, + "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": 0.024093227460980415, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.02205776609480381, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.021155282855033875, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.018890196457505226, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.010840571485459805, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.010032806545495987, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.012917932122945786, + "qparams": { + "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": 0.011750254780054092, + "qparams": { + "group_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": 0.011069677770137787, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.009576349519193172, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.009089337661862373, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.006528294645249844, + "qparams": { + "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": 0.005597924813628197, + "qparams": { + "group_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": 0.005197323393076658, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0050984714180231094, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.0032795891165733337, + "qparams": { + "group_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": 0.0027184016071259975, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.002676595700904727, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.002471152925863862, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.002409183420240879, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.001745866029523313, + "qparams": { + "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": 0.0017502757254987955, + "qparams": { + "group_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": 0.0015964302001520991, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0012126774527132511, + "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": 0.024093227460980415, + "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": 0.024093227460980415, + "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": 0.11295191943645477, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10527406632900238, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10273940861225128, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09302330762147903, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05180591344833374, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04912921413779259, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05814136937260628, + "qparams": { + "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": 0.05367100238800049, + "qparams": { + "group_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": 0.05265607684850693, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.046485014259815216, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04415559023618698, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.029469572007656097, + "qparams": { + "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": 0.025540582835674286, + "qparams": { + "group_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": 0.024699531495571136, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.024502698332071304, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014701592735946178, + "qparams": { + "group_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": 0.012591671198606491, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012524809688329697, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011501616798341274, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011371470056474209, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007584762293845415, + "qparams": { + "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": 0.00746116554364562, + "qparams": { + "group_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": 0.0072710346430540085, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0046890792436897755, + "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": 0.05367100238800049, + "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": 0.05367100238800049, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "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": 0.15896253287792206, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12408009171485901, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10998886078596115, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0899997428059578, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07104656100273132, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05697540566325188, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.09086371958255768, + "qparams": { + "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": 0.08199385553598404, + "qparams": { + "group_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": 0.07601398229598999, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05201096087694168, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04839916154742241, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.046934016048908234, + "qparams": { + "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": 0.040346160531044006, + "qparams": { + "group_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": 0.03542830049991608, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03418769687414169, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.024194758385419846, + "qparams": { + "group_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": 0.0204289760440588, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02007908746600151, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01728402078151703, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01651676930487156, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01433467771857977, + "qparams": { + "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": 0.015618539415299892, + "qparams": { + "group_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": 0.012916292995214462, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.012876119464635849, + "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": 0.05201096087694168, + "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": 0.05201096087694168, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.15512846410274506, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.14789031445980072, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.14580588042736053, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1346779465675354, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07067439705133438, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06848706305027008, + "qparams": { + "group_size": { + "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": 0.07783876359462738, + "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": 0.07186034321784973, + "qparams": { + "group_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": 0.07131492346525192, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06557288765907288, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06342270970344543, + "qparams": { + "group_size": { + "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": 0.03951170668005943, + "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": 0.03425467386841774, + "qparams": { + "group_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": 0.03373301774263382, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03360781818628311, + "qparams": { + "group_size": { + "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": 0.019776007160544395, + "qparams": { + "group_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": 0.017616821452975273, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017578519880771637, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.016629483550786972, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0165561456233263, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.010689999908208847, + "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": 0.010981889441609383, + "qparams": { + "group_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": 0.01052437536418438, + "qparams": { + "group_size": { + "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": 0.007835965603590012, + "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": 0.03951170668005943, + "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": 0.03951170668005943, + "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": 0.17961886525154114, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1713431179523468, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1689334511756897, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15618956089019775, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08196736872196198, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07946739345788956, + "qparams": { + "group_size": { + "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": 0.09023064374923706, + "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": 0.08337012678384781, + "qparams": { + "group_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": 0.08275071531534195, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07609876245260239, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07356023043394089, + "qparams": { + "group_size": { + "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": 0.045609209686517715, + "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": 0.039565153419971466, + "qparams": { + "group_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": 0.03895936533808708, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03882838040590286, + "qparams": { + "group_size": { + "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": 0.022753939032554626, + "qparams": { + "group_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": 0.019870491698384285, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019835136830806732, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018695397302508354, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018608050420880318, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01191517524421215, + "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": 0.011702891439199448, + "qparams": { + "group_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": 0.011721540242433548, + "qparams": { + "group_size": { + "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": 0.007522791624069214, + "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": 0.045609209686517715, + "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": 0.045609209686517715, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.1771124303340912, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15709520876407623, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.14881527423858643, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13356056809425354, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07999896258115768, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07207667082548141, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.09576117247343063, + "qparams": { + "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": 0.08789798617362976, + "qparams": { + "group_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": 0.08291608095169067, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06903580576181412, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06602060049772263, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.04878483712673187, + "qparams": { + "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": 0.04214785248041153, + "qparams": { + "group_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": 0.03858949616551399, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0377134345471859, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.024548716843128204, + "qparams": { + "group_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": 0.020468920469284058, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02015211619436741, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018367119133472443, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017804453149437904, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.013409920036792755, + "qparams": { + "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": 0.013637477532029152, + "qparams": { + "group_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": 0.012293758802115917, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.009818811900913715, + "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": 0.04878483712673187, + "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": 0.04878483712673187, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.044313590973615646, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.04107961803674698, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.0397198386490345, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.03545159846544266, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.020379649475216866, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.019138256087899208, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.024291878566145897, + "qparams": { + "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": 0.021640213206410408, + "qparams": { + "group_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": 0.020730547606945038, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.01824791356921196, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.01722792722284794, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.012497283518314362, + "qparams": { + "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": 0.010407751426100731, + "qparams": { + "group_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": 0.00985395722091198, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.00971850287169218, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.006327944807708263, + "qparams": { + "group_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": 0.005145931616425514, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.005093762651085854, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.00473064836114645, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.004648134112358093, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0033616451546549797, + "qparams": { + "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": 0.0032385208178311586, + "qparams": { + "group_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": 0.003188192145898938, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.002215418964624405, + "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": 0.044313590973615646, + "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": 0.044313590973615646, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.036141566932201385, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.03344086930155754, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.03215152397751808, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.028874188661575317, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.016490330919623375, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.015378538519144058, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.019505150616168976, + "qparams": { + "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": 0.0177733413875103, + "qparams": { + "group_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": 0.016788093373179436, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.01470236387103796, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.014034582301974297, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.009874868206679821, + "qparams": { + "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": 0.008484786376357079, + "qparams": { + "group_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": 0.007926063612103462, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0077895186841487885, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.004954606760293245, + "qparams": { + "group_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": 0.0041612256318330765, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.004101715050637722, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.003811549860984087, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0037277506198734045, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0026466844137758017, + "qparams": { + "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": 0.0026924919802695513, + "qparams": { + "group_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": 0.0024574531707912683, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0018970021046698093, + "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": 0.036141566932201385, + "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": 0.036141566932201385, + "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": 0.15588821470737457, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1456451565027237, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1423969864845276, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1292468011379242, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07200659811496735, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06845349073410034, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.08114556223154068, + "qparams": { + "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": 0.07435014843940735, + "qparams": { + "group_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": 0.07307739555835724, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06475873291492462, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06170668452978134, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.04111761227250099, + "qparams": { + "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": 0.03543238341808319, + "qparams": { + "group_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": 0.03435356542468071, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03409843519330025, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.020550519227981567, + "qparams": { + "group_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": 0.01750798709690571, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017425093799829483, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.016025763005018234, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015862345695495605, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.010656001046299934, + "qparams": { + "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": 0.01034651231020689, + "qparams": { + "group_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": 0.010245312936604023, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.006502197589725256, + "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": 0.04111761227250099, + "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": 0.04111761227250099, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.1499260514974594, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1236352026462555, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11187421530485153, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08871414512395859, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06774207204580307, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05648642033338547, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.08873043954372406, + "qparams": { + "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": 0.07819397747516632, + "qparams": { + "group_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": 0.07071222364902496, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05178605392575264, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04738593101501465, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.045040447264909744, + "qparams": { + "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": 0.039514824748039246, + "qparams": { + "group_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": 0.03496130183339119, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.033714260905981064, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02401203103363514, + "qparams": { + "group_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": 0.021552469581365585, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021121563389897346, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.019051704555749893, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018387984484434128, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.014988405629992485, + "qparams": { + "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": 0.017468392848968506, + "qparams": { + "group_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": 0.0135662155225873, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01521348487585783, + "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": 0.05178605392575264, + "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": 0.05178605392575264, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.16864444315433502, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15998978912830353, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1574074774980545, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14457039535045624, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07846435159444809, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07558790594339371, + "qparams": { + "group_size": { + "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": 0.08660958707332611, + "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": 0.08003672957420349, + "qparams": { + "group_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": 0.07930202037096024, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07201185077428818, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06923655420541763, + "qparams": { + "group_size": { + "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": 0.044374555349349976, + "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": 0.03861832991242409, + "qparams": { + "group_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": 0.03791201487183571, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03774070367217064, + "qparams": { + "group_size": { + "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": 0.022262994199991226, + "qparams": { + "group_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": 0.020296569913625717, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020244114100933075, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01908184587955475, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01897568255662918, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012301593087613583, + "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": 0.013338739052414894, + "qparams": { + "group_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": 0.012082716450095177, + "qparams": { + "group_size": { + "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": 0.010199475102126598, + "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": 0.044374555349349976, + "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": 0.044374555349349976, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.2046610414981842, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.19420303404331207, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.19111984968185425, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1755073070526123, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09495719522237778, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09152212738990784, + "qparams": { + "group_size": { + "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": 0.10491827130317688, + "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": 0.096870057284832, + "qparams": { + "group_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": 0.09599138051271439, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08711691200733185, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08373383432626724, + "qparams": { + "group_size": { + "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": 0.053353264927864075, + "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": 0.04625636339187622, + "qparams": { + "group_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": 0.04542196914553642, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04520401358604431, + "qparams": { + "group_size": { + "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": 0.02667040005326271, + "qparams": { + "group_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": 0.023453643545508385, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023406775668263435, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02189698815345764, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02177530713379383, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.014186525717377663, + "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": 0.014282493852078915, + "qparams": { + "group_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": 0.0139121999964118, + "qparams": { + "group_size": { + "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": 0.009747158735990524, + "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": 0.053353264927864075, + "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": 0.053353264927864075, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.0444239042699337, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.0403570756316185, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.0257367342710495, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.02253265492618084, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.014946918934583664, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.013262458145618439, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.019955355674028397, + "qparams": { + "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": 0.03214935213327408, + "qparams": { + "group_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": 0.015602370724081993, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.012375318445265293, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.01167723536491394, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.009596118703484535, + "qparams": { + "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": 0.008296999149024487, + "qparams": { + "group_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": 0.0074189030565321445, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.007659936789423227, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.004947032779455185, + "qparams": { + "group_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": 0.004915656987577677, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0040867384523153305, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.004526520613580942, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0035720905289053917, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.003731494303792715, + "qparams": { + "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": 0.003985753748565912, + "qparams": { + "group_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": 0.002575997728854418, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.002458923729136586, + "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": 0.0444239042699337, + "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": 0.0444239042699337, + "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": 0.057834822684526443, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.05315409600734711, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.051252979785203934, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.046017538756132126, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.02644466795027256, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.02463720738887787, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.030644329264760017, + "qparams": { + "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": 0.028204036876559258, + "qparams": { + "group_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": 0.026992835104465485, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.023400045931339264, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.022197041660547256, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.01552662719041109, + "qparams": { + "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": 0.013448524288833141, + "qparams": { + "group_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": 0.01266907062381506, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.012480413541197777, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.0077658588998019695, + "qparams": { + "group_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": 0.006562684662640095, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.006490026134997606, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.005956241395324469, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.005838317330926657, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.004070763476192951, + "qparams": { + "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": 0.0041124774143099785, + "qparams": { + "group_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": 0.0038077477365732193, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.002765743527561426, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.05315409600734711, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.05315409600734711, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "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": 0.05110296979546547, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.046936649829149246, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.04517499357461929, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0405578650534153, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.02333960123360157, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.021708883345127106, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.027305038645863533, + "qparams": { + "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": 0.025013884529471397, + "qparams": { + "group_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": 0.023829244077205658, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.02064157836139202, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.019659554585814476, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.01384656596928835, + "qparams": { + "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": 0.011915566399693489, + "qparams": { + "group_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": 0.011185015551745892, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.011003626510500908, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.006925015710294247, + "qparams": { + "group_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": 0.005783586297184229, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.00571112846955657, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.00524700153619051, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.005133653525263071, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0036279219202697277, + "qparams": { + "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": 0.003617748850956559, + "qparams": { + "group_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": 0.0033792448230087757, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.00241145770996809, + "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": 0.05110296979546547, + "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": 0.05110296979546547, + "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.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.16774114966392517, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15632133185863495, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15254279971122742, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13803088665008545, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07778766006231308, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07370185852050781, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.08742524683475494, + "qparams": { + "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": 0.08061164617538452, + "qparams": { + "group_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": 0.07907866686582565, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06962543725967407, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06605177372694016, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.04434945061802864, + "qparams": { + "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": 0.03841869533061981, + "qparams": { + "group_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": 0.037151798605918884, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03684304654598236, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.022127537056803703, + "qparams": { + "group_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": 0.018922816962003708, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01882634125649929, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01723979040980339, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017046507447957993, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.011410702019929886, + "qparams": { + "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": 0.011194526217877865, + "qparams": { + "group_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": 0.010941760614514351, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.00700716394931078, + "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": 0.04434945061802864, + "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": 0.04434945061802864, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.19009168446063995, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16363336145877838, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15240316092967987, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.12282238155603409, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08891483396291733, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07758046686649323, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.10813729465007782, + "qparams": { + "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": 0.09828811883926392, + "qparams": { + "group_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": 0.0919785276055336, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06951475143432617, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0613144189119339, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05631909891963005, + "qparams": { + "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": 0.04816693812608719, + "qparams": { + "group_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": 0.0437997542321682, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04273446649312973, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.028614399954676628, + "qparams": { + "group_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": 0.02419045940041542, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023784061893820763, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02065758779644966, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019982649013400078, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016278963536024094, + "qparams": { + "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": 0.017260676249861717, + "qparams": { + "group_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": 0.014952090568840504, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013557089492678642, + "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": 0.04816693812608719, + "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": 0.04816693812608719, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.1810837835073471, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17117954790592194, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.16819964349269867, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15383607149124146, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08432898670434952, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.080983005464077, + "qparams": { + "group_size": { + "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": 0.09310651570558548, + "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": 0.08618447184562683, + "qparams": { + "group_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": 0.08534429967403412, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07682483643293381, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07344578206539154, + "qparams": { + "group_size": { + "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": 0.047318439930677414, + "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": 0.04112955927848816, + "qparams": { + "group_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": 0.04030189290642738, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.040107354521751404, + "qparams": { + "group_size": { + "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": 0.023622939363121986, + "qparams": { + "group_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": 0.020737379789352417, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020688273012638092, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01922459900379181, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019101537764072418, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012402805499732494, + "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": 0.012494811788201332, + "qparams": { + "group_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": 0.012130658142268658, + "qparams": { + "group_size": { + "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": 0.008331364020705223, + "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": 0.047318439930677414, + "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": 0.047318439930677414, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.2178393453359604, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.20599699020385742, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.20246568322181702, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18529121577739716, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10140149295330048, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09739696979522705, + "qparams": { + "group_size": { + "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": 0.11188576370477676, + "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": 0.10356757044792175, + "qparams": { + "group_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": 0.10262735933065414, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.092430979013443, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08828332275152206, + "qparams": { + "group_size": { + "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": 0.05673804134130478, + "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": 0.049337249249219894, + "qparams": { + "group_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": 0.04834914207458496, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04811875522136688, + "qparams": { + "group_size": { + "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": 0.02828730084002018, + "qparams": { + "group_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": 0.024562466889619827, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02450275421142578, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02270968072116375, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022564051672816277, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.014603422954678535, + "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": 0.01430782675743103, + "qparams": { + "group_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": 0.014274868182837963, + "qparams": { + "group_size": { + "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": 0.008883656933903694, + "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": 0.049337249249219894, + "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": 0.049337249249219894, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.22665436565876007, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19949877262115479, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.18884077668190002, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1660502552986145, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10401583462953568, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09311623126268387, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12326543033123016, + "qparams": { + "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": 0.11333391070365906, + "qparams": { + "group_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": 0.1077374815940857, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08747507631778717, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08223949372768402, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06297136843204498, + "qparams": { + "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": 0.054400209337472916, + "qparams": { + "group_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": 0.05016561970114708, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.049135830253362656, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03165261819958687, + "qparams": { + "group_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": 0.026388315483927727, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026034489274024963, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023160208016633987, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022469623014330864, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017157554626464844, + "qparams": { + "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": 0.01723230443894863, + "qparams": { + "group_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": 0.015818100422620773, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.012116389349102974, + "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": 291983616.0, + "err": 0.054400209337472916, + "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": 291983616.0, + "err": 0.054400209337472916, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.06821820884943008, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.06268829107284546, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.06031455844640732, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.05408114939928055, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.031273771077394485, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.029065337032079697, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.0363335944712162, + "qparams": { + "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": 0.03352447599172592, + "qparams": { + "group_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": 0.03191728889942169, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.027605591341853142, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.026135582476854324, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.018422575667500496, + "qparams": { + "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": 0.016027089208364487, + "qparams": { + "group_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": 0.015017395839095116, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.014775160700082779, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.009222937747836113, + "qparams": { + "group_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": 0.00783463753759861, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.007736014202237129, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.007112570106983185, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.006958067417144775, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.004867785144597292, + "qparams": { + "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": 0.00499774981290102, + "qparams": { + "group_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": 0.004532236140221357, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.003448179457336664, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05408114939928055, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05408114939928055, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "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": 0.05910046398639679, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.054292645305395126, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.05207858607172966, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0467512346804142, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.02712211199104786, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.02514808066189289, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.03192075714468956, + "qparams": { + "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": 0.029262177646160126, + "qparams": { + "group_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": 0.027692295610904694, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.023958493024110794, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.02281012386083603, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.016193872317671776, + "qparams": { + "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": 0.013976438902318478, + "qparams": { + "group_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": 0.0130197424441576, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.012789253145456314, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.008109310641884804, + "qparams": { + "group_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": 0.006766809616237879, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.006667556241154671, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.006138371303677559, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0059921410866081715, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.004264643415808678, + "qparams": { + "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": 0.004291010554879904, + "qparams": { + "group_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": 0.0039447881281375885, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0029108631424605846, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.054292645305395126, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.054292645305395126, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "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": 0.18010368943214417, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16762836277484894, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16349130868911743, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14761441946029663, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0837029293179512, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07917653769254684, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.09429960697889328, + "qparams": { + "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": 0.08680997788906097, + "qparams": { + "group_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": 0.08507147431373596, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07468410581350327, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07072699815034866, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.047881025820970535, + "qparams": { + "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": 0.041412051767110825, + "qparams": { + "group_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": 0.039995599538087845, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03966221585869789, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.023891184478998184, + "qparams": { + "group_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": 0.020381616428494453, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02026895247399807, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018526170402765274, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018312983214855194, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.012332750484347343, + "qparams": { + "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": 0.012081146240234375, + "qparams": { + "group_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": 0.011801160871982574, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.007580784149467945, + "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": 0.047881025820970535, + "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": 0.047881025820970535, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.1919630765914917, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16497927904129028, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15505145490169525, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.12875203788280487, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08926425874233246, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07840702682733536, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.10633117705583572, + "qparams": { + "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": 0.09694711118936539, + "qparams": { + "group_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": 0.09268086403608322, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07134488970041275, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06356155127286911, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.055018190294504166, + "qparams": { + "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": 0.047372013330459595, + "qparams": { + "group_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": 0.04388325661420822, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04302992299199104, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02796856127679348, + "qparams": { + "group_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": 0.024219242855906487, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02395543083548546, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02096615917980671, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02042747475206852, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01586069166660309, + "qparams": { + "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": 0.017128344625234604, + "qparams": { + "group_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": 0.014795360155403614, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013538800179958344, + "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": 0.047372013330459595, + "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": 0.047372013330459595, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.1692008227109909, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15946514904499054, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1565389335155487, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14288334548473358, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07903306931257248, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07568634301424026, + "qparams": { + "group_size": { + "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": 0.08746787905693054, + "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": 0.08094150573015213, + "qparams": { + "group_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": 0.08004429191350937, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07169438153505325, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06841061264276505, + "qparams": { + "group_size": { + "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": 0.04454668611288071, + "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": 0.03874771296977997, + "qparams": { + "group_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": 0.03789553791284561, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03769511356949806, + "qparams": { + "group_size": { + "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": 0.022267112508416176, + "qparams": { + "group_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": 0.01969177834689617, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019635535776615143, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01823529228568077, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01810985989868641, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011814972385764122, + "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": 0.012164453975856304, + "qparams": { + "group_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": 0.011539937928318977, + "qparams": { + "group_size": { + "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": 0.008461147546768188, + "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": 0.04454668611288071, + "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": 0.04454668611288071, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.22425627708435059, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21158315241336823, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2077263593673706, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18958336114883423, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.1046978235244751, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10030564665794373, + "qparams": { + "group_size": { + "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": 0.11580497026443481, + "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": 0.10716500878334045, + "qparams": { + "group_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": 0.10604241490364075, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09500394761562347, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09057055413722992, + "qparams": { + "group_size": { + "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": 0.058763183653354645, + "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": 0.051091939210891724, + "qparams": { + "group_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": 0.04997529834508896, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04971905052661896, + "qparams": { + "group_size": { + "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": 0.029332304373383522, + "qparams": { + "group_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": 0.02539796009659767, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02532218210399151, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02338985539972782, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023222418501973152, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015167384408414364, + "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": 0.014818912371993065, + "qparams": { + "group_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": 0.014792269095778465, + "qparams": { + "group_size": { + "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": 0.00920794252306223, + "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": 0.051091939210891724, + "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": 0.051091939210891724, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2292865663766861, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20165012776851654, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19053904712200165, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1675051599740982, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10527749359607697, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09406206756830215, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12566517293453217, + "qparams": { + "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": 0.11543276160955429, + "qparams": { + "group_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": 0.1092197522521019, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08868379890918732, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08336998522281647, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06438066065311432, + "qparams": { + "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": 0.05551915988326073, + "qparams": { + "group_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": 0.05092666298151016, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04979286715388298, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03244651108980179, + "qparams": { + "group_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": 0.026985660195350647, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02658873423933983, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023765886202454567, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023020587861537933, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017823297530412674, + "qparams": { + "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": 0.017899487167596817, + "qparams": { + "group_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": 0.016391238197684288, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.012832042761147022, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05092666298151016, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05092666298151016, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.08492022752761841, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07888157665729523, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07647791504859924, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.06894369423389435, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.03934159502387047, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03697110339999199, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.04493075609207153, + "qparams": { + "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": 0.04150015860795975, + "qparams": { + "group_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": 0.040019210427999496, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.035066764801740646, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03326351195573807, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02283298596739769, + "qparams": { + "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": 0.019859900698065758, + "qparams": { + "group_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": 0.018894072622060776, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.01866309531033039, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.011431018821895123, + "qparams": { + "group_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": 0.00984464306384325, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.00975156482309103, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.009000899270176888, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.008854136802256107, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006030669901520014, + "qparams": { + "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": 0.0062074665911495686, + "qparams": { + "group_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": 0.005704603623598814, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0042927199974656105, + "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": 0.04493075609207153, + "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": 0.04493075609207153, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.06989777833223343, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.06495256721973419, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.06288769096136093, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.056759510189294815, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.032407235354185104, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.030427390709519386, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.037514571100473404, + "qparams": { + "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": 0.034381166100502014, + "qparams": { + "group_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": 0.03296039626002312, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.028928358107805252, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.02758108638226986, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.019056957215070724, + "qparams": { + "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": 0.016430921852588654, + "qparams": { + "group_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": 0.015550904907286167, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.015337446704506874, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.009536350145936012, + "qparams": { + "group_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": 0.008046752773225307, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.007960316725075245, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0073509495705366135, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.007219870109111071, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.004995821975171566, + "qparams": { + "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": 0.00501400837674737, + "qparams": { + "group_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": 0.004696203861385584, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0033660694025456905, + "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": 0.037514571100473404, + "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": 0.037514571100473404, + "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": 0.1986808329820633, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.18562480807304382, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.18138594925403595, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1642436534166336, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0926915630698204, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08798345178365707, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.10397140681743622, + "qparams": { + "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": 0.09580855071544647, + "qparams": { + "group_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": 0.09411590546369553, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08305253088474274, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0787464901804924, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05277695506811142, + "qparams": { + "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": 0.045697905123233795, + "qparams": { + "group_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": 0.04427739977836609, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04394190013408661, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.026331007480621338, + "qparams": { + "group_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": 0.022520827129483223, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022410960868000984, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.020532304421067238, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020317016169428825, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.013546100817620754, + "qparams": { + "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": 0.013254648074507713, + "qparams": { + "group_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": 0.013008447363972664, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.008222540840506554, + "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": 0.05277695506811142, + "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": 0.05277695506811142, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.19254463911056519, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16224004328250885, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1501728594303131, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.12681594491004944, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08927657455205917, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07682842761278152, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1092245802283287, + "qparams": { + "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": 0.09881865233182907, + "qparams": { + "group_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": 0.09253479540348053, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06894148886203766, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06439739465713501, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05635036900639534, + "qparams": { + "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": 0.048125918954610825, + "qparams": { + "group_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": 0.043804995715618134, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04275146871805191, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.028392357751727104, + "qparams": { + "group_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": 0.024071594700217247, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02368772029876709, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.020379070192575455, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019688675180077553, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0159356240183115, + "qparams": { + "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": 0.01705656386911869, + "qparams": { + "group_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": 0.014569910243153572, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013277203775942326, + "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": 0.048125918954610825, + "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": 0.048125918954610825, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.16606862843036652, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1562240719795227, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1532636433839798, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13964611291885376, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07773511111736298, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07431593537330627, + "qparams": { + "group_size": { + "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": 0.08612263947725296, + "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": 0.07974556088447571, + "qparams": { + "group_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": 0.07878856360912323, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0702865868806839, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06690634042024612, + "qparams": { + "group_size": { + "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": 0.043804965913295746, + "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": 0.03810979798436165, + "qparams": { + "group_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": 0.037215035408735275, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03700035437941551, + "qparams": { + "group_size": { + "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": 0.021880120038986206, + "qparams": { + "group_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": 0.0191448163241148, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019085634499788284, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01763170212507248, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017496267333626747, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011461247690021992, + "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": 0.011546711437404156, + "qparams": { + "group_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": 0.01116291992366314, + "qparams": { + "group_size": { + "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": 0.007675902917981148, + "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": 0.043804965913295746, + "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": 0.043804965913295746, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.22820231318473816, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2148902863264084, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2108226716518402, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19215282797813416, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10681253671646118, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10217104107141495, + "qparams": { + "group_size": { + "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": 0.11833072453737259, + "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": 0.1094735860824585, + "qparams": { + "group_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": 0.1082242876291275, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0966220423579216, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0920323058962822, + "qparams": { + "group_size": { + "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": 0.060163360089063644, + "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": 0.052229028195142746, + "qparams": { + "group_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": 0.05103418231010437, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05075106397271156, + "qparams": { + "group_size": { + "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": 0.03004005365073681, + "qparams": { + "group_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": 0.025973962619900703, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02589975669980049, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023872720077633858, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023690534755587578, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015605534426867962, + "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": 0.015233788639307022, + "qparams": { + "group_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": 0.015201560221612453, + "qparams": { + "group_size": { + "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": 0.00953776203095913, + "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": 0.052229028195142746, + "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": 0.052229028195142746, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.23484815657138824, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20700177550315857, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19589266180992126, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1725335419178009, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10799708217382431, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09671730548143387, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12865135073661804, + "qparams": { + "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": 0.11814934015274048, + "qparams": { + "group_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": 0.11190275847911835, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09115687012672424, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08573085814714432, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06575088948011398, + "qparams": { + "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": 0.05682947114109993, + "qparams": { + "group_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": 0.05223561078310013, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.051110997796058655, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03307738155126572, + "qparams": { + "group_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": 0.02769322507083416, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02729634940624237, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024440471082925797, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02369190752506256, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.018059678375720978, + "qparams": { + "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": 0.018375098705291748, + "qparams": { + "group_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": 0.016617782413959503, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013202234171330929, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05223561078310013, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05223561078310013, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.0834432914853096, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07756802439689636, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07521356642246246, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0677415132522583, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.03867416828870773, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03636825084686279, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.044287584722042084, + "qparams": { + "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": 0.04083087295293808, + "qparams": { + "group_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": 0.039324648678302765, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.034486494958400726, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03268419951200485, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.022483963519334793, + "qparams": { + "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": 0.01951664127409458, + "qparams": { + "group_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": 0.018549129366874695, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.018313972279429436, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01125030592083931, + "qparams": { + "group_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": 0.009585707448422909, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009494100697338581, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.008741335943341255, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.008597332052886486, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.005896029062569141, + "qparams": { + "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": 0.0059389714151620865, + "qparams": { + "group_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": 0.0055675688199698925, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.003974298946559429, + "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": 0.044287584722042084, + "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": 0.044287584722042084, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.06997476518154144, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.06513013690710068, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.06296822428703308, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.05680537596344948, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.032483894377946854, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03046352230012417, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.03753429651260376, + "qparams": { + "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": 0.03457041457295418, + "qparams": { + "group_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": 0.03302118182182312, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.028997622430324554, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.027561955153942108, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.01905485801398754, + "qparams": { + "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": 0.016514843329787254, + "qparams": { + "group_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": 0.01557410042732954, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.015348680317401886, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.009530138224363327, + "qparams": { + "group_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": 0.008047121576964855, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.007949338294565678, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.007339806761592627, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.007199466228485107, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.004979794844985008, + "qparams": { + "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": 0.0050001791678369045, + "qparams": { + "group_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": 0.004665555898100138, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.003319330746307969, + "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": 0.03753429651260376, + "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": 0.03753429651260376, + "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": 0.1965702772140503, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.18364502489566803, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.17941996455192566, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16222290694713593, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09180339425802231, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08710545301437378, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.10333386063575745, + "qparams": { + "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": 0.0949421152472496, + "qparams": { + "group_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": 0.0932132750749588, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08219514042139053, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07789748162031174, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05253031477332115, + "qparams": { + "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": 0.04530717805027962, + "qparams": { + "group_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": 0.04387381672859192, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04353049024939537, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02621757611632347, + "qparams": { + "group_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": 0.02231372892856598, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022199595347046852, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.020316632464528084, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020099230110645294, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.013491383753716946, + "qparams": { + "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": 0.013136664405465126, + "qparams": { + "group_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": 0.012909826822578907, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.008148272521793842, + "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": 0.05253031477332115, + "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": 0.05253031477332115, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.21777154505252838, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.18137222528457642, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16851180791854858, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13788047432899475, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09957730770111084, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0855240598320961, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12018701434135437, + "qparams": { + "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": 0.1095934808254242, + "qparams": { + "group_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": 0.10456483066082001, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07628235965967178, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07037530839443207, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06208959221839905, + "qparams": { + "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": 0.05370563268661499, + "qparams": { + "group_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": 0.04915481433272362, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04802807420492172, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.031611233949661255, + "qparams": { + "group_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": 0.027488738298416138, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02719075232744217, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023385290056467056, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02267533540725708, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01819378137588501, + "qparams": { + "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": 0.01992216147482395, + "qparams": { + "group_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": 0.016859401017427444, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01600697450339794, + "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": 0.05370563268661499, + "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": 0.05370563268661499, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.15987488627433777, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1502363085746765, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.14730317890644073, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1340893805027008, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07499482482671738, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07161716371774673, + "qparams": { + "group_size": { + "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": 0.0832405835390091, + "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": 0.07701036334037781, + "qparams": { + "group_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": 0.07602323591709137, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06772307306528091, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06434093415737152, + "qparams": { + "group_size": { + "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": 0.04235667362809181, + "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": 0.036873046308755875, + "qparams": { + "group_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": 0.03595816344022751, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03574350103735924, + "qparams": { + "group_size": { + "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": 0.021188776940107346, + "qparams": { + "group_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": 0.018595105037093163, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01853097975254059, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.017133165150880814, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0169956237077713, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011174414306879044, + "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": 0.011370646767318249, + "qparams": { + "group_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": 0.010869340971112251, + "qparams": { + "group_size": { + "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": 0.007734110113233328, + "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": 0.04235667362809181, + "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": 0.04235667362809181, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.22820550203323364, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21471308171749115, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21057502925395966, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19176529347896576, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10719648003578186, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10246621817350388, + "qparams": { + "group_size": { + "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": 0.11891195178031921, + "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": 0.10998591780662537, + "qparams": { + "group_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": 0.10863761603832245, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09685520082712173, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09211622923612595, + "qparams": { + "group_size": { + "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": 0.06058633700013161, + "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": 0.05262058973312378, + "qparams": { + "group_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": 0.05137455835938454, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05109056085348129, + "qparams": { + "group_size": { + "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": 0.030270220711827278, + "qparams": { + "group_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": 0.02653442695736885, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026450786739587784, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024444127455353737, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02426028996706009, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01594315469264984, + "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": 0.016146674752235413, + "qparams": { + "group_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": 0.015526976436376572, + "qparams": { + "group_size": { + "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": 0.010900040157139301, + "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": 0.05262058973312378, + "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": 0.05262058973312378, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.22904238104820251, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20121194422245026, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19017294049263, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16722993552684784, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10519776493310928, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09396785497665405, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12681064009666443, + "qparams": { + "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": 0.11531597375869751, + "qparams": { + "group_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": 0.1091565266251564, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08867595344781876, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08347082138061523, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06462428718805313, + "qparams": { + "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": 0.05580499395728111, + "qparams": { + "group_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": 0.051201753318309784, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05008789524435997, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03307688981294632, + "qparams": { + "group_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": 0.027706801891326904, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027309684082865715, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02460847795009613, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023865144699811935, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.018849628046154976, + "qparams": { + "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": 0.01906225085258484, + "qparams": { + "group_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": 0.01737385056912899, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014391045086085796, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051201753318309784, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051201753318309784, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.0963093563914299, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08964257687330246, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08695869147777557, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.07839340716600418, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0446842685341835, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04207388684153557, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05109484866261482, + "qparams": { + "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": 0.047066595405340195, + "qparams": { + "group_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": 0.04541873559355736, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03987335041165352, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.037836238741874695, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.025920920073986053, + "qparams": { + "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": 0.022492749616503716, + "qparams": { + "group_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": 0.021424932405352592, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.021163344383239746, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.012969596311450005, + "qparams": { + "group_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": 0.011074919253587723, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0109727643430233, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01011115312576294, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009947462938725948, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006799998693168163, + "qparams": { + "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": 0.006858211476355791, + "qparams": { + "group_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": 0.006434875074774027, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004604839254170656, + "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": 0.05109484866261482, + "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": 0.05109484866261482, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.07876427471637726, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07329236716032028, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07094539701938629, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.06398996710777283, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.03656802698969841, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03434095159173012, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.042134832590818405, + "qparams": { + "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": 0.03881317004561424, + "qparams": { + "group_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": 0.037173643708229065, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03266793489456177, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.031050989404320717, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.021397335454821587, + "qparams": { + "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": 0.01855807937681675, + "qparams": { + "group_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": 0.017538148909807205, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.017294203862547874, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.010703862644731998, + "qparams": { + "group_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": 0.00906376726925373, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.008960215374827385, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.008279175497591496, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.008124122396111488, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.005605640355497599, + "qparams": { + "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": 0.005631471052765846, + "qparams": { + "group_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": 0.005275517702102661, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.00374901806935668, + "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": 0.042134832590818405, + "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": 0.042134832590818405, + "qparams": { + "group_size": { + "3": 128 + }, + "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": 0.21111765503883362, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19731980562210083, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19281016290187836, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17436626553535461, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09860936552286148, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09359389543533325, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11121270805597305, + "qparams": { + "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": 0.10200387239456177, + "qparams": { + "group_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": 0.10011444240808487, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08832657337188721, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08382081240415573, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05652749910950661, + "qparams": { + "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": 0.04867741838097572, + "qparams": { + "group_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": 0.047125913202762604, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.046751994639635086, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.028228607028722763, + "qparams": { + "group_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": 0.02397179789841175, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023852555081248283, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021843817085027695, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021603938192129135, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.014549379236996174, + "qparams": { + "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": 0.014109605923295021, + "qparams": { + "group_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": 0.013911250978708267, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.008755192160606384, + "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": 0.04867741838097572, + "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": 0.04867741838097572, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.20430415868759155, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.17811617255210876, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16907159984111786, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14512360095977783, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09528185427188873, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08517888933420181, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11062924563884735, + "qparams": { + "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": 0.10159627348184586, + "qparams": { + "group_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": 0.09826862812042236, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07792507112026215, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07061505317687988, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05732924863696098, + "qparams": { + "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": 0.04958989471197128, + "qparams": { + "group_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": 0.046768687665462494, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04608479142189026, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02902543731033802, + "qparams": { + "group_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": 0.025636959820985794, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025448143482208252, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.022435056045651436, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022008016705513, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016461214050650597, + "qparams": { + "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": 0.01783587969839573, + "qparams": { + "group_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": 0.015623608604073524, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014062856324017048, + "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": 0.04958989471197128, + "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": 0.04958989471197128, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.16519325971603394, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15527305006980896, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15217463672161102, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13841666281223297, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07774422317743301, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07420440763235092, + "qparams": { + "group_size": { + "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": 0.08642483502626419, + "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": 0.07988035678863525, + "qparams": { + "group_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": 0.07879422605037689, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07015787065029144, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06671343743801117, + "qparams": { + "group_size": { + "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": 0.044135045260190964, + "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": 0.03835385665297508, + "qparams": { + "group_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": 0.03737955167889595, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03714924305677414, + "qparams": { + "group_size": { + "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": 0.022088998928666115, + "qparams": { + "group_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": 0.019490860402584076, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01942134089767933, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.017985811457037926, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017841123044490814, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011780287139117718, + "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": 0.012158309109508991, + "qparams": { + "group_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": 0.011460219509899616, + "qparams": { + "group_size": { + "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": 0.008548134006559849, + "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": 0.044135045260190964, + "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": 0.044135045260190964, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.23049457371234894, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21675005555152893, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21252669394016266, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19341886043548584, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10816116631031036, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10329120606184006, + "qparams": { + "group_size": { + "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": 0.12021661549806595, + "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": 0.11102529615163803, + "qparams": { + "group_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": 0.10962865501642227, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09763144701719284, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09281133860349655, + "qparams": { + "group_size": { + "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": 0.06111808121204376, + "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": 0.05303634703159332, + "qparams": { + "group_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": 0.0517427884042263, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05143368989229202, + "qparams": { + "group_size": { + "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": 0.030555542558431625, + "qparams": { + "group_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": 0.02639027312397957, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026303565129637718, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024214109405875206, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024020833894610405, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015938740223646164, + "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": 0.015573423355817795, + "qparams": { + "group_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": 0.015496416948735714, + "qparams": { + "group_size": { + "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": 0.009865347295999527, + "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": 0.05303634703159332, + "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": 0.05303634703159332, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.23682761192321777, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20813557505607605, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19655132293701172, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1730465590953827, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10915850847959518, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09729139506816864, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1311756819486618, + "qparams": { + "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": 0.11975889652967453, + "qparams": { + "group_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": 0.11317124217748642, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0918019637465477, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08647242933511734, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06700948625802994, + "qparams": { + "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": 0.05790836364030838, + "qparams": { + "group_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": 0.05308404937386513, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.051889922469854355, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03385896608233452, + "qparams": { + "group_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": 0.028567444533109665, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028144769370555878, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025279542431235313, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024509821087121964, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01878461055457592, + "qparams": { + "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": 0.01949349045753479, + "qparams": { + "group_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": 0.01726367324590683, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014564304612576962, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05308404937386513, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05308404937386513, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.10296766459941864, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09586217999458313, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09300252050161362, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08393023908138275, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04787357524037361, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0450921393930912, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05470837280154228, + "qparams": { + "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": 0.05044836550951004, + "qparams": { + "group_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": 0.04867032915353775, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04278293251991272, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04059445858001709, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.027861129492521286, + "qparams": { + "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": 0.02422931231558323, + "qparams": { + "group_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": 0.02308080531656742, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.022806620225310326, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013977442868053913, + "qparams": { + "group_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": 0.01214736606925726, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012034653685986996, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011144322343170643, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010974155738949776, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007457961328327656, + "qparams": { + "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": 0.007806214038282633, + "qparams": { + "group_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": 0.0070732333697378635, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005596235394477844, + "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": 0.05470837280154228, + "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": 0.05470837280154228, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.08779054880142212, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08179745078086853, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07927406579256058, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0716157779097557, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04088371619582176, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03843807429075241, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.046993959695100784, + "qparams": { + "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": 0.043270356953144073, + "qparams": { + "group_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": 0.04154245927929878, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.036547377705574036, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.034780390560626984, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.023879937827587128, + "qparams": { + "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": 0.020697638392448425, + "qparams": { + "group_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": 0.019618647173047066, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.01935536041855812, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.011951178312301636, + "qparams": { + "group_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": 0.010151083581149578, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010042215697467327, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.009280634112656116, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.00911582913249731, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006272947881370783, + "qparams": { + "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": 0.006312097422778606, + "qparams": { + "group_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": 0.0059170955792069435, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004232075531035662, + "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": 0.046993959695100784, + "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": 0.046993959695100784, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.21007075905799866, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19646525382995605, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19194667041301727, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1735534965991974, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09823327511548996, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09321583807468414, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11084293574094772, + "qparams": { + "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": 0.10167647153139114, + "qparams": { + "group_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": 0.09971502423286438, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08801442384719849, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08350663632154465, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05635913833975792, + "qparams": { + "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": 0.04854094609618187, + "qparams": { + "group_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": 0.04696948826313019, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04659224674105644, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.028133854269981384, + "qparams": { + "group_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": 0.023943154141306877, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02381826750934124, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021827565506100655, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0215925220400095, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.014544092118740082, + "qparams": { + "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": 0.014175252057611942, + "qparams": { + "group_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": 0.013915140181779861, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.008901631459593773, + "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": 0.04854094609618187, + "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": 0.04854094609618187, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.21395230293273926, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.18234556913375854, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16984865069389343, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13784882426261902, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09847508370876312, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08582881838083267, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12119082361459732, + "qparams": { + "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": 0.10986663401126862, + "qparams": { + "group_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": 0.10253249853849411, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07784691452980042, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06954774260520935, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06267646700143814, + "qparams": { + "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": 0.0539720356464386, + "qparams": { + "group_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": 0.04887113720178604, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04760519042611122, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.031942687928676605, + "qparams": { + "group_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": 0.027595870196819305, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027162866666913033, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024006925523281097, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023231862112879753, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.018545327708125114, + "qparams": { + "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": 0.02035299688577652, + "qparams": { + "group_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": 0.01699390634894371, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.016526082530617714, + "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": 0.0539720356464386, + "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": 0.0539720356464386, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.16668078303337097, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15670843422412872, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15356232225894928, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13969533145427704, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07844991236925125, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07486648112535477, + "qparams": { + "group_size": { + "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": 0.08714473992586136, + "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": 0.08063355833292007, + "qparams": { + "group_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": 0.07951056957244873, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07077556103467941, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0672944039106369, + "qparams": { + "group_size": { + "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": 0.04449525102972984, + "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": 0.038692668080329895, + "qparams": { + "group_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": 0.03768942877650261, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03745810315012932, + "qparams": { + "group_size": { + "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": 0.02225475199520588, + "qparams": { + "group_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": 0.019632473587989807, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01956257037818432, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01810591295361519, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017961662262678146, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011827737092971802, + "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": 0.012215600349009037, + "qparams": { + "group_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": 0.011497057043015957, + "qparams": { + "group_size": { + "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": 0.008546330034732819, + "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": 0.04449525102972984, + "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": 0.04449525102972984, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.23120971024036407, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2174905389547348, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21320629119873047, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19402886927127838, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10857181996107101, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1036863774061203, + "qparams": { + "group_size": { + "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": 0.1206742599606514, + "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": 0.11146187782287598, + "qparams": { + "group_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": 0.11002703756093979, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0979928970336914, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09315969794988632, + "qparams": { + "group_size": { + "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": 0.06135580688714981, + "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": 0.05324774235486984, + "qparams": { + "group_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": 0.05193129926919937, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.051627885550260544, + "qparams": { + "group_size": { + "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": 0.030646448954939842, + "qparams": { + "group_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": 0.026486830785870552, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026398083195090294, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02430049143731594, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02410212717950344, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015957847237586975, + "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": 0.01561729796230793, + "qparams": { + "group_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": 0.015509642660617828, + "qparams": { + "group_size": { + "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": 0.00987457949668169, + "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": 0.05324774235486984, + "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": 0.05324774235486984, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2437131106853485, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21514734625816345, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20357313752174377, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1791887879371643, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11270838230848312, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10089629143476486, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13441802561283112, + "qparams": { + "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": 0.12325332313776016, + "qparams": { + "group_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": 0.11676362156867981, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09505954384803772, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08938305824995041, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06908033043146133, + "qparams": { + "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": 0.05972785875201225, + "qparams": { + "group_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": 0.05493598431348801, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.053776662796735764, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.034966498613357544, + "qparams": { + "group_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": 0.0298128891736269, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029400795698165894, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026504937559366226, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025743145495653152, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019614070653915405, + "qparams": { + "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": 0.020578626543283463, + "qparams": { + "group_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": 0.018121927976608276, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015667347237467766, + "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": 0.053776662796735764, + "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": 0.053776662796735764, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.10846634954214096, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10103587806224823, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09810873121023178, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08844274282455444, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05040323734283447, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04750584065914154, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.057409968227148056, + "qparams": { + "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": 0.053002651780843735, + "qparams": { + "group_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": 0.05122901871800423, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04502131789922714, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04264850914478302, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02919538877904415, + "qparams": { + "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": 0.025368470698595047, + "qparams": { + "group_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": 0.024214021861553192, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.023931842297315598, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01461342815309763, + "qparams": { + "group_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": 0.012588917277753353, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012481854297220707, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011516053229570389, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011343113146722317, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007702920585870743, + "qparams": { + "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": 0.00788659043610096, + "qparams": { + "group_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": 0.00731688505038619, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005421914625912905, + "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": 0.053002651780843735, + "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": 0.053002651780843735, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "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": 0.08818583935499191, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08212624490261078, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07952532917261124, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.07174882292747498, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0409831777215004, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.038523510098457336, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.04712769389152527, + "qparams": { + "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": 0.0434737391769886, + "qparams": { + "group_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": 0.041644953191280365, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03663262352347374, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.034791842103004456, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.023904401808977127, + "qparams": { + "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": 0.020773716270923615, + "qparams": { + "group_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": 0.019666962325572968, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.019395824521780014, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.011964196339249611, + "qparams": { + "group_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": 0.01017466839402914, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010059267282485962, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.009298844262957573, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009127981960773468, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006279551424086094, + "qparams": { + "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": 0.006329125724732876, + "qparams": { + "group_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": 0.005919856950640678, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0042371382005512714, + "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": 0.04712769389152527, + "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": 0.04712769389152527, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2109672576189041, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19708824157714844, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19222797453403473, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1737474948167801, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09855414927005768, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09338624775409698, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1112426146864891, + "qparams": { + "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": 0.10243508219718933, + "qparams": { + "group_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": 0.10005634278059006, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08827199786901474, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08352755755186081, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05648016184568405, + "qparams": { + "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": 0.048916224390268326, + "qparams": { + "group_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": 0.047143127769231796, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04671128839254379, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02820458635687828, + "qparams": { + "group_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": 0.02407289296388626, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023921340703964233, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02194441482424736, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02167867124080658, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.014593166299164295, + "qparams": { + "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": 0.014348956756293774, + "qparams": { + "group_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": 0.013949102722108364, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.009086857549846172, + "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": 0.048916224390268326, + "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": 0.048916224390268326, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.22814339399337769, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2058849334716797, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.195932075381279, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16801683604717255, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10753247886896133, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09803571552038193, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12762516736984253, + "qparams": { + "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": 0.11687332391738892, + "qparams": { + "group_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": 0.1099323183298111, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09073030948638916, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08253564685583115, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06593406200408936, + "qparams": { + "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": 0.05707388371229172, + "qparams": { + "group_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": 0.05278848111629486, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05174286663532257, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03342317417263985, + "qparams": { + "group_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": 0.029086505994200706, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028671786189079285, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026024017482995987, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025393471121788025, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.018898067995905876, + "qparams": { + "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": 0.0205441452562809, + "qparams": { + "group_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": 0.01761350780725479, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01620393805205822, + "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": 0.05278848111629486, + "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": 0.05278848111629486, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.16464568674564362, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1545810103416443, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15139667689800262, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.137520432472229, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07745862752199173, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07384350150823593, + "qparams": { + "group_size": { + "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": 0.0862089991569519, + "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": 0.07970709353685379, + "qparams": { + "group_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": 0.07853715866804123, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06977230310440063, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06626185029745102, + "qparams": { + "group_size": { + "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": 0.04404656961560249, + "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": 0.038319192826747894, + "qparams": { + "group_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": 0.03729980066418648, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.037063054740428925, + "qparams": { + "group_size": { + "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": 0.02205391228199005, + "qparams": { + "group_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": 0.019549086689949036, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019479285925626755, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018026383593678474, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01788119226694107, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011800698935985565, + "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": 0.012331515550613403, + "qparams": { + "group_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": 0.011474702507257462, + "qparams": { + "group_size": { + "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": 0.008819054812192917, + "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": 0.04404656961560249, + "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": 0.04404656961560249, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.23008133471012115, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21609003841876984, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2117738425731659, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1923685520887375, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10794153809547424, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10296355932950974, + "qparams": { + "group_size": { + "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": 0.11999491602182388, + "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": 0.1108890175819397, + "qparams": { + "group_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": 0.10942371189594269, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09721139818429947, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0922975018620491, + "qparams": { + "group_size": { + "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": 0.06094968691468239, + "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": 0.052950914949178696, + "qparams": { + "group_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": 0.05162106081843376, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05130316689610481, + "qparams": { + "group_size": { + "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": 0.030452115461230278, + "qparams": { + "group_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": 0.02630426175892353, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026212558150291443, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024083811789751053, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023882541805505753, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01580744795501232, + "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": 0.015483931638300419, + "qparams": { + "group_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": 0.015351944603025913, + "qparams": { + "group_size": { + "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": 0.009740624576807022, + "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": 0.052950914949178696, + "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": 0.052950914949178696, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.25075989961624146, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22217363119125366, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2111022025346756, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1858300268650055, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11603955924510956, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10452666133642197, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13668401539325714, + "qparams": { + "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": 0.12580229341983795, + "qparams": { + "group_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": 0.11994213610887527, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09808328747749329, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09196900576353073, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07016317546367645, + "qparams": { + "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": 0.06077998876571655, + "qparams": { + "group_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": 0.056314438581466675, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05524394288659096, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03547796979546547, + "qparams": { + "group_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": 0.03015740029513836, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02977759763598442, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02673555165529251, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02601546235382557, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019743233919143677, + "qparams": { + "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": 0.02029629424214363, + "qparams": { + "group_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": 0.018410777673125267, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015042390674352646, + "qparams": { + "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": 0.03547796979546547, + "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": 0.03547796979546547, + "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": 0.10835405439138412, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10088972002267838, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09795866906642914, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08826480060815811, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.050408706068992615, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04749706760048866, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.057477112859487534, + "qparams": { + "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": 0.0530475378036499, + "qparams": { + "group_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": 0.05122130364179611, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04499569162726402, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04260735586285591, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02919469028711319, + "qparams": { + "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": 0.025376709178090096, + "qparams": { + "group_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": 0.024178002029657364, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02389855869114399, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014614409767091274, + "qparams": { + "group_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": 0.012504291720688343, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01239111926406622, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011413958854973316, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011233644559979439, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0076629770919680595, + "qparams": { + "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": 0.00774361751973629, + "qparams": { + "group_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": 0.007257077377289534, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0051937890239059925, + "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": 0.0530475378036499, + "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": 0.0530475378036499, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "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": 0.08967132866382599, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08358044177293777, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08095157146453857, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.07301785051822662, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04174153506755829, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.039225976914167404, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.047920964658260345, + "qparams": { + "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": 0.04428906738758087, + "qparams": { + "group_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": 0.04240090772509575, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0372987799346447, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.035381585359573364, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02433166839182377, + "qparams": { + "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": 0.021166188642382622, + "qparams": { + "group_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": 0.020018162205815315, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.019744448363780975, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.012171530164778233, + "qparams": { + "group_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": 0.01032037753611803, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010198798961937428, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.009421071037650108, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009245073422789574, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006366219837218523, + "qparams": { + "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": 0.0063745067454874516, + "qparams": { + "group_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": 0.005995661020278931, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004196366295218468, + "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": 0.047920964658260345, + "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": 0.047920964658260345, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2124849557876587, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19840551912784576, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19337983429431915, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1745954155921936, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09926963597536087, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09395039081573486, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11223609000444412, + "qparams": { + "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": 0.10328277945518494, + "qparams": { + "group_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": 0.10080229490995407, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08874183148145676, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08395414054393768, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05697621405124664, + "qparams": { + "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": 0.0493009015917778, + "qparams": { + "group_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": 0.0475020632147789, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.047074854373931885, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.028469916433095932, + "qparams": { + "group_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": 0.02426827885210514, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024107545614242554, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02209487557411194, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021814711391925812, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.014727383852005005, + "qparams": { + "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": 0.014487517066299915, + "qparams": { + "group_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": 0.014058459550142288, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.009193096309900284, + "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": 0.0493009015917778, + "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": 0.0493009015917778, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.22464118897914886, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20012512803077698, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1891089826822281, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16098599135875702, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10455101728439331, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09448233991861343, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1270054578781128, + "qparams": { + "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": 0.11600039899349213, + "qparams": { + "group_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": 0.10865373909473419, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08798598498106003, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08094445616006851, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06547016650438309, + "qparams": { + "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": 0.05654952675104141, + "qparams": { + "group_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": 0.05125394091010094, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04995392635464668, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.032957110553979874, + "qparams": { + "group_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": 0.028086811304092407, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027627522125840187, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025036433711647987, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024224843829870224, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01829634979367256, + "qparams": { + "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": 0.019856812432408333, + "qparams": { + "group_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": 0.016678309068083763, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015369387343525887, + "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": 0.05125394091010094, + "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": 0.05125394091010094, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.1658404916524887, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1556185483932495, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1523059606552124, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13823431730270386, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07808014750480652, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07437513023614883, + "qparams": { + "group_size": { + "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": 0.08700675517320633, + "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": 0.08046244084835052, + "qparams": { + "group_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": 0.0791732668876648, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07024296373128891, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06675736606121063, + "qparams": { + "group_size": { + "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": 0.04448961466550827, + "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": 0.038703735917806625, + "qparams": { + "group_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": 0.037632256746292114, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03738158941268921, + "qparams": { + "group_size": { + "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": 0.02228773757815361, + "qparams": { + "group_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": 0.019764157012104988, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019685791805386543, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01822824217379093, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018070310354232788, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011956540867686272, + "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": 0.012540686875581741, + "qparams": { + "group_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": 0.011605764739215374, + "qparams": { + "group_size": { + "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": 0.009042241610586643, + "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": 0.04448961466550827, + "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": 0.04448961466550827, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.22719286382198334, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21319237351417542, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.20885087549686432, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1895359307527542, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10659855604171753, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10163213312625885, + "qparams": { + "group_size": { + "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": 0.11870637536048889, + "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": 0.10963312536478043, + "qparams": { + "group_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": 0.10810388624668121, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09590505063533783, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09091156721115112, + "qparams": { + "group_size": { + "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": 0.06024644523859024, + "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": 0.05236267298460007, + "qparams": { + "group_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": 0.05099468678236008, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05067155882716179, + "qparams": { + "group_size": { + "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": 0.030104901641607285, + "qparams": { + "group_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": 0.026020998135209084, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025926515460014343, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023809032514691353, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023606397211551666, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015606590546667576, + "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": 0.015382136218249798, + "qparams": { + "group_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": 0.015132106840610504, + "qparams": { + "group_size": { + "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": 0.00974990800023079, + "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": 0.05236267298460007, + "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": 0.05236267298460007, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2543743848800659, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2262922078371048, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21531721949577332, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18895044922828674, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11800484359264374, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10661186277866364, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13909654319286346, + "qparams": { + "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": 0.12779659032821655, + "qparams": { + "group_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": 0.12184534221887589, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09989941865205765, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09354714304208755, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07155216485261917, + "qparams": { + "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": 0.06175753101706505, + "qparams": { + "group_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": 0.05724445730447769, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05613947659730911, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.0362374521791935, + "qparams": { + "group_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": 0.03055511601269245, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.030177870765328407, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02708377316594124, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026365462690591812, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.020225297659635544, + "qparams": { + "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": 0.02045154757797718, + "qparams": { + "group_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": 0.018858499825000763, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015059763565659523, + "qparams": { + "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": 0.0362374521791935, + "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": 0.0362374521791935, + "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": 0.1159093827009201, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1079614907503128, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10493112355470657, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09449709951877594, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05398081988096237, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.050882987678050995, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.061348363757133484, + "qparams": { + "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": 0.05664729326963425, + "qparams": { + "group_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": 0.05483487993478775, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04814782366156578, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.045539695769548416, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.031155647709965706, + "qparams": { + "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": 0.027073992416262627, + "qparams": { + "group_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": 0.02586176060140133, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.025571925565600395, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.015589497983455658, + "qparams": { + "group_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": 0.013313796371221542, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013200363144278526, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01213385071605444, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011950455605983734, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008134379051625729, + "qparams": { + "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": 0.008148596622049809, + "qparams": { + "group_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": 0.007722825277596712, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0053601195104420185, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05398081988096237, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05398081988096237, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.09544687718153, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08897733688354492, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08624942600727081, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.07777231931686401, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.044479843229055405, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04182204231619835, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05094916746020317, + "qparams": { + "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": 0.047092702239751816, + "qparams": { + "group_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": 0.04518789425492287, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0397372804582119, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.037693385034799576, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.025862768292427063, + "qparams": { + "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": 0.02250760793685913, + "qparams": { + "group_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": 0.021320628002285957, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.021036705002188683, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.012928266078233719, + "qparams": { + "group_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": 0.01097314152866602, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010853956453502178, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010012144222855568, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009829926304519176, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006748787593096495, + "qparams": { + "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": 0.006749298889189959, + "qparams": { + "group_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": 0.006368409842252731, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004417881369590759, + "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": 0.05094916746020317, + "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": 0.05094916746020317, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2179342657327652, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20333079993724823, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19824206829071045, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17877376079559326, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10185681283473969, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09634695947170258, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11521556973457336, + "qparams": { + "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": 0.10602506250143051, + "qparams": { + "group_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": 0.10343112051486969, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.090951107442379, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08603246510028839, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05851443484425545, + "qparams": { + "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": 0.05070232227444649, + "qparams": { + "group_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": 0.04875481501221657, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.048295535147190094, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.029279887676239014, + "qparams": { + "group_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": 0.024961311370134354, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024787327274680138, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.022729435935616493, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0224265418946743, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.015228992328047752, + "qparams": { + "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": 0.01499471627175808, + "qparams": { + "group_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": 0.014527644962072372, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.009604893624782562, + "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": 0.05070232227444649, + "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": 0.05070232227444649, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2316020131111145, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20625807344913483, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1973475068807602, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17265214025974274, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10853278636932373, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0986219197511673, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12619367241859436, + "qparams": { + "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": 0.11568791419267654, + "qparams": { + "group_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": 0.11142948269844055, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09159255027770996, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08445370942354202, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06482509523630142, + "qparams": { + "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": 0.05616588145494461, + "qparams": { + "group_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": 0.05296119302511215, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05219074338674545, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03260833024978638, + "qparams": { + "group_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": 0.0286873672157526, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02842659503221512, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025504885241389275, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025032881647348404, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01793479733169079, + "qparams": { + "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": 0.019558411091566086, + "qparams": { + "group_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": 0.016936080530285835, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015047561377286911, + "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": 0.05296119302511215, + "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": 0.05296119302511215, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.1648566871881485, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15464013814926147, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1513255387544632, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13727276027202606, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0776795893907547, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07399512827396393, + "qparams": { + "group_size": { + "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": 0.0865662693977356, + "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": 0.08005155622959137, + "qparams": { + "group_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": 0.07877101749181747, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06986814737319946, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06633317470550537, + "qparams": { + "group_size": { + "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": 0.04426577687263489, + "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": 0.03853688761591911, + "qparams": { + "group_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": 0.037473950535058975, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03722295165061951, + "qparams": { + "group_size": { + "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": 0.022173531353473663, + "qparams": { + "group_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": 0.019765131175518036, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0196919497102499, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018235061317682266, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018084973096847534, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011922821402549744, + "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": 0.012639745138585567, + "qparams": { + "group_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": 0.01157198753207922, + "qparams": { + "group_size": { + "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": 0.009219778701663017, + "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": 0.04426577687263489, + "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": 0.04426577687263489, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.22701239585876465, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21294912695884705, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.20853587985038757, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18915005028247833, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10659844428300858, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10156892240047455, + "qparams": { + "group_size": { + "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": 0.11866264790296555, + "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": 0.10961359739303589, + "qparams": { + "group_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": 0.10808949917554855, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09580609202384949, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09085095673799515, + "qparams": { + "group_size": { + "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": 0.060330137610435486, + "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": 0.05236116796731949, + "qparams": { + "group_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": 0.050979677587747574, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.050656430423259735, + "qparams": { + "group_size": { + "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": 0.030141523107886314, + "qparams": { + "group_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": 0.025982705876231194, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02588672563433647, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023755354806780815, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023548975586891174, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015631087124347687, + "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": 0.01531226560473442, + "qparams": { + "group_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": 0.01514601893723011, + "qparams": { + "group_size": { + "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": 0.009635680355131626, + "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": 0.05236116796731949, + "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": 0.05236116796731949, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.25063207745552063, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2232481837272644, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21256259083747864, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1871233880519867, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11640055477619171, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10526542365550995, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13716313242912292, + "qparams": { + "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": 0.1258689910173416, + "qparams": { + "group_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": 0.12018230557441711, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09885655343532562, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09280452877283096, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07065563648939133, + "qparams": { + "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": 0.06088196486234665, + "qparams": { + "group_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": 0.05651559308171272, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.055450670421123505, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03590194880962372, + "qparams": { + "group_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": 0.03023979626595974, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029878271743655205, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026884205639362335, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026194948703050613, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.02022593654692173, + "qparams": { + "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": 0.020320363342761993, + "qparams": { + "group_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": 0.01890011504292488, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015071042813360691, + "qparams": { + "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": 0.03590194880962372, + "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": 0.03590194880962372, + "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": 0.12276840209960938, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11438333988189697, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11121811717748642, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10023435205221176, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.057224784046411514, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05393899232149124, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.0649670735001564, + "qparams": { + "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": 0.05995824560523033, + "qparams": { + "group_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": 0.05810554325580597, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05106721073389053, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04835033044219017, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03305141627788544, + "qparams": { + "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": 0.02872057817876339, + "qparams": { + "group_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": 0.02746760845184326, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.027170278131961823, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.0165332593023777, + "qparams": { + "group_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": 0.014258146286010742, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01414039358496666, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013030449859797955, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01284382026642561, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008700430393218994, + "qparams": { + "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": 0.008883959613740444, + "qparams": { + "group_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": 0.008279599249362946, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.006063423585146666, + "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": 0.05106721073389053, + "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": 0.05106721073389053, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09807553142309189, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09139134734869003, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08843209594488144, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.07974173873662949, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04564545676112175, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04288041591644287, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05255328118801117, + "qparams": { + "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": 0.04854630306363106, + "qparams": { + "group_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": 0.04638899117708206, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04078986123204231, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0387263260781765, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.026684507727622986, + "qparams": { + "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": 0.02320924960076809, + "qparams": { + "group_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": 0.02191184088587761, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.021596485748887062, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013348329812288284, + "qparams": { + "group_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": 0.0113051263615489, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011168763972818851, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010321244597434998, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010121911764144897, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0069910152815282345, + "qparams": { + "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": 0.007004198618233204, + "qparams": { + "group_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": 0.006582589820027351, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004623998422175646, + "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": 0.05255328118801117, + "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": 0.05255328118801117, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2279258817434311, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21286237239837646, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20759904384613037, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18733899295330048, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10668858140707016, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10103938728570938, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12051427364349365, + "qparams": { + "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": 0.11081913858652115, + "qparams": { + "group_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": 0.10832711309194565, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09540797770023346, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09025140106678009, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.061296455562114716, + "qparams": { + "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": 0.0530320480465889, + "qparams": { + "group_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": 0.05109944939613342, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.050629232078790665, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.030643556267023087, + "qparams": { + "group_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": 0.02616894617676735, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026000957936048508, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02385120652616024, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023548457771539688, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01595214568078518, + "qparams": { + "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": 0.01573142409324646, + "qparams": { + "group_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": 0.015244923532009125, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010122938081622124, + "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": 0.0530320480465889, + "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": 0.0530320480465889, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.23249410092830658, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20550866425037384, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1966814547777176, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16861602663993835, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10906998068094254, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09917015582323074, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12667454779148102, + "qparams": { + "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": 0.11561643332242966, + "qparams": { + "group_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": 0.11232500523328781, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09064875543117523, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0831482782959938, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06574496626853943, + "qparams": { + "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": 0.05652159824967384, + "qparams": { + "group_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": 0.05351713299751282, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05279965326189995, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.033245258033275604, + "qparams": { + "group_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": 0.02922673150897026, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02903626300394535, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025817718356847763, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0253668911755085, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.018665608018636703, + "qparams": { + "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": 0.02019353397190571, + "qparams": { + "group_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": 0.017771068960428238, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01582319661974907, + "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": 0.05351713299751282, + "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": 0.05351713299751282, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.1644126921892166, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1541285216808319, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1508168876171112, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.136811301112175, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0775391012430191, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0738193690776825, + "qparams": { + "group_size": { + "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": 0.08638562262058258, + "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": 0.07987150549888611, + "qparams": { + "group_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": 0.0786459818482399, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06969477236270905, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06613218039274216, + "qparams": { + "group_size": { + "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": 0.044169921427965164, + "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": 0.038472022861242294, + "qparams": { + "group_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": 0.037410542368888855, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03716311603784561, + "qparams": { + "group_size": { + "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": 0.022140534594655037, + "qparams": { + "group_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": 0.019701344892382622, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019623108208179474, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01815955340862274, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018005017191171646, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011912578716874123, + "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": 0.012567946687340736, + "qparams": { + "group_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": 0.011569226160645485, + "qparams": { + "group_size": { + "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": 0.009125723503530025, + "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": 0.044169921427965164, + "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": 0.044169921427965164, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.22744639217853546, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21330395340919495, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.20888864994049072, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18950854241847992, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10683050751686096, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1017865464091301, + "qparams": { + "group_size": { + "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": 0.11879006773233414, + "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": 0.10981673002243042, + "qparams": { + "group_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": 0.10836070030927658, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09598791599273682, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0909748300909996, + "qparams": { + "group_size": { + "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": 0.06042467802762985, + "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": 0.05246065557003021, + "qparams": { + "group_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": 0.05109531804919243, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05077332630753517, + "qparams": { + "group_size": { + "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": 0.030188539996743202, + "qparams": { + "group_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": 0.02604488842189312, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025951657444238663, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023808462545275688, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023597434163093567, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01567772962152958, + "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": 0.015349740162491798, + "qparams": { + "group_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": 0.015208550728857517, + "qparams": { + "group_size": { + "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": 0.00965795200318098, + "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": 0.05246065557003021, + "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": 0.05246065557003021, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2526835799217224, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22457508742809296, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21378640830516815, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18830430507659912, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11689647287130356, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10568614304065704, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1381731629371643, + "qparams": { + "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": 0.12662753462791443, + "qparams": { + "group_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": 0.12081700563430786, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09928776323795319, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09328600019216537, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07078490406274796, + "qparams": { + "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": 0.061073414981365204, + "qparams": { + "group_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": 0.05665760487318039, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05557241663336754, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.035600416362285614, + "qparams": { + "group_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": 0.030172107741236687, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029802775010466576, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026774631813168526, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026064284145832062, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019465811550617218, + "qparams": { + "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": 0.020099185407161713, + "qparams": { + "group_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": 0.018058085814118385, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014695587567985058, + "qparams": { + "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": 0.035600416362285614, + "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": 0.035600416362285614, + "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": 0.12451472133398056, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11616980284452438, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11297300457954407, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10200241208076477, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.058078303933143616, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05485674738883972, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06581670790910721, + "qparams": { + "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": 0.06078829988837242, + "qparams": { + "group_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": 0.058978497982025146, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05194691941142082, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04915093258023262, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03344649076461792, + "qparams": { + "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": 0.02905847318470478, + "qparams": { + "group_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": 0.027839036658406258, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02754611149430275, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01672748290002346, + "qparams": { + "group_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": 0.014336660504341125, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014226974919438362, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013091678731143475, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012904316186904907, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008734490722417831, + "qparams": { + "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": 0.008767681196331978, + "qparams": { + "group_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": 0.008314236998558044, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0057906233705580235, + "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": 0.05194691941142082, + "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": 0.05194691941142082, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.09788767993450165, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.0913216844201088, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.0884811282157898, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.07989095896482468, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.045632559806108475, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.042911674827337265, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05230686441063881, + "qparams": { + "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": 0.04837960749864578, + "qparams": { + "group_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": 0.046335458755493164, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0408146008849144, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03871094807982445, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02653556875884533, + "qparams": { + "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": 0.0230989009141922, + "qparams": { + "group_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": 0.0218697190284729, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.021569406613707542, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013265399262309074, + "qparams": { + "group_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": 0.0112417321652174, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011112191714346409, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010265444405376911, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010072953067719936, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0069145564921200275, + "qparams": { + "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": 0.006894981954246759, + "qparams": { + "group_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": 0.006519767921417952, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004482707940042019, + "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": 0.05230686441063881, + "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": 0.05230686441063881, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.23075620830059052, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21576111018657684, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21033869683742523, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19011947512626648, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10825590044260025, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10244671255350113, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12250781059265137, + "qparams": { + "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": 0.11270033568143845, + "qparams": { + "group_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": 0.10986828804016113, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09692059457302094, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09182000160217285, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06229348108172417, + "qparams": { + "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": 0.05396422743797302, + "qparams": { + "group_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": 0.05183655768632889, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05133725702762604, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03114703856408596, + "qparams": { + "group_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": 0.026490161195397377, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026291655376553535, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02415180206298828, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023828091099858284, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016184039413928986, + "qparams": { + "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": 0.015856336802244186, + "qparams": { + "group_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": 0.015429510734975338, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010057945735752583, + "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": 0.05396422743797302, + "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": 0.05396422743797302, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2378692328929901, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2145403027534485, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20701779425144196, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1777936965227127, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11116304248571396, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10239820927381516, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12721991539001465, + "qparams": { + "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": 0.1163020208477974, + "qparams": { + "group_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": 0.11363708972930908, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09384462237358093, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.085903100669384, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06585049629211426, + "qparams": { + "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": 0.0569179505109787, + "qparams": { + "group_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": 0.054586198180913925, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.054015837609767914, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.033693358302116394, + "qparams": { + "group_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": 0.03003300540149212, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029885221272706985, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026866208761930466, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026524167507886887, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019592495635151863, + "qparams": { + "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": 0.020903104916214943, + "qparams": { + "group_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": 0.018907465040683746, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01669202744960785, + "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": 0.054586198180913925, + "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": 0.054586198180913925, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.16945967078208923, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1589520126581192, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15556354820728302, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14117729663848877, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07985977828502655, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07603935152292252, + "qparams": { + "group_size": { + "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": 0.08894848823547363, + "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": 0.08227132260799408, + "qparams": { + "group_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": 0.08100009709596634, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07179353386163712, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06814122200012207, + "qparams": { + "group_size": { + "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": 0.04545842483639717, + "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": 0.039551135152578354, + "qparams": { + "group_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": 0.03847257420420647, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03821241855621338, + "qparams": { + "group_size": { + "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": 0.022750262171030045, + "qparams": { + "group_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": 0.020142171531915665, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02006755955517292, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01853906363248825, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018384192138910294, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012151684612035751, + "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": 0.012686717323958874, + "qparams": { + "group_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": 0.011793715879321098, + "qparams": { + "group_size": { + "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": 0.009036828763782978, + "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": 0.04545842483639717, + "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": 0.04545842483639717, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.22825723886489868, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2141815572977066, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.20978359878063202, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19038282334804535, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10725879669189453, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10221412777900696, + "qparams": { + "group_size": { + "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": 0.11923383921384811, + "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": 0.11025045812129974, + "qparams": { + "group_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": 0.10877066105604172, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09642859548330307, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0914444550871849, + "qparams": { + "group_size": { + "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": 0.06059536710381508, + "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": 0.05268415808677673, + "qparams": { + "group_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": 0.051308032125234604, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05098205804824829, + "qparams": { + "group_size": { + "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": 0.030262233689427376, + "qparams": { + "group_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": 0.026145905256271362, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026050018146634102, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02391122654080391, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023701544851064682, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015655890107154846, + "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": 0.015402043238282204, + "qparams": { + "group_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": 0.015184872783720493, + "qparams": { + "group_size": { + "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": 0.009685229510068893, + "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": 0.05268415808677673, + "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": 0.05268415808677673, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2545127272605896, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22504378855228424, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21346145868301392, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18770352005958557, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11768563091754913, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10571326315402985, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1393214464187622, + "qparams": { + "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": 0.1280357837677002, + "qparams": { + "group_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": 0.12180899828672409, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.099358931183815, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09313693642616272, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07142234593629837, + "qparams": { + "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": 0.061698682606220245, + "qparams": { + "group_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": 0.056971050798892975, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05580520257353783, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03611058369278908, + "qparams": { + "group_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": 0.03019261360168457, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029790867120027542, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026634762063622475, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025862039998173714, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01987280882894993, + "qparams": { + "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": 0.01997644454240799, + "qparams": { + "group_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": 0.018401378765702248, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014370234683156013, + "qparams": { + "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": 0.03611058369278908, + "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": 0.03611058369278908, + "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": 0.12003068625926971, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11203113198280334, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10893034189939499, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09842952340841293, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05603238195180893, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05292164534330368, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06354755163192749, + "qparams": { + "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": 0.058687176555395126, + "qparams": { + "group_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": 0.05689121410250664, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05011115223169327, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.047527533024549484, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03229081258177757, + "qparams": { + "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": 0.028039341792464256, + "qparams": { + "group_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": 0.02685135416686535, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02656432054936886, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01615782640874386, + "qparams": { + "group_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": 0.01381139736622572, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01370344590395689, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.012616032734513283, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012434285134077072, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008440444245934486, + "qparams": { + "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": 0.00842669140547514, + "qparams": { + "group_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": 0.008043549954891205, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005534138530492783, + "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": 0.05011115223169327, + "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": 0.05011115223169327, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.09737459570169449, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09090263396501541, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08810397982597351, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.07957137376070023, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.045413579791784286, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04272455349564552, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.051979437470436096, + "qparams": { + "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": 0.04810919240117073, + "qparams": { + "group_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": 0.04611467570066452, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04062846675515175, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.038568560034036636, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02636142633855343, + "qparams": { + "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": 0.022979870438575745, + "qparams": { + "group_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": 0.021774442866444588, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.021480141207575798, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013185464777052402, + "qparams": { + "group_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": 0.011205343529582024, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011079264804720879, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010234725661575794, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01004746276885271, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006879559718072414, + "qparams": { + "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": 0.006882158108055592, + "qparams": { + "group_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": 0.0064974878914654255, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004496802110224962, + "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": 0.051979437470436096, + "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": 0.051979437470436096, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2358996719121933, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22071634232997894, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21533000469207764, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19489140808582306, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11071367561817169, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10490909218788147, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12489936500787735, + "qparams": { + "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": 0.11496555805206299, + "qparams": { + "group_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": 0.11233188211917877, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09917987138032913, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09398877620697021, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06351965665817261, + "qparams": { + "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": 0.054995615035295486, + "qparams": { + "group_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": 0.05300584435462952, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05252481624484062, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03175141289830208, + "qparams": { + "group_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": 0.027061017230153084, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026893436908721924, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024688955396413803, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024379778653383255, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016460571438074112, + "qparams": { + "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": 0.016135629266500473, + "qparams": { + "group_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": 0.015727797523140907, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010208791121840477, + "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": 0.05300584435462952, + "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": 0.05300584435462952, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.22883830964565277, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20614838600158691, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19631795585155487, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17070499062538147, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10725858807563782, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09757325798273087, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12847140431404114, + "qparams": { + "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": 0.11707864701747894, + "qparams": { + "group_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": 0.11022605001926422, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09129232913255692, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08431360125541687, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0659470185637474, + "qparams": { + "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": 0.056595634669065475, + "qparams": { + "group_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": 0.05212985724210739, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05103999748826027, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03310806676745415, + "qparams": { + "group_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": 0.02778417430818081, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027371233329176903, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024677520617842674, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02400573343038559, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017836380749940872, + "qparams": { + "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": 0.018613560125231743, + "qparams": { + "group_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": 0.016372842714190483, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0136397210881114, + "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": 0.05212985724210739, + "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": 0.05212985724210739, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.17070423066616058, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1601247489452362, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1567060351371765, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14220643043518066, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0805446058511734, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07670557498931885, + "qparams": { + "group_size": { + "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": 0.08963237702846527, + "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": 0.08294012397527695, + "qparams": { + "group_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": 0.08167549222707748, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07240411639213562, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06871314346790314, + "qparams": { + "group_size": { + "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": 0.04582861065864563, + "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": 0.03991442918777466, + "qparams": { + "group_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": 0.03882840648293495, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03857194259762764, + "qparams": { + "group_size": { + "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": 0.022968223318457603, + "qparams": { + "group_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": 0.02039516717195511, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02031974494457245, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018786322325468063, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018626917153596878, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012307541444897652, + "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": 0.012937263585627079, + "qparams": { + "group_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": 0.011955013498663902, + "qparams": { + "group_size": { + "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": 0.009316863492131233, + "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": 0.04582861065864563, + "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": 0.04582861065864563, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.23005414009094238, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2158929854631424, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21145154535770416, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19184234738349915, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10818520933389664, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10310325771570206, + "qparams": { + "group_size": { + "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": 0.12027016282081604, + "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": 0.11120354384183884, + "qparams": { + "group_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": 0.10968996584415436, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0972684696316719, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09218144416809082, + "qparams": { + "group_size": { + "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": 0.06114460900425911, + "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": 0.05313685163855553, + "qparams": { + "group_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": 0.051769159734249115, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05143589526414871, + "qparams": { + "group_size": { + "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": 0.030546458438038826, + "qparams": { + "group_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": 0.02641225978732109, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02631765976548195, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024157334119081497, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02395021729171276, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01584354043006897, + "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": 0.01560877449810505, + "qparams": { + "group_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": 0.01537677738815546, + "qparams": { + "group_size": { + "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": 0.00988816749304533, + "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": 0.05313685163855553, + "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": 0.05313685163855553, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2560979127883911, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2266838401556015, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21510906517505646, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18948157131671906, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11844578385353088, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10658077895641327, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1402490735054016, + "qparams": { + "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": 0.1288377046585083, + "qparams": { + "group_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": 0.12267769873142242, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10011512786149979, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09405114501714706, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07202824205160141, + "qparams": { + "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": 0.062029946595430374, + "qparams": { + "group_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": 0.05725345015525818, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05610313639044762, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.036371950060129166, + "qparams": { + "group_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": 0.030240055173635483, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029833057895302773, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02666296251118183, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02587430737912655, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.020037880167365074, + "qparams": { + "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": 0.0198736023157835, + "qparams": { + "group_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": 0.018557829782366753, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014110292308032513, + "qparams": { + "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": 0.036371950060129166, + "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": 0.036371950060129166, + "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": 0.12764336168766022, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1192680150270462, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11607543379068375, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10495202243328094, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05972214415669441, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05651498958468437, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06759195029735565, + "qparams": { + "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": 0.062371790409088135, + "qparams": { + "group_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": 0.06061685457825661, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05347774177789688, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05077451840043068, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03441784903407097, + "qparams": { + "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": 0.029874319210648537, + "qparams": { + "group_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": 0.02867325209081173, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02838963270187378, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.017238063737750053, + "qparams": { + "group_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": 0.01483115740120411, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014726657420396805, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013577736914157867, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013398392125964165, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009049257263541222, + "qparams": { + "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": 0.009154352359473705, + "qparams": { + "group_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": 0.008648284710943699, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.006184062920510769, + "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": 0.05347774177789688, + "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": 0.05347774177789688, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.10252892971038818, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09576575458049774, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09289272874593735, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08394007384777069, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04789251461625099, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04511416703462601, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05465804785490036, + "qparams": { + "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": 0.05058291554450989, + "qparams": { + "group_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": 0.048620518296957016, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04287077486515045, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.040685031563043594, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02774953469634056, + "qparams": { + "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": 0.02417556755244732, + "qparams": { + "group_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": 0.022961415350437164, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.022672811523079872, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013882599771022797, + "qparams": { + "group_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": 0.011817790567874908, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011696689762175083, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010801131837069988, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010616479441523552, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007246277295053005, + "qparams": { + "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": 0.007250087335705757, + "qparams": { + "group_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": 0.006863855291157961, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004754131659865379, + "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": 0.05465804785490036, + "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": 0.05465804785490036, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "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": 0.24701283872127533, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23132389783859253, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22596144676208496, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20463497936725616, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11614230275154114, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11027976870536804, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13069172203540802, + "qparams": { + "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": 0.12037727236747742, + "qparams": { + "group_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": 0.11786208301782608, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1041986271739006, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09882481396198273, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06646964699029922, + "qparams": { + "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": 0.05758286267518997, + "qparams": { + "group_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": 0.05560174956917763, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05514172837138176, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03320659324526787, + "qparams": { + "group_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": 0.028370248153805733, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0282001793384552, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025899790227413177, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02559661492705345, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017176056280732155, + "qparams": { + "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": 0.016841422766447067, + "qparams": { + "group_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": 0.016454501077532768, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010606949217617512, + "qparams": { + "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": 0.03320659324526787, + "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": 0.03320659324526787, + "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": 0.23188184201717377, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2040960043668747, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19489829242229462, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16376931965351105, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10777938365936279, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09729942679405212, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12566745281219482, + "qparams": { + "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": 0.11443621665239334, + "qparams": { + "group_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": 0.11118555068969727, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08683770149946213, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07858094573020935, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06480423361063004, + "qparams": { + "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": 0.05563483387231827, + "qparams": { + "group_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": 0.05259286239743233, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.051864445209503174, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.032871924340724945, + "qparams": { + "group_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": 0.028458692133426666, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02826067991554737, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024535775184631348, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024057110771536827, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01858743466436863, + "qparams": { + "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": 0.0193475391715765, + "qparams": { + "group_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": 0.017686165869235992, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014840971678495407, + "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": 0.05259286239743233, + "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": 0.05259286239743233, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.1682455837726593, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15800228714942932, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1547509729862213, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14054182171821594, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.079339399933815, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07559914886951447, + "qparams": { + "group_size": { + "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": 0.08833207935094833, + "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": 0.08165712654590607, + "qparams": { + "group_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": 0.08045759797096252, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07138161361217499, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06784658133983612, + "qparams": { + "group_size": { + "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": 0.04510077089071274, + "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": 0.039260003715753555, + "qparams": { + "group_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": 0.03821207582950592, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.037961263209581375, + "qparams": { + "group_size": { + "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": 0.022595804184675217, + "qparams": { + "group_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": 0.019998112693428993, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01992572844028473, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01842043548822403, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018265215680003166, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012073345482349396, + "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": 0.012577490881085396, + "qparams": { + "group_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": 0.011734707280993462, + "qparams": { + "group_size": { + "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": 0.008953314274549484, + "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": 0.04510077089071274, + "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": 0.04510077089071274, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.22691476345062256, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21312344074249268, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.20887191593647003, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18963658809661865, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10664501786231995, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10172374546527863, + "qparams": { + "group_size": { + "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": 0.11860771477222443, + "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": 0.10958611220121384, + "qparams": { + "group_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": 0.10815861076116562, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09599953144788742, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09114781022071838, + "qparams": { + "group_size": { + "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": 0.06029902771115303, + "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": 0.0523652508854866, + "qparams": { + "group_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": 0.05103211849927902, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05072113126516342, + "qparams": { + "group_size": { + "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": 0.030172957107424736, + "qparams": { + "group_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": 0.026036448776721954, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02595346048474312, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023834919556975365, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023632455617189407, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015749722719192505, + "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": 0.01537692453712225, + "qparams": { + "group_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": 0.015303075313568115, + "qparams": { + "group_size": { + "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": 0.009740648791193962, + "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": 0.0523652508854866, + "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": 0.0523652508854866, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.24918413162231445, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21933238208293915, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20707868039608002, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18174654245376587, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11494186520576477, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10255999118089676, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13768699765205383, + "qparams": { + "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": 0.12625989317893982, + "qparams": { + "group_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": 0.11924880743026733, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09656966477632523, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09071796387434006, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07066839933395386, + "qparams": { + "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": 0.06081222742795944, + "qparams": { + "group_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": 0.055635519325733185, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05435940995812416, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03570428863167763, + "qparams": { + "group_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": 0.029463455080986023, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02900095097720623, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025888066738843918, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025036271661520004, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01965976320207119, + "qparams": { + "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": 0.01955435611307621, + "qparams": { + "group_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": 0.018027206882834435, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013959326781332493, + "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": 0.05435940995812416, + "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": 0.05435940995812416, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.1247434988617897, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11670483648777008, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11364098638296127, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10287600755691528, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05828666687011719, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.055208027362823486, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06605937331914902, + "qparams": { + "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": 0.060932084918022156, + "qparams": { + "group_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": 0.05917009711265564, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05236673727631569, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04976861551403999, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03360184282064438, + "qparams": { + "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": 0.029146570712327957, + "qparams": { + "group_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": 0.027942713350057602, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02765742316842079, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.016804853454232216, + "qparams": { + "group_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": 0.014362158253788948, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014248182997107506, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013155810534954071, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012974121607840061, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008764353580772877, + "qparams": { + "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": 0.008737596683204174, + "qparams": { + "group_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": 0.008365233428776264, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.00572108943015337, + "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": 0.05236673727631569, + "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": 0.05236673727631569, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.10045016556978226, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09394112229347229, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09111958742141724, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08241774141788483, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04690574109554291, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04422730579972267, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05363238602876663, + "qparams": { + "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": 0.049604736268520355, + "qparams": { + "group_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": 0.047610457986593246, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04209963604807854, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.040018994361162186, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.027225106954574585, + "qparams": { + "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": 0.023721370846033096, + "qparams": { + "group_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": 0.02248840406537056, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0221969336271286, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013622920960187912, + "qparams": { + "group_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": 0.011588672176003456, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011461575515568256, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010613450780510902, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010426003485918045, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.00712403142824769, + "qparams": { + "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": 0.007135761436074972, + "qparams": { + "group_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": 0.006737841293215752, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004690940026193857, + "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": 0.05363238602876663, + "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": 0.05363238602876663, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2407330572605133, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22585603594779968, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22081398963928223, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20022806525230408, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11315697431564331, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10761665552854538, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12703271210193634, + "qparams": { + "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": 0.11701583862304688, + "qparams": { + "group_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": 0.11473140120506287, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10170383751392365, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0965653508901596, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06458360701799393, + "qparams": { + "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": 0.05592375248670578, + "qparams": { + "group_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": 0.054129619151353836, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.053702425211668015, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03226263076066971, + "qparams": { + "group_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": 0.027593648061156273, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0274448711425066, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02522885799407959, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024957189336419106, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016672074794769287, + "qparams": { + "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": 0.016333870589733124, + "qparams": { + "group_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": 0.01600068435072899, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010252362117171288, + "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": 0.054129619151353836, + "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": 0.054129619151353836, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.23090194165706635, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20458285510540009, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19626623392105103, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.161235049366951, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10844495892524719, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09883848577737808, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12446988373994827, + "qparams": { + "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": 0.11391567438840866, + "qparams": { + "group_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": 0.11124526709318161, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08815298974514008, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0763155072927475, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06419172883033752, + "qparams": { + "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": 0.05544794723391533, + "qparams": { + "group_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": 0.052954819053411484, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0523681566119194, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.032384242862463, + "qparams": { + "group_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": 0.028600363060832024, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028440048918128014, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02476891502737999, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02438882179558277, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.018072044476866722, + "qparams": { + "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": 0.019314924255013466, + "qparams": { + "group_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": 0.0173446387052536, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014859561808407307, + "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": 0.052954819053411484, + "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": 0.052954819053411484, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.1738031804561615, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.16322264075279236, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1598883718252182, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14521893858909607, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08195927739143372, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07812372595071793, + "qparams": { + "group_size": { + "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": 0.09113750606775284, + "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": 0.08429676294326782, + "qparams": { + "group_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": 0.08309756964445114, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07375714182853699, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07007640600204468, + "qparams": { + "group_size": { + "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": 0.046505432575941086, + "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": 0.04047655686736107, + "qparams": { + "group_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": 0.03942807391285896, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0391780324280262, + "qparams": { + "group_size": { + "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": 0.02332654409110546, + "qparams": { + "group_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": 0.02055349200963974, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020480815321207047, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01892334595322609, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01876964047551155, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012469031848013401, + "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": 0.012807806953787804, + "qparams": { + "group_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": 0.012125118635594845, + "qparams": { + "group_size": { + "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": 0.00899199303239584, + "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": 0.046505432575941086, + "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": 0.046505432575941086, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.23443186283111572, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22027404606342316, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21588706970214844, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19614525139331818, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11022452265024185, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10515344142913818, + "qparams": { + "group_size": { + "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": 0.12258116900920868, + "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": 0.11317969858646393, + "qparams": { + "group_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": 0.11172617226839066, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09927696734666824, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09427142888307571, + "qparams": { + "group_size": { + "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": 0.06229319050908089, + "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": 0.054100196808576584, + "qparams": { + "group_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": 0.05275103449821472, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.052436042577028275, + "qparams": { + "group_size": { + "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": 0.03121803142130375, + "qparams": { + "group_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": 0.02694082260131836, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02684902213513851, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024684512987732887, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02448016032576561, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016377834603190422, + "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": 0.015938812866806984, + "qparams": { + "group_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": 0.01593790017068386, + "qparams": { + "group_size": { + "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": 0.01015576533973217, + "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": 0.054100196808576584, + "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": 0.054100196808576584, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2544979453086853, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2234029620885849, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21073946356773376, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18461494147777557, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11736588925123215, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10444607585668564, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.140091672539711, + "qparams": { + "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": 0.12905777990818024, + "qparams": { + "group_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": 0.12188544124364853, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09829459339380264, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09242445230484009, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07185409963130951, + "qparams": { + "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": 0.062401000410318375, + "qparams": { + "group_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": 0.0569782480597496, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05563053488731384, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.036287251859903336, + "qparams": { + "group_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": 0.030455609783530235, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02997898869216442, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02679538168013096, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025914499536156654, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019611716270446777, + "qparams": { + "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": 0.020561207085847855, + "qparams": { + "group_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": 0.017828719690442085, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015062633901834488, + "qparams": { + "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": 0.036287251859903336, + "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": 0.036287251859903336, + "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": 0.1251174658536911, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11724715679883957, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11418858170509338, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1034843921661377, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05854576826095581, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0554809495806694, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06628162413835526, + "qparams": { + "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": 0.06114521622657776, + "qparams": { + "group_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": 0.05941658839583397, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05262251943349838, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05007287859916687, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03371120244264603, + "qparams": { + "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": 0.029258856549859047, + "qparams": { + "group_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": 0.028073646128177643, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.027789754793047905, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.016866452991962433, + "qparams": { + "group_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": 0.014457948505878448, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014346457086503506, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013259127736091614, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013077843934297562, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008830240927636623, + "qparams": { + "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": 0.008835313841700554, + "qparams": { + "group_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": 0.008444799110293388, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005837839562445879, + "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": 0.05262251943349838, + "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": 0.05262251943349838, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10384964197874069, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09729038923978806, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09451214224100113, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08557596802711487, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.048562850803136826, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0458730086684227, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05533070117235184, + "qparams": { + "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": 0.051142677664756775, + "qparams": { + "group_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": 0.04926951602101326, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04361788183450699, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04148295521736145, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.028102964162826538, + "qparams": { + "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": 0.02444986253976822, + "qparams": { + "group_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": 0.023264307528734207, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.022984663024544716, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014047850854694843, + "qparams": { + "group_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": 0.011957906186580658, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011840284802019596, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0109567791223526, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010777895338833332, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007329082116484642, + "qparams": { + "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": 0.007300631143152714, + "qparams": { + "group_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": 0.006951743271201849, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004750834312289953, + "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": 0.051142677664756775, + "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": 0.051142677664756775, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2438056319952011, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22888967394828796, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22400321066379547, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20324450731277466, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11460530757904053, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10909276455640793, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12842528522014618, + "qparams": { + "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": 0.11827848851680756, + "qparams": { + "group_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": 0.11619982868432999, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10310618579387665, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09797562658786774, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06528449803590775, + "qparams": { + "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": 0.05650123581290245, + "qparams": { + "group_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": 0.05479864776134491, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.054387416690588, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.0325801782310009, + "qparams": { + "group_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": 0.02784855291247368, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027713973075151443, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025466129183769226, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02520359493792057, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016764633357524872, + "qparams": { + "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": 0.016342082992196083, + "qparams": { + "group_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": 0.016124768182635307, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010098564438521862, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054387416690588, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054387416690588, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.241488516330719, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20706094801425934, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19562792778015137, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16180016100406647, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11246494203805923, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09899784624576569, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13099324703216553, + "qparams": { + "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": 0.12007168680429459, + "qparams": { + "group_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": 0.11650204658508301, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08852086961269379, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07903455197811127, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06779814511537552, + "qparams": { + "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": 0.05868835002183914, + "qparams": { + "group_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": 0.05526489391922951, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05442359670996666, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.034554872661828995, + "qparams": { + "group_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": 0.030470093712210655, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.03027106262743473, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02614763006567955, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025636082515120506, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019951775670051575, + "qparams": { + "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": 0.02139880135655403, + "qparams": { + "group_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": 0.01897195354104042, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.017018228769302368, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05442359670996666, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05442359670996666, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.18203367292881012, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1710335910320282, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1675550937652588, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15236406028270721, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08587610721588135, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08190087974071503, + "qparams": { + "group_size": { + "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": 0.09535513818264008, + "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": 0.08825366944074631, + "qparams": { + "group_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": 0.0870676189661026, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07734258472919464, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07345744967460632, + "qparams": { + "group_size": { + "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": 0.048645634204149246, + "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": 0.042352598160505295, + "qparams": { + "group_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": 0.04127296432852745, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.041016582399606705, + "qparams": { + "group_size": { + "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": 0.024328989908099174, + "qparams": { + "group_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": 0.02144465409219265, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021370714530348778, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.019735010340809822, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0195760577917099, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012863144278526306, + "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": 0.013263463042676449, + "qparams": { + "group_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": 0.012506713159382343, + "qparams": { + "group_size": { + "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": 0.009193268604576588, + "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": 0.048645634204149246, + "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": 0.048645634204149246, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.23911651968955994, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22470882534980774, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22028981149196625, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20028521120548248, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.1125127524137497, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10735974460840225, + "qparams": { + "group_size": { + "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": 0.12472090870141983, + "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": 0.11548600345849991, + "qparams": { + "group_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": 0.1140332892537117, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1013857051730156, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09616755694150925, + "qparams": { + "group_size": { + "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": 0.06340529769659042, + "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": 0.055172793567180634, + "qparams": { + "group_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": 0.05381399765610695, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.053499914705753326, + "qparams": { + "group_size": { + "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": 0.03165449947118759, + "qparams": { + "group_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": 0.02741776965558529, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027326589450240135, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025114046409726143, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024908067658543587, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016347892582416534, + "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": 0.016122182831168175, + "qparams": { + "group_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": 0.015894096344709396, + "qparams": { + "group_size": { + "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": 0.010129058733582497, + "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": 0.05381399765610695, + "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": 0.05381399765610695, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.25812289118766785, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22554022073745728, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21199607849121094, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18569651246070862, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11891868710517883, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10532049089670181, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14275644719600677, + "qparams": { + "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": 0.13145099580287933, + "qparams": { + "group_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": 0.12366810441017151, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09916314482688904, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09314604103565216, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07298161834478378, + "qparams": { + "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": 0.06327079236507416, + "qparams": { + "group_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": 0.05755087360739708, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05613936856389046, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03669676557183266, + "qparams": { + "group_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": 0.030432214960455894, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029920488595962524, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026574725285172462, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025630196556448936, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019747933372855186, + "qparams": { + "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": 0.02019270323216915, + "qparams": { + "group_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": 0.01793496124446392, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014305224642157555, + "qparams": { + "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": 0.03669676557183266, + "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": 0.03669676557183266, + "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": 0.1245843917131424, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11665260046720505, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11362786591053009, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10294230282306671, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05823281779885292, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05516969412565231, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06613060086965561, + "qparams": { + "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": 0.06092027947306633, + "qparams": { + "group_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": 0.0590788833796978, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.052355118095874786, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04984664171934128, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03362191841006279, + "qparams": { + "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": 0.029166456311941147, + "qparams": { + "group_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": 0.027945026755332947, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02765311300754547, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01684313639998436, + "qparams": { + "group_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": 0.014432096853852272, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014313248917460442, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013241316191852093, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01305784098803997, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008831907995045185, + "qparams": { + "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": 0.008874508552253246, + "qparams": { + "group_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": 0.008435068652033806, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005931640975177288, + "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": 0.052355118095874786, + "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": 0.052355118095874786, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1047024130821228, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09804024547338486, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09515413641929626, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0861523374915123, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.048950277268886566, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.046155866235494614, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05596860125660896, + "qparams": { + "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": 0.05174915865063667, + "qparams": { + "group_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": 0.04967067390680313, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04397224262356758, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04182273522019386, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.028421781957149506, + "qparams": { + "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": 0.024728836491703987, + "qparams": { + "group_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": 0.023464908823370934, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02315647341310978, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014212450012564659, + "qparams": { + "group_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": 0.012064266949892044, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011935305781662464, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01105408649891615, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010861311107873917, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007424641400575638, + "qparams": { + "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": 0.007394286338239908, + "qparams": { + "group_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": 0.007022975478321314, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004821982700377703, + "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": 0.05174915865063667, + "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": 0.05174915865063667, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.24482642114162445, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22976826131343842, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2248135507106781, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.2040613740682602, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11505477875471115, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10949547588825226, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12898698449134827, + "qparams": { + "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": 0.11881615221500397, + "qparams": { + "group_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": 0.11667591333389282, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10351972281932831, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0983290821313858, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06564072519540787, + "qparams": { + "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": 0.056769706308841705, + "qparams": { + "group_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": 0.0550200492143631, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.054604314267635345, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03273392096161842, + "qparams": { + "group_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": 0.02801297977566719, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027877338230609894, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025631634518504143, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025365538895130157, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016866890713572502, + "qparams": { + "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": 0.016517912968993187, + "qparams": { + "group_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": 0.016206610947847366, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010315459221601486, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054604314267635345, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.054604314267635345, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.22716698050498962, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19882555305957794, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1896362006664276, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15797430276870728, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10585196316242218, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0950184315443039, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12264914810657501, + "qparams": { + "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": 0.11218367516994476, + "qparams": { + "group_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": 0.1092434898018837, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0857740044593811, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07551734149456024, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06329656392335892, + "qparams": { + "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": 0.054399363696575165, + "qparams": { + "group_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": 0.05157171189785004, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05088777840137482, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03200998157262802, + "qparams": { + "group_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": 0.027699917554855347, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027513636276125908, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023876547813415527, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02342456392943859, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017915841192007065, + "qparams": { + "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": 0.0185695867985487, + "qparams": { + "group_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": 0.01706988923251629, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014039356261491776, + "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": 0.054399363696575165, + "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": 0.054399363696575165, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.18610337376594543, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1749633252620697, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17149263620376587, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1559852957725525, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08777108788490295, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08375386148691177, + "qparams": { + "group_size": { + "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": 0.09760969877243042, + "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": 0.09017280489206314, + "qparams": { + "group_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": 0.08899983763694763, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07914624363183975, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07526344060897827, + "qparams": { + "group_size": { + "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": 0.04977108910679817, + "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": 0.04331713169813156, + "qparams": { + "group_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": 0.042226675897836685, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04196038097143173, + "qparams": { + "group_size": { + "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": 0.024960339069366455, + "qparams": { + "group_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": 0.022023184224963188, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021949781104922295, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02030099742114544, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020140113309025764, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013341221958398819, + "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": 0.013736364431679249, + "qparams": { + "group_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": 0.012993598356842995, + "qparams": { + "group_size": { + "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": 0.009662414900958538, + "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": 0.04977108910679817, + "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": 0.04977108910679817, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.23858889937400818, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22433382272720337, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21996094286441803, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1999732106924057, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11223126947879791, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1071384996175766, + "qparams": { + "group_size": { + "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": 0.12459976971149445, + "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": 0.1151878759264946, + "qparams": { + "group_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": 0.11377640068531036, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10121379792690277, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0961129441857338, + "qparams": { + "group_size": { + "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": 0.0633716955780983, + "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": 0.05506478250026703, + "qparams": { + "group_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": 0.05371712148189545, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05340283364057541, + "qparams": { + "group_size": { + "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": 0.03172137588262558, + "qparams": { + "group_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": 0.02741788886487484, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027329890057444572, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02514323592185974, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024940432980656624, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016594193875789642, + "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": 0.01620355248451233, + "qparams": { + "group_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": 0.016150031238794327, + "qparams": { + "group_size": { + "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": 0.010302101261913776, + "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": 0.05371712148189545, + "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": 0.05371712148189545, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.25660356879234314, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22323386371135712, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20882165431976318, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18282949924468994, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11805450171232224, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10379429161548615, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14292746782302856, + "qparams": { + "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": 0.13160502910614014, + "qparams": { + "group_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": 0.12293535470962524, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09801627695560455, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09223046153783798, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07306475192308426, + "qparams": { + "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": 0.06326965987682343, + "qparams": { + "group_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": 0.05710584670305252, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05556521937251091, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03667064756155014, + "qparams": { + "group_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": 0.03012881800532341, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029566748067736626, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0262162983417511, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025185763835906982, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019706429913640022, + "qparams": { + "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": 0.020004257559776306, + "qparams": { + "group_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": 0.017751235514879227, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014030076563358307, + "qparams": { + "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": 0.03667064756155014, + "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": 0.03667064756155014, + "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": 0.11974316835403442, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11220652610063553, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1092623621225357, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09896641224622726, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.055965546518564224, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05298115313053131, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06355712562799454, + "qparams": { + "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": 0.05858816206455231, + "qparams": { + "group_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": 0.0567389652132988, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05030430853366852, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04789016395807266, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.032281771302223206, + "qparams": { + "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": 0.02800123207271099, + "qparams": { + "group_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": 0.0268031544983387, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.026513323187828064, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01614466868340969, + "qparams": { + "group_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": 0.013766053132712841, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013652858324348927, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.012614747509360313, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012439131736755371, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008428066968917847, + "qparams": { + "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": 0.008367926813662052, + "qparams": { + "group_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": 0.008027960546314716, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0054575358517467976, + "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": 0.05030430853366852, + "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": 0.05030430853366852, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10088218003511429, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09443405270576477, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09165769070386887, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08300009369850159, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.047109946608543396, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.044445209205150604, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05396438017487526, + "qparams": { + "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": 0.04984349384903908, + "qparams": { + "group_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": 0.04783164709806442, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.042365703731775284, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.040327172726392746, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02738593891263008, + "qparams": { + "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": 0.023825448006391525, + "qparams": { + "group_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": 0.022587573155760765, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.022283708676695824, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013693587854504585, + "qparams": { + "group_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": 0.011617937125265598, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01148951891809702, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010649939998984337, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010461139492690563, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007153074722737074, + "qparams": { + "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": 0.00712814973667264, + "qparams": { + "group_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": 0.006759923417121172, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0046503134071826935, + "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": 0.05396438017487526, + "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": 0.05396438017487526, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.23754894733428955, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22295908629894257, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2179608941078186, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19770626723766327, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11136309057474136, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10589370131492615, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1250784695148468, + "qparams": { + "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": 0.1152767539024353, + "qparams": { + "group_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": 0.11292257159948349, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1002265065908432, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09523595124483109, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06357395648956299, + "qparams": { + "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": 0.05504821985960007, + "qparams": { + "group_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": 0.05324292927980423, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0528254434466362, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03172184154391289, + "qparams": { + "group_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": 0.027126159518957138, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026971688494086266, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0248190276324749, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024545593187212944, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016385817900300026, + "qparams": { + "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": 0.016015777364373207, + "qparams": { + "group_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": 0.015731625258922577, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.009988311678171158, + "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": 0.05324292927980423, + "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": 0.05324292927980423, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.23098459839820862, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19751963019371033, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.18590101599693298, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1571427285671234, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10699304193258286, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09408961981534958, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12650831043720245, + "qparams": { + "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": 0.11567706614732742, + "qparams": { + "group_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": 0.1113395020365715, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08535896986722946, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07728289812803268, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06525006145238876, + "qparams": { + "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": 0.056209009140729904, + "qparams": { + "group_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": 0.05228744447231293, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.051344387233257294, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.033096395432949066, + "qparams": { + "group_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": 0.028376247733831406, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02810034155845642, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02432546205818653, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023723356425762177, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01880616508424282, + "qparams": { + "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": 0.019499968737363815, + "qparams": { + "group_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": 0.017655232921242714, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014962534420192242, + "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": 0.05228744447231293, + "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": 0.05228744447231293, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.18768401443958282, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17646104097366333, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17292198538780212, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15733295679092407, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08842188119888306, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0844227597117424, + "qparams": { + "group_size": { + "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": 0.09829694032669067, + "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": 0.09084248542785645, + "qparams": { + "group_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": 0.08965487033128738, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07976675778627396, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07585449516773224, + "qparams": { + "group_size": { + "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": 0.05014796927571297, + "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": 0.043582092970609665, + "qparams": { + "group_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": 0.04249606281518936, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04223986342549324, + "qparams": { + "group_size": { + "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": 0.025132587179541588, + "qparams": { + "group_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": 0.02209222875535488, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022017499431967735, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02035793848335743, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02019456773996353, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013397708535194397, + "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": 0.013670407235622406, + "qparams": { + "group_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": 0.0130468113347888, + "qparams": { + "group_size": { + "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": 0.009493163786828518, + "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": 0.05014796927571297, + "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": 0.05014796927571297, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.23594027757644653, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22193795442581177, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21761305630207062, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19796578586101532, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11090567708015442, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10590306669473648, + "qparams": { + "group_size": { + "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": 0.12315940856933594, + "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": 0.11381588131189346, + "qparams": { + "group_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": 0.11239467561244965, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10008232295513153, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09521079808473587, + "qparams": { + "group_size": { + "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": 0.06260016560554504, + "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": 0.05441632866859436, + "qparams": { + "group_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": 0.0531010776758194, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.052784208208322525, + "qparams": { + "group_size": { + "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": 0.031312864273786545, + "qparams": { + "group_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": 0.027154773473739624, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02706173248589039, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024922911077737808, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024723883718252182, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016306398436427116, + "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": 0.016121579334139824, + "qparams": { + "group_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": 0.015865610912442207, + "qparams": { + "group_size": { + "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": 0.01034936960786581, + "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": 0.05441632866859436, + "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": 0.05441632866859436, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.25466400384902954, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22062449157238007, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20571495592594147, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17984797060489655, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11707169562578201, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10254448652267456, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14338721334934235, + "qparams": { + "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": 0.13126330077648163, + "qparams": { + "group_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": 0.122194804251194, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0970403254032135, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0914347842335701, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07344075292348862, + "qparams": { + "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": 0.06339483708143234, + "qparams": { + "group_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": 0.056946516036987305, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05533263832330704, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.036966852843761444, + "qparams": { + "group_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": 0.030572231858968735, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029987724497914314, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026762064546346664, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025703787803649902, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.020166601985692978, + "qparams": { + "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": 0.020963842049241066, + "qparams": { + "group_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": 0.01809416338801384, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015398278832435608, + "qparams": { + "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": 0.036966852843761444, + "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": 0.036966852843761444, + "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": 0.12323003262281418, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11541715264320374, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11240791529417038, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1018994078040123, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05757400766015053, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05456015095114708, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.0653509795665741, + "qparams": { + "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": 0.060225870460271835, + "qparams": { + "group_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": 0.058404870331287384, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.051820430904626846, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04933144524693489, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03324265778064728, + "qparams": { + "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": 0.028809929266572, + "qparams": { + "group_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": 0.027595937252044678, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02730892226099968, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.016637662425637245, + "qparams": { + "group_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": 0.014208076521754265, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01409322489053011, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013038471341133118, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01285715401172638, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.00872820895165205, + "qparams": { + "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": 0.008680197410285473, + "qparams": { + "group_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": 0.008335541002452374, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005720635876059532, + "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": 0.051820430904626846, + "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": 0.051820430904626846, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10251093655824661, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09600619971752167, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09308840334415436, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08435745537281036, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0478711761534214, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04511217027902603, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05497784540057182, + "qparams": { + "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": 0.050787679851055145, + "qparams": { + "group_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": 0.048553939908742905, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04304318502545357, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04099308326840401, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.027919085696339607, + "qparams": { + "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": 0.024272294715046883, + "qparams": { + "group_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": 0.022954588755965233, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02263728156685829, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013976361602544785, + "qparams": { + "group_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": 0.011829997412860394, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01169236283749342, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010851984843611717, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010652566328644753, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.00733196921646595, + "qparams": { + "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": 0.007302162237465382, + "qparams": { + "group_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": 0.00691575650125742, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004801610950380564, + "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": 0.050787679851055145, + "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": 0.050787679851055145, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2481706738471985, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23300684988498688, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2280217409133911, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20709332823753357, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11656612157821655, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11093559861183167, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13047707080841064, + "qparams": { + "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": 0.12026292085647583, + "qparams": { + "group_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": 0.11818817257881165, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10488445311784744, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09965384006500244, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06627453863620758, + "qparams": { + "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": 0.05743426829576492, + "qparams": { + "group_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": 0.05570748448371887, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0552927628159523, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03309565782546997, + "qparams": { + "group_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": 0.028312284499406815, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028185417875647545, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025898098945617676, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025638075545430183, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01704574190080166, + "qparams": { + "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": 0.016612466424703598, + "qparams": { + "group_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": 0.016400286927819252, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0102682551369071, + "qparams": { + "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": 0.03309565782546997, + "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": 0.03309565782546997, + "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": 0.2059377133846283, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1741868555545807, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1641428917646408, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13508771359920502, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09491831809282303, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08329226076602936, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1106928288936615, + "qparams": { + "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": 0.10170801728963852, + "qparams": { + "group_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": 0.09867100417613983, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07401631772518158, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06519503891468048, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05713238939642906, + "qparams": { + "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": 0.04977092146873474, + "qparams": { + "group_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": 0.046690043061971664, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04593968018889427, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02907789684832096, + "qparams": { + "group_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": 0.025900118052959442, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025722263380885124, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02217128314077854, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02170473523437977, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016760550439357758, + "qparams": { + "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": 0.018392860889434814, + "qparams": { + "group_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": 0.015847206115722656, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014756363816559315, + "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": 0.04977092146873474, + "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": 0.04977092146873474, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.17947813868522644, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.16881133615970612, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.16553764045238495, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15070176124572754, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08455751091241837, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08075286448001862, + "qparams": { + "group_size": { + "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": 0.0938139483332634, + "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": 0.08682437986135483, + "qparams": { + "group_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": 0.0856974869966507, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07633363455533981, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07258257269859314, + "qparams": { + "group_size": { + "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": 0.04787905886769295, + "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": 0.04162030667066574, + "qparams": { + "group_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": 0.04059319198131561, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04034968838095665, + "qparams": { + "group_size": { + "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": 0.02393762581050396, + "qparams": { + "group_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": 0.021026184782385826, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020956015214323997, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0193736981600523, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019221343100070953, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012647309340536594, + "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": 0.01290916372090578, + "qparams": { + "group_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": 0.012315385974943638, + "qparams": { + "group_size": { + "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": 0.008846567012369633, + "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": 0.04787905886769295, + "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": 0.04787905886769295, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.22857244312763214, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.215154230594635, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2109716385602951, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19211067259311676, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10745642334222794, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10268812626600266, + "qparams": { + "group_size": { + "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": 0.11912760883569717, + "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": 0.11026980727910995, + "qparams": { + "group_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": 0.10893380641937256, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09705017507076263, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09224261343479156, + "qparams": { + "group_size": { + "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": 0.06058460474014282, + "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": 0.05266205966472626, + "qparams": { + "group_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": 0.05139957368373871, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05109362304210663, + "qparams": { + "group_size": { + "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": 0.030238989740610123, + "qparams": { + "group_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": 0.026171905919909477, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026087122038006783, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02401062101125717, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02382074110209942, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015651976689696312, + "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": 0.015368727967143059, + "qparams": { + "group_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": 0.015229551121592522, + "qparams": { + "group_size": { + "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": 0.009637171402573586, + "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": 0.05266205966472626, + "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": 0.05266205966472626, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.24542368948459625, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21158316731452942, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19671794772148132, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17240944504737854, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11242632567882538, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09786234050989151, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1380675882101059, + "qparams": { + "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": 0.12646138668060303, + "qparams": { + "group_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": 0.1174054741859436, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09283912181854248, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08783715218305588, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07071446627378464, + "qparams": { + "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": 0.0609353706240654, + "qparams": { + "group_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": 0.05455928295850754, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05296160653233528, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.0356895886361599, + "qparams": { + "group_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": 0.02909149043262005, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028501814231276512, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02532612904906273, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024276062846183777, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019618259742856026, + "qparams": { + "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": 0.01975914277136326, + "qparams": { + "group_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": 0.017607949674129486, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014245321974158287, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05455928295850754, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05455928295850754, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.12696292996406555, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11930805444717407, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1163850948214531, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10578782856464386, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.059460368007421494, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.056499578058719635, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.0669645369052887, + "qparams": { + "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": 0.06191486492753029, + "qparams": { + "group_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": 0.06031458079814911, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.053624477237463, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05099857971072197, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03399279713630676, + "qparams": { + "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": 0.029570462182164192, + "qparams": { + "group_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": 0.028459027409553528, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.028187589719891548, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01699742116034031, + "qparams": { + "group_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": 0.014545867219567299, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014445154927670956, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013351162895560265, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013180617243051529, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008842285722494125, + "qparams": { + "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": 0.008711690083146095, + "qparams": { + "group_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": 0.008468423038721085, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005552799440920353, + "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": 0.053624477237463, + "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": 0.053624477237463, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.10982583463191986, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10312791168689728, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10046863555908203, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09133461862802505, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0514155775308609, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.048771340399980545, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05833518132567406, + "qparams": { + "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": 0.05381203442811966, + "qparams": { + "group_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": 0.05214870721101761, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04638329893350601, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04421081393957138, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02961021661758423, + "qparams": { + "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": 0.025696605443954468, + "qparams": { + "group_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": 0.02461959421634674, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.024353913962841034, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014791857451200485, + "qparams": { + "group_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": 0.012587379664182663, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012482995167374611, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011553642340004444, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011392607353627682, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007677543442696333, + "qparams": { + "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": 0.007563456427305937, + "qparams": { + "group_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": 0.007316931150853634, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004816466011106968, + "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": 0.05381203442811966, + "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": 0.05381203442811966, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.24592293798923492, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23146526515483856, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22670023143291473, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.206340029835701, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11561065912246704, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11031409353017807, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12901251018047333, + "qparams": { + "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": 0.11896985024213791, + "qparams": { + "group_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": 0.11716578155755997, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1043185368180275, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09921962022781372, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06549452245235443, + "qparams": { + "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": 0.05680166557431221, + "qparams": { + "group_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": 0.055237166583538055, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.054858800023794174, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.032690200954675674, + "qparams": { + "group_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": 0.02806740440428257, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027944037690758705, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02572183683514595, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025480443611741066, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016845857724547386, + "qparams": { + "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": 0.01641746051609516, + "qparams": { + "group_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": 0.01626776158809662, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010122577659785748, + "qparams": { + "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": 0.032690200954675674, + "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": 0.032690200954675674, + "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": 0.2133505642414093, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.17294125258922577, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15876136720180511, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.12445921450853348, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09720859676599503, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0818527564406395, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11673885583877563, + "qparams": { + "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": 0.10708265751600266, + "qparams": { + "group_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": 0.10268072783946991, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0699356198310852, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06373675167560577, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06006627157330513, + "qparams": { + "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": 0.0518244244158268, + "qparams": { + "group_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": 0.047318778932094574, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04621502757072449, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03023160994052887, + "qparams": { + "group_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": 0.025449199602007866, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0251711867749691, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02035978063941002, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019575105980038643, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016657767817378044, + "qparams": { + "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": 0.017376858741044998, + "qparams": { + "group_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": 0.015272795222699642, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.012936239130795002, + "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": 0.0518244244158268, + "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": 0.0518244244158268, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.18863710761070251, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17763324081897736, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17418259382247925, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1586817502975464, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08887583017349243, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08493968844413757, + "qparams": { + "group_size": { + "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": 0.09869535267353058, + "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": 0.09122920036315918, + "qparams": { + "group_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": 0.09008591622114182, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0803178921341896, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07642015069723129, + "qparams": { + "group_size": { + "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": 0.050318360328674316, + "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": 0.043717917054891586, + "qparams": { + "group_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": 0.04266095906496048, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04240795597434044, + "qparams": { + "group_size": { + "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": 0.025161199271678925, + "qparams": { + "group_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": 0.02208888530731201, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022016864269971848, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02036431059241295, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020202336832880974, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013298936188220978, + "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": 0.013536317273974419, + "qparams": { + "group_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": 0.012948565185070038, + "qparams": { + "group_size": { + "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": 0.009253064170479774, + "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": 0.050318360328674316, + "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": 0.050318360328674316, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.235043004155159, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2214175909757614, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21726809442043304, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1979786902666092, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11051853001117706, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10565662384033203, + "qparams": { + "group_size": { + "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": 0.12286070734262466, + "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": 0.11333835124969482, + "qparams": { + "group_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": 0.11195341497659683, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09990767389535904, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09519817680120468, + "qparams": { + "group_size": { + "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": 0.06245657429099083, + "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": 0.05415594205260277, + "qparams": { + "group_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": 0.05286875739693642, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05256590619683266, + "qparams": { + "group_size": { + "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": 0.031252145767211914, + "qparams": { + "group_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": 0.026995878666639328, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026910560205578804, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024807406589388847, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024614639580249786, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.0163801908493042, + "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": 0.015962963923811913, + "qparams": { + "group_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": 0.015955600887537003, + "qparams": { + "group_size": { + "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": 0.01017359271645546, + "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": 0.05415594205260277, + "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": 0.05415594205260277, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.24747593700885773, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21341276168823242, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1976483166217804, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1734105348587036, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.1133929118514061, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09831378608942032, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14061026275157928, + "qparams": { + "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": 0.1289234161376953, + "qparams": { + "group_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": 0.11863832175731659, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09393993765115738, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08892501890659332, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07186517119407654, + "qparams": { + "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": 0.062133509665727615, + "qparams": { + "group_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": 0.055027857422828674, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.053247060626745224, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03616069257259369, + "qparams": { + "group_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": 0.02930871769785881, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028630705550312996, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025562142953276634, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024378124624490738, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01960202492773533, + "qparams": { + "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": 0.01996956393122673, + "qparams": { + "group_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": 0.017346743494272232, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014258009381592274, + "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": 0.053247060626745224, + "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": 0.053247060626745224, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.12435349076986313, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11667585372924805, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11370652168989182, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10326826572418213, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.058222927153110504, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05519571155309677, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06570006906986237, + "qparams": { + "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": 0.06074734777212143, + "qparams": { + "group_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": 0.059029366821050644, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05243782699108124, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04988223314285278, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03337618708610535, + "qparams": { + "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": 0.029045408591628075, + "qparams": { + "group_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": 0.02785632759332657, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.027572978287935257, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01669137552380562, + "qparams": { + "group_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": 0.014274592511355877, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014167679473757744, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01310008205473423, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012920191511511803, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008699393831193447, + "qparams": { + "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": 0.00861845351755619, + "qparams": { + "group_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": 0.008313899859786034, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005558414850383997, + "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": 0.05243782699108124, + "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": 0.05243782699108124, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.10456854104995728, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.0980893149971962, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09537395089864731, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08663544058799744, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.048934206366539, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04628770053386688, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05591043084859848, + "qparams": { + "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": 0.0515601746737957, + "qparams": { + "group_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": 0.049643322825431824, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04410761967301369, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0420520044863224, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02837914600968361, + "qparams": { + "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": 0.02462868206202984, + "qparams": { + "group_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": 0.02344975620508194, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.023162273690104485, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014197693206369877, + "qparams": { + "group_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": 0.012031116522848606, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01191095169633627, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01104162447154522, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010862603783607483, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.00740100396797061, + "qparams": { + "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": 0.007312243338674307, + "qparams": { + "group_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": 0.007017349824309349, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004713288974016905, + "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": 0.0515601746737957, + "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": 0.0515601746737957, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2449677586555481, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23026418685913086, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2255387306213379, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20497193932533264, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11505324393510818, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10971969366073608, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12850530445575714, + "qparams": { + "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": 0.11855675280094147, + "qparams": { + "group_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": 0.11661285907030106, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10374706983566284, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09865323454141617, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06529373675584793, + "qparams": { + "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": 0.05660274997353554, + "qparams": { + "group_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": 0.05499975383281708, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05462128669023514, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.032557420432567596, + "qparams": { + "group_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": 0.027944859117269516, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027822865173220634, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025589952245354652, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025347838178277016, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01674024388194084, + "qparams": { + "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": 0.016362158581614494, + "qparams": { + "group_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": 0.016149131581187248, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010102387517690659, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05462128669023514, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05462128669023514, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.2082900106906891, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16679570078849792, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15069295465946198, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1213822290301323, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09614074230194092, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07900063693523407, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11622384935617447, + "qparams": { + "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": 0.10671786218881607, + "qparams": { + "group_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": 0.09982267022132874, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06741705536842346, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0632680281996727, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05984595790505409, + "qparams": { + "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": 0.05167461186647415, + "qparams": { + "group_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": 0.046889130026102066, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04569099843502045, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.030175060033798218, + "qparams": { + "group_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": 0.02533883973956108, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024905364960432053, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.020056476816534996, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019236531108617783, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016700174659490585, + "qparams": { + "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": 0.017505057156085968, + "qparams": { + "group_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": 0.015233516693115234, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013167821802198887, + "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": 0.05167461186647415, + "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": 0.05167461186647415, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.1933848261833191, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18217679858207703, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1787167340517044, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16278283298015594, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0911293625831604, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08712182939052582, + "qparams": { + "group_size": { + "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": 0.10108339786529541, + "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": 0.09347648918628693, + "qparams": { + "group_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": 0.0923340693116188, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0823562815785408, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07837597280740738, + "qparams": { + "group_size": { + "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": 0.05151912569999695, + "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": 0.044787753373384476, + "qparams": { + "group_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": 0.04372700676321983, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04346916079521179, + "qparams": { + "group_size": { + "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": 0.025767076760530472, + "qparams": { + "group_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": 0.022603098303079605, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022528868168592453, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.020829487591981888, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02067294530570507, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013602446764707565, + "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": 0.013796520419418812, + "qparams": { + "group_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": 0.013253397308290005, + "qparams": { + "group_size": { + "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": 0.009370341897010803, + "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": 0.05151912569999695, + "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": 0.05151912569999695, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2373647540807724, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22364673018455505, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21943795680999756, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19997882843017578, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11162618547677994, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10674288123846054, + "qparams": { + "group_size": { + "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": 0.12367626279592514, + "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": 0.11445643752813339, + "qparams": { + "group_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": 0.11306794732809067, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10090696811676025, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09599818289279938, + "qparams": { + "group_size": { + "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": 0.06292277574539185, + "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": 0.05465381592512131, + "qparams": { + "group_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": 0.05336901172995567, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05306401476264, + "qparams": { + "group_size": { + "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": 0.03140103071928024, + "qparams": { + "group_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": 0.02717055380344391, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027081195265054703, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02496138960123062, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02476654201745987, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016280874609947205, + "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": 0.01595207303762436, + "qparams": { + "group_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": 0.015852229669690132, + "qparams": { + "group_size": { + "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": 0.010012071579694748, + "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": 0.05465381592512131, + "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": 0.05465381592512131, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.24983832240104675, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21607911586761475, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2001928687095642, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17602995038032532, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11449858546257019, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09958389401435852, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14248812198638916, + "qparams": { + "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": 0.13035015761852264, + "qparams": { + "group_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": 0.11976308375597, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09529438614845276, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09027355909347534, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07276619225740433, + "qparams": { + "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": 0.06282643228769302, + "qparams": { + "group_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": 0.05565118044614792, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.053834136575460434, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03702933341264725, + "qparams": { + "group_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": 0.029744209721684456, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02905300259590149, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02605910412967205, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024868307635188103, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.020809078589081764, + "qparams": { + "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": 0.020391689613461494, + "qparams": { + "group_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": 0.01866987720131874, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0146986935287714, + "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": 0.053834136575460434, + "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": 0.053834136575460434, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.12548525631427765, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11786383390426636, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11488029360771179, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1044115200638771, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05878397822380066, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.055765777826309204, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.0667145699262619, + "qparams": { + "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": 0.06144168972969055, + "qparams": { + "group_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": 0.059581901878118515, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05300486087799072, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.050537366420030594, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.033866412937641144, + "qparams": { + "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": 0.02938505820930004, + "qparams": { + "group_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": 0.028173571452498436, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02788485214114189, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.016972888261079788, + "qparams": { + "group_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": 0.014454228803515434, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014334645122289658, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013273055665194988, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013086169958114624, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008879927918314934, + "qparams": { + "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": 0.008745312690734863, + "qparams": { + "group_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": 0.008476163260638714, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005663158372044563, + "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": 0.05300486087799072, + "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": 0.05300486087799072, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.10733355581760406, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10077439993619919, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09812083095312119, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0891391932964325, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05029978230595589, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.047662168741226196, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.057169437408447266, + "qparams": { + "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": 0.05278217792510986, + "qparams": { + "group_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": 0.05102330073714256, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0453471839427948, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.043248068541288376, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.029038704931735992, + "qparams": { + "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": 0.02521168626844883, + "qparams": { + "group_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": 0.024096176028251648, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.023821551352739334, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014514639042317867, + "qparams": { + "group_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": 0.012323293834924698, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012214499525725842, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011304903775453568, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011136164888739586, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0075355954468250275, + "qparams": { + "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": 0.007420965936034918, + "qparams": { + "group_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": 0.007167706731706858, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004727623425424099, + "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": 0.05278217792510986, + "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": 0.05278217792510986, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "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": 0.2404746562242508, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2260812520980835, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2215137928724289, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20149606466293335, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11297371238470078, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10771667957305908, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12615947425365448, + "qparams": { + "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": 0.11643011122941971, + "qparams": { + "group_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": 0.11451262980699539, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10190524905920029, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09696260839700699, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06404666602611542, + "qparams": { + "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": 0.05555453151464462, + "qparams": { + "group_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": 0.05398477241396904, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05360865965485573, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03196835517883301, + "qparams": { + "group_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": 0.027415160089731216, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02729923091828823, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025112194940447807, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024881400167942047, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016451900824904442, + "qparams": { + "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": 0.01603705808520317, + "qparams": { + "group_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": 0.0158732570707798, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.00988390576094389, + "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": 0.05398477241396904, + "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": 0.05398477241396904, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.2104046493768692, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1704159677028656, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15728704631328583, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1350259631872177, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09444073587656021, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07884619385004044, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11367577314376831, + "qparams": { + "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": 0.10315004736185074, + "qparams": { + "group_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": 0.0994453951716423, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07200773060321808, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06507574766874313, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05817661061882973, + "qparams": { + "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": 0.05036836862564087, + "qparams": { + "group_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": 0.04650857672095299, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04555564001202583, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.030027588829398155, + "qparams": { + "group_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": 0.025793531909585, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02556898258626461, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021788576617836952, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02116904780268669, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017678458243608475, + "qparams": { + "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": 0.018390530720353127, + "qparams": { + "group_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": 0.016614513471722603, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014618166722357273, + "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": 0.05036836862564087, + "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": 0.05036836862564087, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.19986604154109955, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18829578161239624, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1847010850906372, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16832466423511505, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09425145387649536, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0900576263666153, + "qparams": { + "group_size": { + "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": 0.10449259728193283, + "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": 0.09670138359069824, + "qparams": { + "group_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": 0.09549997746944427, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08517605066299438, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08102382719516754, + "qparams": { + "group_size": { + "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": 0.05328821390867233, + "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": 0.04630879685282707, + "qparams": { + "group_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": 0.04519739747047424, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04492934048175812, + "qparams": { + "group_size": { + "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": 0.02664208598434925, + "qparams": { + "group_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": 0.02331054024398327, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0232376866042614, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021474022418260574, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021305467933416367, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01401466317474842, + "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": 0.01414608582854271, + "qparams": { + "group_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": 0.013653748668730259, + "qparams": { + "group_size": { + "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": 0.00950024463236332, + "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": 0.05328821390867233, + "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": 0.05328821390867233, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.24228057265281677, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22821059823036194, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22397376596927643, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20412924885749817, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11401575803756714, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10899525880813599, + "qparams": { + "group_size": { + "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": 0.12640655040740967, + "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": 0.11686796694993973, + "qparams": { + "group_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": 0.115499347448349, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10307169705629349, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0981488972902298, + "qparams": { + "group_size": { + "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": 0.06443152576684952, + "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": 0.055890053510665894, + "qparams": { + "group_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": 0.05457343906164169, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05426209047436714, + "qparams": { + "group_size": { + "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": 0.03220779821276665, + "qparams": { + "group_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": 0.027887752279639244, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02779729664325714, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025637127459049225, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02543214149773121, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016834517940878868, + "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": 0.01651681959629059, + "qparams": { + "group_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": 0.016402365639805794, + "qparams": { + "group_size": { + "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": 0.010576554574072361, + "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": 0.05457343906164169, + "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": 0.05426209047436714, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.2543475031852722, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2211930900812149, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2057252824306488, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18115270137786865, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11661098897457123, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10199172794818878, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1435667723417282, + "qparams": { + "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": 0.13215070962905884, + "qparams": { + "group_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": 0.12178990244865417, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09755339473485947, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09236323088407516, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07338355481624603, + "qparams": { + "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": 0.06344180554151535, + "qparams": { + "group_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": 0.05637365207076073, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05459444224834442, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03686334565281868, + "qparams": { + "group_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": 0.029568929225206375, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02889418788254261, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025799620896577835, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02461402863264084, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019785858690738678, + "qparams": { + "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": 0.01958794891834259, + "qparams": { + "group_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": 0.017544975504279137, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013353300280869007, + "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": 0.05459444224834442, + "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": 0.05459444224834442, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.12729960680007935, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11946793645620346, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11650270223617554, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10592285543680191, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05963604152202606, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05658005177974701, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06735263019800186, + "qparams": { + "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": 0.06222426891326904, + "qparams": { + "group_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": 0.06048073247075081, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05375909060239792, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05117260292172432, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.034208573400974274, + "qparams": { + "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": 0.02973310835659504, + "qparams": { + "group_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": 0.028556086122989655, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.028274960815906525, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.017107391729950905, + "qparams": { + "group_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": 0.014633839018642902, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014525721780955791, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013426857069134712, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013245481997728348, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008910594508051872, + "qparams": { + "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": 0.008819537237286568, + "qparams": { + "group_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": 0.008520355448126793, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005681207403540611, + "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": 0.05375909060239792, + "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": 0.05375909060239792, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.1078648567199707, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1012139767408371, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09851136803627014, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08957034349441528, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0505484864115715, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04787341505289078, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.057546377182006836, + "qparams": { + "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": 0.053065136075019836, + "qparams": { + "group_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": 0.0512797050178051, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04556316137313843, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.043465062975883484, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.029237613081932068, + "qparams": { + "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": 0.02536391280591488, + "qparams": { + "group_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": 0.02420787699520588, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.023937104269862175, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014614280313253403, + "qparams": { + "group_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": 0.012386806309223175, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012274215929210186, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011365671642124653, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011188345029950142, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007596504874527454, + "qparams": { + "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": 0.007470393553376198, + "qparams": { + "group_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": 0.007217117585241795, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004757164046168327, + "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": 0.053065136075019836, + "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": 0.053065136075019836, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2460312396287918, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23128463327884674, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2265210896730423, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20611454546451569, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11555349826812744, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11018174141645432, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1288907527923584, + "qparams": { + "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": 0.11896085739135742, + "qparams": { + "group_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": 0.11714737117290497, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10422573238611221, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0991087555885315, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06544890999794006, + "qparams": { + "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": 0.0567697137594223, + "qparams": { + "group_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": 0.05521797761321068, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.054842088371515274, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03265504166483879, + "qparams": { + "group_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": 0.02803473174571991, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027914199978113174, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025674795731902122, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02543923445045948, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01676468551158905, + "qparams": { + "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": 0.016375264152884483, + "qparams": { + "group_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": 0.016190433874726295, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010068090632557869, + "qparams": { + "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": 0.03265504166483879, + "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": 0.03265504166483879, + "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": 0.21910691261291504, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1696912795305252, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1508432775735855, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.12576204538345337, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0975479707121849, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07528048753738403, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12193092703819275, + "qparams": { + "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": 0.11197485774755478, + "qparams": { + "group_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": 0.1052502766251564, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07236320525407791, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06546422839164734, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06269488483667374, + "qparams": { + "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": 0.05400114879012108, + "qparams": { + "group_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": 0.047464821487665176, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.045810528099536896, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.031564440578222275, + "qparams": { + "group_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": 0.025326760485768318, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024889858439564705, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02056865580379963, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019418958574533463, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017261218279600143, + "qparams": { + "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": 0.017440633848309517, + "qparams": { + "group_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": 0.015276137739419937, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.012535701505839825, + "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": 0.05400114879012108, + "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": 0.05400114879012108, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.20076939463615417, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1890779435634613, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18548989295959473, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1690904200077057, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09470213204622269, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09051431715488434, + "qparams": { + "group_size": { + "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": 0.10525842010974884, + "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": 0.0971723347902298, + "qparams": { + "group_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": 0.0959797352552414, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0855754166841507, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08155542612075806, + "qparams": { + "group_size": { + "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": 0.053688034415245056, + "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": 0.046556174755096436, + "qparams": { + "group_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": 0.045436371117830276, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04517263174057007, + "qparams": { + "group_size": { + "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": 0.026883909478783607, + "qparams": { + "group_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": 0.02347414568066597, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023399269208312035, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021625084802508354, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021458355709910393, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.014259415678679943, + "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": 0.014296111650764942, + "qparams": { + "group_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": 0.013900712132453918, + "qparams": { + "group_size": { + "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": 0.009667742066085339, + "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": 0.053688034415245056, + "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": 0.053688034415245056, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.2444920688867569, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.23029889166355133, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22596843540668488, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20600688457489014, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11508745700120926, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11000483483076096, + "qparams": { + "group_size": { + "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": 0.12768223881721497, + "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": 0.11798335611820221, + "qparams": { + "group_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": 0.1165943294763565, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1040077656507492, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09910093992948532, + "qparams": { + "group_size": { + "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": 0.06497766822576523, + "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": 0.05639983341097832, + "qparams": { + "group_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": 0.055058255791664124, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.054740142077207565, + "qparams": { + "group_size": { + "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": 0.03245048597455025, + "qparams": { + "group_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": 0.02808697521686554, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02799893170595169, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02580134943127632, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025603745132684708, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016889411956071854, + "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": 0.01656452566385269, + "qparams": { + "group_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": 0.01644834876060486, + "qparams": { + "group_size": { + "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": 0.010503489524126053, + "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": 0.054740142077207565, + "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": 0.054740142077207565, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.25217700004577637, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22011935710906982, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20408596098423004, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1801174134016037, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11558585613965988, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10093256086111069, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14386437833309174, + "qparams": { + "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": 0.1324736624956131, + "qparams": { + "group_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": 0.12069730460643768, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09735577553510666, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09240002185106277, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07338203489780426, + "qparams": { + "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": 0.06353878229856491, + "qparams": { + "group_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": 0.05591557174921036, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05399090796709061, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.036779701709747314, + "qparams": { + "group_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": 0.029410559684038162, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028630517423152924, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025819681584835052, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024535221979022026, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01959196850657463, + "qparams": { + "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": 0.01967618055641651, + "qparams": { + "group_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": 0.01712537184357643, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013450324535369873, + "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": 0.05399090796709061, + "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": 0.05399090796709061, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.12626676261425018, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11855881661176682, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11555858701467514, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10509706288576126, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05918693169951439, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.056120362132787704, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06712210178375244, + "qparams": { + "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": 0.061835505068302155, + "qparams": { + "group_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": 0.06004675477743149, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05336521565914154, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05093621090054512, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03410404548048973, + "qparams": { + "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": 0.02956315129995346, + "qparams": { + "group_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": 0.028349434956908226, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.028062738478183746, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.017056668177247047, + "qparams": { + "group_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": 0.014513418078422546, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014397692866623402, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013314586132764816, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013133067637681961, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008878685534000397, + "qparams": { + "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": 0.008732559159398079, + "qparams": { + "group_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": 0.008480108343064785, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005585538223385811, + "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": 0.05336521565914154, + "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": 0.05336521565914154, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.10839751362800598, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1018410474061966, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09914285689592361, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09015500545501709, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.050821732729673386, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.048132430762052536, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05808781459927559, + "qparams": { + "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": 0.05333369970321655, + "qparams": { + "group_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": 0.05153832212090492, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04581774026155472, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04383915662765503, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.029473017901182175, + "qparams": { + "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": 0.025490351021289825, + "qparams": { + "group_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": 0.024356659501791, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.024085495620965958, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014749194495379925, + "qparams": { + "group_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": 0.012472248636186123, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012357592582702637, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011449234560132027, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011273500509560108, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007691759150475264, + "qparams": { + "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": 0.007524420972913504, + "qparams": { + "group_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": 0.007323212455958128, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004808076191693544, + "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": 0.05333369970321655, + "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": 0.05333369970321655, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "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": 0.24646157026290894, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23179109394550323, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2271142303943634, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20668712258338928, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11594215780496597, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11053670942783356, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1293235421180725, + "qparams": { + "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": 0.11932267248630524, + "qparams": { + "group_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": 0.11754611134529114, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10454672574996948, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0995408147573471, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06577195972204208, + "qparams": { + "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": 0.05699129402637482, + "qparams": { + "group_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": 0.055431898683309555, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05506553128361702, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03284379094839096, + "qparams": { + "group_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": 0.028246305882930756, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028132367879152298, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025884175673127174, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025652805343270302, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01696164533495903, + "qparams": { + "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": 0.01665470376610756, + "qparams": { + "group_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": 0.016380663961172104, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01046463567763567, + "qparams": { + "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": 0.03284379094839096, + "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": 0.03284379094839096, + "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": 0.19690483808517456, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.16227057576179504, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1497134268283844, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11877986788749695, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08922430127859116, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07521678507328033, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1096009910106659, + "qparams": { + "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": 0.09970568120479584, + "qparams": { + "group_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": 0.09438610076904297, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06774352490901947, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05969906225800514, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0563533790409565, + "qparams": { + "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": 0.04861324653029442, + "qparams": { + "group_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": 0.04391861706972122, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04275503754615784, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02847742661833763, + "qparams": { + "group_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": 0.02423911914229393, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023911412805318832, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.020352985709905624, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01960977166891098, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016068924218416214, + "qparams": { + "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": 0.017331503331661224, + "qparams": { + "group_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": 0.014648471027612686, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013585129752755165, + "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": 0.04861324653029442, + "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": 0.04861324653029442, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.20102788507938385, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1893322765827179, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18570128083229065, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16929970681667328, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09481564909219742, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09060628712177277, + "qparams": { + "group_size": { + "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": 0.10527589172124863, + "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": 0.09729483723640442, + "qparams": { + "group_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": 0.09612005949020386, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0856887549161911, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08163155615329742, + "qparams": { + "group_size": { + "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": 0.05368034169077873, + "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": 0.046605441719293594, + "qparams": { + "group_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": 0.04547595605254173, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.045201376080513, + "qparams": { + "group_size": { + "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": 0.026828551664948463, + "qparams": { + "group_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": 0.023445621132850647, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023375146090984344, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.0215871911495924, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02142677642405033, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.014127840287983418, + "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": 0.014226099476218224, + "qparams": { + "group_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": 0.013760032132267952, + "qparams": { + "group_size": { + "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": 0.009540731087327003, + "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": 0.05368034169077873, + "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": 0.05368034169077873, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.24475936591625214, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2305239737033844, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22612780332565308, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.2061513513326645, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11515876650810242, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11007406562566757, + "qparams": { + "group_size": { + "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": 0.12772053480148315, + "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": 0.11809427291154861, + "qparams": { + "group_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": 0.11669167131185532, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10408613085746765, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09913763403892517, + "qparams": { + "group_size": { + "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": 0.0649360790848732, + "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": 0.05641993507742882, + "qparams": { + "group_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": 0.05508344992995262, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.054763760417699814, + "qparams": { + "group_size": { + "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": 0.03244480490684509, + "qparams": { + "group_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": 0.028058074414730072, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027967344969511032, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02576637640595436, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025562282651662827, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016867857426404953, + "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": 0.016493095085024834, + "qparams": { + "group_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": 0.01642373390495777, + "qparams": { + "group_size": { + "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": 0.010372033342719078, + "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": 0.054763760417699814, + "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": 0.054763760417699814, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.25043538212776184, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21909885108470917, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20346859097480774, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17993928492069244, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.1150846853852272, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10068383812904358, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14292289316654205, + "qparams": { + "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": 0.13151732087135315, + "qparams": { + "group_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": 0.12008211761713028, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09726861864328384, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09243831783533096, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07333024591207504, + "qparams": { + "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": 0.06341846287250519, + "qparams": { + "group_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": 0.055858414620161057, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.053955819457769394, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03709203377366066, + "qparams": { + "group_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": 0.029720421880483627, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028972841799259186, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026266805827617645, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025037303566932678, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.020199788734316826, + "qparams": { + "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": 0.02026238664984703, + "qparams": { + "group_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": 0.017794648185372353, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014396406710147858, + "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": 0.053955819457769394, + "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": 0.053955819457769394, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.13521310687065125, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1270046979188919, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12400881946086884, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11275891214609146, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0634727030992508, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06026412546634674, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07136358320713043, + "qparams": { + "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": 0.06603989750146866, + "qparams": { + "group_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": 0.06433337926864624, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05719839408993721, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05438673496246338, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03620029240846634, + "qparams": { + "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": 0.03153254836797714, + "qparams": { + "group_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": 0.030364539474248886, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.030083294957876205, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.018083330243825912, + "qparams": { + "group_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": 0.015489314682781696, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015383210964500904, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.014196526259183884, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014017963781952858, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.00935463048517704, + "qparams": { + "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": 0.009221379645168781, + "qparams": { + "group_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": 0.00896717794239521, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005800248123705387, + "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": 0.05438673496246338, + "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": 0.05438673496246338, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "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": 0.11428263783454895, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10735150426626205, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10466445982456207, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09522062540054321, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05367084592580795, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05090508610010147, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06077626347541809, + "qparams": { + "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": 0.056048858910799026, + "qparams": { + "group_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": 0.05444210395216942, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04840777441859245, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04617278277873993, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.030887272208929062, + "qparams": { + "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": 0.026783451437950134, + "qparams": { + "group_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": 0.02570226788520813, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02544279955327511, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.015437239781022072, + "qparams": { + "group_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": 0.013131813146173954, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013029532507061958, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.012045461684465408, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01187961921095848, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008008015342056751, + "qparams": { + "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": 0.007859896868467331, + "qparams": { + "group_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": 0.007648324128240347, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0049798376858234406, + "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": 0.05444210395216942, + "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": 0.05444210395216942, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "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": 0.2507553696632385, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23573185503482819, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23104271292686462, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.2103252410888672, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.1179569885134697, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11247393488883972, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13134661316871643, + "qparams": { + "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": 0.12128379940986633, + "qparams": { + "group_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": 0.11952206492424011, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10636529326438904, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10119882225990295, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06678129732608795, + "qparams": { + "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": 0.05789410322904587, + "qparams": { + "group_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": 0.05636371299624443, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05600126460194588, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.033308856189250946, + "qparams": { + "group_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": 0.028647838160395622, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028534911572933197, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02623859792947769, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02600787952542305, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01714988425374031, + "qparams": { + "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": 0.016757385805249214, + "qparams": { + "group_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": 0.016583232209086418, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010363989509642124, + "qparams": { + "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": 0.033308856189250946, + "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": 0.033308856189250946, + "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": 0.19211173057556152, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.14677993953227997, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12775124609470367, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10215387493371964, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08460283279418945, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06694833189249039, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.10880966484546661, + "qparams": { + "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": 0.10018674284219742, + "qparams": { + "group_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": 0.09208814799785614, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05970306321978569, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.056808263063430786, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05600612238049507, + "qparams": { + "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": 0.04818800836801529, + "qparams": { + "group_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": 0.04106695204973221, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.039229679852724075, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.028141804039478302, + "qparams": { + "group_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": 0.021789973601698875, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021240154281258583, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.017057053744792938, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01570683903992176, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01521947979927063, + "qparams": { + "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": 0.015068232081830502, + "qparams": { + "group_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": 0.012974104844033718, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010424361564218998, + "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": 0.04818800836801529, + "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": 0.04818800836801529, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.20068438351154327, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18899023532867432, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1853330433368683, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16880778968334198, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09471377730369568, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09047341346740723, + "qparams": { + "group_size": { + "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": 0.10520250350236893, + "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": 0.09724340587854385, + "qparams": { + "group_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": 0.09599483013153076, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08550328016281128, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08136726915836334, + "qparams": { + "group_size": { + "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": 0.05365070328116417, + "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": 0.0465831533074379, + "qparams": { + "group_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": 0.04544181376695633, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.045168496668338776, + "qparams": { + "group_size": { + "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": 0.026839153841137886, + "qparams": { + "group_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": 0.023429373279213905, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023356102406978607, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021561656147241592, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021391218528151512, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.014166674576699734, + "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": 0.01421182882040739, + "qparams": { + "group_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": 0.01379281934350729, + "qparams": { + "group_size": { + "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": 0.009530789218842983, + "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": 0.05365070328116417, + "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": 0.05365070328116417, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.24594438076019287, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.23160208761692047, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.227179154753685, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20704184472560883, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11579892784357071, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11064186692237854, + "qparams": { + "group_size": { + "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": 0.1284273862838745, + "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": 0.11877661943435669, + "qparams": { + "group_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": 0.11734438687562943, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10457607358694077, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09947250783443451, + "qparams": { + "group_size": { + "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": 0.06532412767410278, + "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": 0.056752074509859085, + "qparams": { + "group_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": 0.055388133972883224, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05506100505590439, + "qparams": { + "group_size": { + "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": 0.03260036185383797, + "qparams": { + "group_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": 0.02818930521607399, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028099747374653816, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025864260271191597, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025656109675765038, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016882628202438354, + "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": 0.016540654003620148, + "qparams": { + "group_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": 0.016431007534265518, + "qparams": { + "group_size": { + "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": 0.010349090211093426, + "qparams": { + "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": 0.03260036185383797, + "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": 0.03260036185383797, + "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": 0.25094205141067505, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22018083930015564, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20466214418411255, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1813737004995346, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11521399766206741, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10108152031898499, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1429746150970459, + "qparams": { + "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": 0.13166871666908264, + "qparams": { + "group_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": 0.12012434005737305, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09760348498821259, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09273362904787064, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07290981709957123, + "qparams": { + "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": 0.06310320645570755, + "qparams": { + "group_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": 0.05561013147234917, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05372467637062073, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03658388555049896, + "qparams": { + "group_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": 0.029031388461589813, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028269022703170776, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025524046272039413, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02426784485578537, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019467221572995186, + "qparams": { + "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": 0.019145499914884567, + "qparams": { + "group_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": 0.017063871026039124, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.012758743949234486, + "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": 0.05372467637062073, + "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": 0.05372467637062073, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.13281847536563873, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12464777380228043, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12151850759983063, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11043430119752884, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.062375571578741074, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05912870913743973, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07039966434240341, + "qparams": { + "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": 0.06506653130054474, + "qparams": { + "group_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": 0.06326800584793091, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05613293498754501, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05342159420251846, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03576085343956947, + "qparams": { + "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": 0.03112831525504589, + "qparams": { + "group_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": 0.02987966313958168, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02957765758037567, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01789577305316925, + "qparams": { + "group_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": 0.015304175205528736, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015186620876193047, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.014027951285243034, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013838955201208591, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009329789318144321, + "qparams": { + "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": 0.009233057498931885, + "qparams": { + "group_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": 0.008911265060305595, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005933708045631647, + "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": 0.05342159420251846, + "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": 0.05342159420251846, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "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": 0.11261323094367981, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10570210963487625, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10293536633253098, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09352768212556839, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05285800248384476, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.050055935978889465, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.060029275715351105, + "qparams": { + "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": 0.05535624176263809, + "qparams": { + "group_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": 0.05360865220427513, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04759025573730469, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.045377038419246674, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03049275651574135, + "qparams": { + "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": 0.026438971981406212, + "qparams": { + "group_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": 0.025310613214969635, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02503247931599617, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.015240189619362354, + "qparams": { + "group_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": 0.012938623316586018, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012830770574510098, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011859005317091942, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011686475947499275, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007904251106083393, + "qparams": { + "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": 0.007775309961289167, + "qparams": { + "group_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": 0.007526765577495098, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004935927223414183, + "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": 0.05360865220427513, + "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": 0.05360865220427513, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "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": 0.24523420631885529, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23034296929836273, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22559747099876404, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20506475865840912, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11515491455793381, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10974577069282532, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12855122983455658, + "qparams": { + "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": 0.11862795054912567, + "qparams": { + "group_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": 0.11679717153310776, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10371759533882141, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0987187996506691, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06529372930526733, + "qparams": { + "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": 0.05661532282829285, + "qparams": { + "group_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": 0.05504085123538971, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05466607213020325, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03257643058896065, + "qparams": { + "group_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": 0.027931077405810356, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027814922854304314, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025558192282915115, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025315912440419197, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016732877120375633, + "qparams": { + "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": 0.016303518787026405, + "qparams": { + "group_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": 0.016153354197740555, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.009989858604967594, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05466607213020325, + "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": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05466607213020325, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.2009768784046173, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.14447547495365143, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12073185294866562, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09624459594488144, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08530225604772568, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05954844132065773, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11425340175628662, + "qparams": { + "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": 0.1036701574921608, + "qparams": { + "group_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": 0.0952487513422966, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.058833047747612, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.056807950139045715, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05817653611302376, + "qparams": { + "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": 0.05028766766190529, + "qparams": { + "group_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": 0.0420440211892128, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.039864979684352875, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.029537931084632874, + "qparams": { + "group_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": 0.02317562885582447, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02262542024254799, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018460845574736595, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.016953516751527786, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01648661494255066, + "qparams": { + "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": 0.017106834799051285, + "qparams": { + "group_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": 0.014000789262354374, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.012876779772341251, + "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": 0.05028766766190529, + "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": 0.05028766766190529, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2020021229982376, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1901685744524002, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18646234273910522, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16985921561717987, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09537307918071747, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09110167622566223, + "qparams": { + "group_size": { + "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": 0.10602763295173645, + "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": 0.09790948778390884, + "qparams": { + "group_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": 0.09667623788118362, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0861022025346756, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08196716755628586, + "qparams": { + "group_size": { + "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": 0.05411384627223015, + "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": 0.046935904771089554, + "qparams": { + "group_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": 0.04578578472137451, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0455172024667263, + "qparams": { + "group_size": { + "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": 0.027128275483846664, + "qparams": { + "group_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": 0.023657385259866714, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023575635626912117, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021772796288132668, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021598709747195244, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.014459718950092793, + "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": 0.0144136818125844, + "qparams": { + "group_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": 0.014092476107180119, + "qparams": { + "group_size": { + "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": 0.009742793627083302, + "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": 0.05411384627223015, + "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": 0.05411384627223015, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.2485390305519104, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2340535968542099, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22954396903514862, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20912617444992065, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11708391457796097, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11184914410114288, + "qparams": { + "group_size": { + "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": 0.13020066916942596, + "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": 0.12012844532728195, + "qparams": { + "group_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": 0.11866176128387451, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10572927445173264, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1006539836525917, + "qparams": { + "group_size": { + "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": 0.06620092689990997, + "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": 0.05747942999005318, + "qparams": { + "group_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": 0.05609713867306709, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05576339364051819, + "qparams": { + "group_size": { + "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": 0.03318461775779724, + "qparams": { + "group_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": 0.028703104704618454, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028609082102775574, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026361895725131035, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026152536273002625, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01748821884393692, + "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": 0.017065808176994324, + "qparams": { + "group_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": 0.01703587919473648, + "qparams": { + "group_size": { + "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": 0.011000415310263634, + "qparams": { + "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": 0.03318461775779724, + "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": 0.03318461775779724, + "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": 0.25424259901046753, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22443944215774536, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20893509685993195, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1853107362985611, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11732182651758194, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10331547260284424, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14543533325195312, + "qparams": { + "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": 0.13408826291561127, + "qparams": { + "group_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": 0.12204252183437347, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10001938045024872, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09507272392511368, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07441012561321259, + "qparams": { + "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": 0.0647754892706871, + "qparams": { + "group_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": 0.05707314610481262, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05513306334614754, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.0375179685652256, + "qparams": { + "group_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": 0.030581418424844742, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029788441956043243, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.027256038039922714, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026014283299446106, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.020123444497585297, + "qparams": { + "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": 0.021081607788801193, + "qparams": { + "group_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": 0.017605243250727654, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015254423953592777, + "qparams": { + "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": 0.0375179685652256, + "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": 0.0375179685652256, + "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": 0.13579007983207703, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1274946928024292, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12450145184993744, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11316826194524765, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06377927958965302, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06057948246598244, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07169981300830841, + "qparams": { + "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": 0.06632229685783386, + "qparams": { + "group_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": 0.06470461934804916, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05743499472737312, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05457683280110359, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.036401472985744476, + "qparams": { + "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": 0.031679462641477585, + "qparams": { + "group_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": 0.03053015097975731, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.030249793082475662, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.018183600157499313, + "qparams": { + "group_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": 0.015554441139101982, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015451343730092049, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.014238393865525723, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01406205352395773, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009386634454131126, + "qparams": { + "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": 0.009229796007275581, + "qparams": { + "group_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": 0.008998840115964413, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005773785524070263, + "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": 0.05457683280110359, + "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": 0.05457683280110359, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11570437252521515, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10868144035339355, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10600978881120682, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09638149291276932, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05435338243842125, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05158202722668648, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.061457082629203796, + "qparams": { + "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": 0.056676216423511505, + "qparams": { + "group_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": 0.05512577295303345, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.048959244042634964, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04665984958410263, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.031200537458062172, + "qparams": { + "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": 0.027075646445155144, + "qparams": { + "group_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": 0.026031972840428352, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.025781577453017235, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01559173408895731, + "qparams": { + "group_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": 0.013288289308547974, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01319092232733965, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01217788364738226, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012016624212265015, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0080756526440382, + "qparams": { + "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": 0.007925991900265217, + "qparams": { + "group_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": 0.007726472336798906, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004994930233806372, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05435338243842125, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05435338243842125, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.25043171644210815, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2354646772146225, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23068511486053467, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.2097877711057663, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11781537532806396, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11236085742712021, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13123756647109985, + "qparams": { + "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": 0.12118025869131088, + "qparams": { + "group_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": 0.1194700300693512, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10618065297603607, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10096418857574463, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06668184697628021, + "qparams": { + "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": 0.05785693600773811, + "qparams": { + "group_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": 0.05631275475025177, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.055943217128515244, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03327125683426857, + "qparams": { + "group_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": 0.028569865971803665, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02845715917646885, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026140592992305756, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02590888924896717, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017091788351535797, + "qparams": { + "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": 0.0166472215205431, + "qparams": { + "group_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": 0.016519397497177124, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010182559490203857, + "qparams": { + "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": 0.03327125683426857, + "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": 0.03327125683426857, + "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": 0.21461215615272522, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15322080254554749, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.125774085521698, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10143444687128067, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09080882370471954, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06682364642620087, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12436046451330185, + "qparams": { + "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": 0.11415960639715195, + "qparams": { + "group_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": 0.10338573902845383, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0637798011302948, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.062433432787656784, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06426142156124115, + "qparams": { + "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": 0.05531827360391617, + "qparams": { + "group_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": 0.044701144099235535, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04185623303055763, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03243987634778023, + "qparams": { + "group_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": 0.024585282430052757, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023846866562962532, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.019756149500608444, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017801081761717796, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01809055730700493, + "qparams": { + "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": 0.01840090937912464, + "qparams": { + "group_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": 0.014914391562342644, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013613601215183735, + "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": 0.044701144099235535, + "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": 0.044701144099235535, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.20243242383003235, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.19044367969036102, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18670733273029327, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17003348469734192, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.095565065741539, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09118643403053284, + "qparams": { + "group_size": { + "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": 0.10627459734678268, + "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": 0.09811583161354065, + "qparams": { + "group_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": 0.09687851369380951, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08621419221162796, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08205410093069077, + "qparams": { + "group_size": { + "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": 0.05419255048036575, + "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": 0.047024257481098175, + "qparams": { + "group_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": 0.04586084559559822, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04558464512228966, + "qparams": { + "group_size": { + "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": 0.02712642028927803, + "qparams": { + "group_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": 0.023687614127993584, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02360525168478489, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021786293014883995, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02161087654531002, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.014353378675878048, + "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": 0.014424136839807034, + "qparams": { + "group_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": 0.01397489383816719, + "qparams": { + "group_size": { + "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": 0.00972976628690958, + "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": 0.05419255048036575, + "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": 0.05419255048036575, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.24846898019313812, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.23383738100528717, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2293100357055664, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.2087697833776474, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11706278473138809, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11177321523427963, + "qparams": { + "group_size": { + "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": 0.12986280024051666, + "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": 0.12014442682266235, + "qparams": { + "group_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": 0.11864721029996872, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10560395568609238, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10039521008729935, + "qparams": { + "group_size": { + "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": 0.06607340276241302, + "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": 0.05741109326481819, + "qparams": { + "group_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": 0.05601339414715767, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05567708611488342, + "qparams": { + "group_size": { + "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": 0.03298984840512276, + "qparams": { + "group_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": 0.028536105528473854, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028441565111279488, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02616208605468273, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02595202811062336, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.017111387103796005, + "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": 0.01678546331822872, + "qparams": { + "group_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": 0.01665167324244976, + "qparams": { + "group_size": { + "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": 0.010556770488619804, + "qparams": { + "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": 0.03298984840512276, + "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": 0.03298984840512276, + "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": 0.2512092590332031, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22228595614433289, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2064048945903778, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18302905559539795, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11584469676017761, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1019163429737091, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14621271193027496, + "qparams": { + "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": 0.13353002071380615, + "qparams": { + "group_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": 0.12043413519859314, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09904354065656662, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0943317636847496, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07496947050094604, + "qparams": { + "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": 0.06429198384284973, + "qparams": { + "group_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": 0.0562237985432148, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05417077988386154, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03822002559900284, + "qparams": { + "group_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": 0.02986418642103672, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028994427993893623, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026591304689645767, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02526506967842579, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.02140241116285324, + "qparams": { + "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": 0.020382210612297058, + "qparams": { + "group_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": 0.018883340060710907, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014329250901937485, + "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": 0.05417077988386154, + "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": 0.05417077988386154, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.14054648578166962, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.13202731311321259, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12883777916431427, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1171383410692215, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06606822460889816, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06274303793907166, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07441394031047821, + "qparams": { + "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": 0.06873184442520142, + "qparams": { + "group_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": 0.06701730191707611, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05947956442832947, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.056583184748888016, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03779444098472595, + "qparams": { + "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": 0.032847125083208084, + "qparams": { + "group_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": 0.03164129704236984, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03135903924703598, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01889565959572792, + "qparams": { + "group_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": 0.016164813190698624, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01605680212378502, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01480462308973074, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014625201001763344, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009801715612411499, + "qparams": { + "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": 0.009656542912125587, + "qparams": { + "group_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": 0.009385800920426846, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.006133196875452995, + "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": 0.03779444098472595, + "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": 0.03779444098472595, + "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": 0.11983094364404678, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11250709742307663, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1097593605518341, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09983810782432556, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.056366126984357834, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05348535627126694, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06378830224275589, + "qparams": { + "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": 0.05877012386918068, + "qparams": { + "group_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": 0.057170819491147995, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05075012519955635, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04840884730219841, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03243407607078552, + "qparams": { + "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": 0.028099656105041504, + "qparams": { + "group_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": 0.02700827270746231, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.026746226474642754, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01621437631547451, + "qparams": { + "group_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": 0.013797439634799957, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013696270994842052, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.012640339322388172, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012475499883294106, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008416497148573399, + "qparams": { + "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": 0.008250298909842968, + "qparams": { + "group_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": 0.008052822202444077, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005228801630437374, + "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": 0.05075012519955635, + "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": 0.05075012519955635, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.25291532278060913, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23770064115524292, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2329043596982956, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.21183638274669647, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11897456645965576, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11344139277935028, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13247179985046387, + "qparams": { + "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": 0.12234116345643997, + "qparams": { + "group_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": 0.12062256783246994, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10714336484670639, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10190334916114807, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06731832027435303, + "qparams": { + "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": 0.058410514146089554, + "qparams": { + "group_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": 0.05687208101153374, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05651329830288887, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03358563035726547, + "qparams": { + "group_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": 0.0288652665913105, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028763165697455406, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026410426944494247, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026173647493124008, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017275122925639153, + "qparams": { + "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": 0.01685716025531292, + "qparams": { + "group_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": 0.01671125553548336, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01037449762225151, + "qparams": { + "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": 0.03358563035726547, + "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": 0.03358563035726547, + "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": 0.23043803870677948, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.169979989528656, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.14363522827625275, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11446984112262726, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10205025225877762, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07638048380613327, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13240846991539001, + "qparams": { + "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": 0.12163448333740234, + "qparams": { + "group_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": 0.11114230751991272, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06596130132675171, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06672512739896774, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06807361543178558, + "qparams": { + "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": 0.058606602251529694, + "qparams": { + "group_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": 0.049750834703445435, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.047448620200157166, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.034156303852796555, + "qparams": { + "group_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": 0.026692509651184082, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025994420051574707, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.020062560215592384, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018295863643288612, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01845129393041134, + "qparams": { + "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": 0.01885807327926159, + "qparams": { + "group_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": 0.0156609658151865, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013449827209115028, + "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": 0.049750834703445435, + "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": 0.049750834703445435, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.19737866520881653, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18569231033325195, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1820429414510727, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16578418016433716, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0932314395904541, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08897209912538528, + "qparams": { + "group_size": { + "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": 0.10357436537742615, + "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": 0.09573713690042496, + "qparams": { + "group_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": 0.09451359510421753, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08407357335090637, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0800013393163681, + "qparams": { + "group_size": { + "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": 0.05290171131491661, + "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": 0.045884281396865845, + "qparams": { + "group_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": 0.04474036023020744, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.044472016394138336, + "qparams": { + "group_size": { + "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": 0.026457449421286583, + "qparams": { + "group_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": 0.023090966045856476, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02301645651459694, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.021235471591353416, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02106350287795067, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013980293646454811, + "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": 0.014039277099072933, + "qparams": { + "group_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": 0.013605832122266293, + "qparams": { + "group_size": { + "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": 0.0094474907964468, + "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": 0.05290171131491661, + "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": 0.05290171131491661, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.24810880422592163, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.23350343108177185, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22896720468997955, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20845867693424225, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11698342859745026, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11164098978042603, + "qparams": { + "group_size": { + "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": 0.1298523098230362, + "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": 0.12005588412284851, + "qparams": { + "group_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": 0.11856115609407425, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10550552606582642, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1003318652510643, + "qparams": { + "group_size": { + "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": 0.06618467718362808, + "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": 0.05740465596318245, + "qparams": { + "group_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": 0.055997334420681, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05566314235329628, + "qparams": { + "group_size": { + "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": 0.03303898498415947, + "qparams": { + "group_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": 0.02859521470963955, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028499562293291092, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026229670271277428, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026013022288680077, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01720571704208851, + "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": 0.016917582601308823, + "qparams": { + "group_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": 0.016741683706641197, + "qparams": { + "group_size": { + "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": 0.010784750804305077, + "qparams": { + "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": 0.03303898498415947, + "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": 0.03303898498415947, + "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": 0.2532700002193451, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.224094957113266, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20880834758281708, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1854490488767624, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11676687747240067, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10311414301395416, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14500030875205994, + "qparams": { + "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": 0.1336502879858017, + "qparams": { + "group_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": 0.12146072089672089, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10003767162561417, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09492500126361847, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0739392340183258, + "qparams": { + "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": 0.06440751254558563, + "qparams": { + "group_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": 0.05667075514793396, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05472835898399353, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03722040727734566, + "qparams": { + "group_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": 0.030117932707071304, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029312031343579292, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026861924678087234, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025600001215934753, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.020025020465254784, + "qparams": { + "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": 0.02050095982849598, + "qparams": { + "group_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": 0.01760510355234146, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014483772218227386, + "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": 0.05472835898399353, + "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": 0.05472835898399353, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.13582664728164673, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12750649452209473, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12441255152225494, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11299487948417664, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06387868523597717, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.060602668672800064, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07187756896018982, + "qparams": { + "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": 0.06651293486356735, + "qparams": { + "group_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": 0.06479153037071228, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05744372308254242, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05455997586250305, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0364803671836853, + "qparams": { + "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": 0.031773168593645096, + "qparams": { + "group_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": 0.03057866171002388, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03029567375779152, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01823091134428978, + "qparams": { + "group_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": 0.0156093155965209, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015499397180974483, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01427986565977335, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014097745530307293, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009447328746318817, + "qparams": { + "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": 0.009303191676735878, + "qparams": { + "group_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": 0.009049070999026299, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0058694747276604176, + "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": 0.05455997586250305, + "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": 0.05455997586250305, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "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": 0.11451268196105957, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10748336464166641, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10483575612306595, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09517086297273636, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05384743586182594, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05105515196919441, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.060952022671699524, + "qparams": { + "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": 0.05618526414036751, + "qparams": { + "group_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": 0.05463015288114548, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04844897612929344, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04614811763167381, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.030967295169830322, + "qparams": { + "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": 0.0268628541380167, + "qparams": { + "group_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": 0.025795994326472282, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.025543441995978355, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.015476522035896778, + "qparams": { + "group_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": 0.013186290860176086, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013087286613881588, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.012071949429810047, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011907951906323433, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008035809732973576, + "qparams": { + "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": 0.007897119969129562, + "qparams": { + "group_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": 0.0076818922534585, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005010991357266903, + "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": 0.05463015288114548, + "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": 0.05463015288114548, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "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": 0.2535451352596283, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2381598800420761, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.233271062374115, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.21208904683589935, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.1192842647433281, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11371837556362152, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13283342123031616, + "qparams": { + "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": 0.12266384065151215, + "qparams": { + "group_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": 0.12097998708486557, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10735684633255005, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10204647481441498, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0674804225564003, + "qparams": { + "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": 0.05858315899968147, + "qparams": { + "group_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": 0.05701799318194389, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.056650735437870026, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03366969898343086, + "qparams": { + "group_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": 0.02893744967877865, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02882547676563263, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026454482227563858, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02620929479598999, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017281996086239815, + "qparams": { + "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": 0.016869105398654938, + "qparams": { + "group_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": 0.01670457236468792, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010328032076358795, + "qparams": { + "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": 0.03366969898343086, + "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": 0.03366969898343086, + "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": 0.19805143773555756, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.14668293297290802, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12282244116067886, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10629651695489883, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08966653794050217, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06708036363124847, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11625876277685165, + "qparams": { + "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": 0.10635507106781006, + "qparams": { + "group_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": 0.09550253301858902, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06296191364526749, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05972570553421974, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06008707359433174, + "qparams": { + "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": 0.05130694806575775, + "qparams": { + "group_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": 0.04365224391222, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.041691023856401443, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03003607876598835, + "qparams": { + "group_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": 0.023305708542466164, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02255999855697155, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018270321190357208, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01683889329433441, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016142146661877632, + "qparams": { + "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": 0.01628701575100422, + "qparams": { + "group_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": 0.013638317584991455, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.011491715908050537, + "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": 0.05130694806575775, + "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": 0.05130694806575775, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.19438327848911285, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18276731669902802, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.179073765873909, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16296854615211487, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09182476997375488, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08761857450008392, + "qparams": { + "group_size": { + "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": 0.1020183116197586, + "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": 0.09432704746723175, + "qparams": { + "group_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": 0.09309758991003036, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08273951709270477, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07864801585674286, + "qparams": { + "group_size": { + "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": 0.05204077437520027, + "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": 0.045213691890239716, + "qparams": { + "group_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": 0.044084332883358, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04381707310676575, + "qparams": { + "group_size": { + "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": 0.026058152318000793, + "qparams": { + "group_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": 0.02276158332824707, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022682825103402138, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02091371826827526, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020742325112223625, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01375561486929655, + "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": 0.013854359276592731, + "qparams": { + "group_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": 0.01338205300271511, + "qparams": { + "group_size": { + "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": 0.00934047345072031, + "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": 0.05204077437520027, + "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": 0.05204077437520027, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.24725928902626038, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.23256200551986694, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.227958545088768, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20745015144348145, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11655845493078232, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11122249066829681, + "qparams": { + "group_size": { + "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": 0.12932874262332916, + "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": 0.11961759626865387, + "qparams": { + "group_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": 0.11812259256839752, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10500496625900269, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09976664930582047, + "qparams": { + "group_size": { + "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": 0.06580217182636261, + "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": 0.057166967540979385, + "qparams": { + "group_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": 0.055750831961631775, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.055426161736249924, + "qparams": { + "group_size": { + "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": 0.032839544117450714, + "qparams": { + "group_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": 0.028384936973452568, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0282910093665123, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02599930949509144, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02578883431851864, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016982536762952805, + "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": 0.016664937138557434, + "qparams": { + "group_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": 0.016513854265213013, + "qparams": { + "group_size": { + "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": 0.010428043082356453, + "qparams": { + "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": 0.032839544117450714, + "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": 0.032839544117450714, + "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": 0.25299787521362305, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22375303506851196, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20777195692062378, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18461567163467407, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11670860648155212, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10276374220848083, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1484961360692978, + "qparams": { + "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": 0.1345912218093872, + "qparams": { + "group_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": 0.12154880166053772, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10002957284450531, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09546759724617004, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0754501074552536, + "qparams": { + "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": 0.06500394642353058, + "qparams": { + "group_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": 0.056871190667152405, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.054817166179418564, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03830619156360626, + "qparams": { + "group_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": 0.030608540400862694, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029750598594546318, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02742154709994793, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02610272727906704, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.02129531279206276, + "qparams": { + "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": 0.021318262442946434, + "qparams": { + "group_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": 0.018727706745266914, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.015524892136454582, + "qparams": { + "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": 0.03830619156360626, + "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": 0.03830619156360626, + "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": 0.13776999711990356, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12920960783958435, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1260225921869278, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11436820775270462, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06476006656885147, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.061394259333610535, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07298967987298965, + "qparams": { + "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": 0.06749608367681503, + "qparams": { + "group_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": 0.06572265177965164, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05818971246480942, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.055259522050619125, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03704899922013283, + "qparams": { + "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": 0.032262299209833145, + "qparams": { + "group_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": 0.031012514606118202, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.030716175213456154, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.0185115784406662, + "qparams": { + "group_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": 0.015842875465750694, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015727423131465912, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.014480500482022762, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01429070346057415, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009592029266059399, + "qparams": { + "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": 0.009469597600400448, + "qparams": { + "group_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": 0.009175357408821583, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.00598762184381485, + "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": 0.03704899922013283, + "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": 0.03704899922013283, + "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": 0.11496268212795258, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1077851951122284, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10499243438243866, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09535612165927887, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05405009910464287, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.051193658262491226, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06132502853870392, + "qparams": { + "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": 0.056512631475925446, + "qparams": { + "group_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": 0.05484457314014435, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04859849438071251, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.046280719339847565, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03116856887936592, + "qparams": { + "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": 0.027022231370210648, + "qparams": { + "group_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": 0.02589055709540844, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.025633495301008224, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.015578309074044228, + "qparams": { + "group_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": 0.013235745020210743, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013128818944096565, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.012110631912946701, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011941082775592804, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0080937584862113, + "qparams": { + "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": 0.007934964261949062, + "qparams": { + "group_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": 0.0077214594930410385, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005036793649196625, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05405009910464287, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05405009910464287, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.259222149848938, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.24331903457641602, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23835395276546478, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.21652913093566895, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.12204226851463318, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11627698689699173, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.135910302400589, + "qparams": { + "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": 0.12550115585327148, + "qparams": { + "group_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": 0.12379930913448334, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10973939299583435, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10420064628124237, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06907916069030762, + "qparams": { + "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": 0.05996100604534149, + "qparams": { + "group_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": 0.058352090418338776, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05797233060002327, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03447098657488823, + "qparams": { + "group_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": 0.029629172757267952, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02951575070619583, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02706177718937397, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026811132207512856, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017713986337184906, + "qparams": { + "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": 0.017293399199843407, + "qparams": { + "group_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": 0.017119646072387695, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010634281672537327, + "qparams": { + "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": 0.03447098657488823, + "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": 0.03447098657488823, + "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": 0.21334728598594666, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15002219378948212, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1194286122918129, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10093197226524353, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0906180888414383, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06427019834518433, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12640129029750824, + "qparams": { + "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": 0.11553927510976791, + "qparams": { + "group_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": 0.1025695949792862, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.061841633170843124, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06345473229885101, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06523874402046204, + "qparams": { + "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": 0.05557244271039963, + "qparams": { + "group_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": 0.044167134910821915, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04103124514222145, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03268597647547722, + "qparams": { + "group_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": 0.023583978414535522, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022637562826275826, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018201490864157677, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015859369188547134, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01756424270570278, + "qparams": { + "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": 0.01713709346950054, + "qparams": { + "group_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": 0.013882068917155266, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.011634161695837975, + "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": 0.044167134910821915, + "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": 0.044167134910821915, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.18952509760856628, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17806009948253632, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17453601956367493, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15875528752803802, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08954982459545135, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08540395647287369, + "qparams": { + "group_size": { + "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": 0.09949809312820435, + "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": 0.09201734513044357, + "qparams": { + "group_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": 0.09082095324993134, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08065181970596313, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07667817175388336, + "qparams": { + "group_size": { + "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": 0.050815399736166, + "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": 0.044118065387010574, + "qparams": { + "group_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": 0.043002065271139145, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04273974150419235, + "qparams": { + "group_size": { + "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": 0.025428486987948418, + "qparams": { + "group_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": 0.02221277728676796, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022136813029646873, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02040531113743782, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020235925912857056, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013420679606497288, + "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": 0.013540912419557571, + "qparams": { + "group_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": 0.013056032359600067, + "qparams": { + "group_size": { + "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": 0.009156150743365288, + "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": 0.050815399736166, + "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": 0.050815399736166, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.2443184107542038, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2296917885541916, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2251949906349182, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20480483770370483, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.1152353584766388, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.109974205493927, + "qparams": { + "group_size": { + "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": 0.12792757153511047, + "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": 0.11832819879055023, + "qparams": { + "group_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": 0.11685522645711899, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1038200706243515, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09872609376907349, + "qparams": { + "group_size": { + "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": 0.0652107372879982, + "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": 0.05659865587949753, + "qparams": { + "group_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": 0.05519789829850197, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05486171320080757, + "qparams": { + "group_size": { + "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": 0.03256290778517723, + "qparams": { + "group_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": 0.028179189190268517, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0280862245708704, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025818055495619774, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025595787912607193, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016937077045440674, + "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": 0.016667140647768974, + "qparams": { + "group_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": 0.016472147777676582, + "qparams": { + "group_size": { + "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": 0.010606415569782257, + "qparams": { + "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": 0.03256290778517723, + "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": 0.03256290778517723, + "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": 0.2448022961616516, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2165357768535614, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.200750470161438, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17852911353111267, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11268401890993118, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09907282888889313, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14211636781692505, + "qparams": { + "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": 0.13056765496730804, + "qparams": { + "group_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": 0.11744226515293121, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09666067361831665, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09212586283683777, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07267846912145615, + "qparams": { + "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": 0.0627269446849823, + "qparams": { + "group_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": 0.05461172014474869, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05254082381725311, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03670765832066536, + "qparams": { + "group_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": 0.028842177242040634, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027966098859906197, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02569461241364479, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02434317208826542, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019965827465057373, + "qparams": { + "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": 0.01954493299126625, + "qparams": { + "group_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": 0.017422178760170937, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013461301103234291, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05461172014474869, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05461172014474869, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.13190285861492157, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1236809492111206, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12053656578063965, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10938906669616699, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06203637272119522, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05875483527779579, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07010623067617416, + "qparams": { + "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": 0.0648016557097435, + "qparams": { + "group_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": 0.06295444071292877, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05572834238409996, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05297784134745598, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03563380986452103, + "qparams": { + "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": 0.03098875842988491, + "qparams": { + "group_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": 0.029717687517404556, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.029415559023618698, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.017802594229578972, + "qparams": { + "group_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": 0.015179900452494621, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015066588297486305, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013883794657886028, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013686960563063622, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009239108301699162, + "qparams": { + "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": 0.00909516867250204, + "qparams": { + "group_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": 0.008819321170449257, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005755476653575897, + "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": 0.05297784134745598, + "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": 0.05297784134745598, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "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": 0.11181365698575974, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10483033955097198, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10212725400924683, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09268708527088165, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.052574869245290756, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04978933930397034, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05952294170856476, + "qparams": { + "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": 0.05497109144926071, + "qparams": { + "group_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": 0.05334807559847832, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04725922271609306, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04493887722492218, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.030251838266849518, + "qparams": { + "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": 0.026282329112291336, + "qparams": { + "group_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": 0.0251869298517704, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.024924788624048233, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01512083038687706, + "qparams": { + "group_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": 0.012875326909124851, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012769750319421291, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011776932515203953, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011608573608100414, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0078438026830554, + "qparams": { + "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": 0.007720509078353643, + "qparams": { + "group_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": 0.007479433435946703, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004894856829196215, + "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": 0.05334807559847832, + "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": 0.05334807559847832, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "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": 0.25030624866485596, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23497164249420166, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23014713823795319, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.2090689092874527, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11787525564432144, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11227242648601532, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13126853108406067, + "qparams": { + "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": 0.1212972104549408, + "qparams": { + "group_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": 0.11952534317970276, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10597695410251617, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10054299235343933, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0667029544711113, + "qparams": { + "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": 0.05793971195816994, + "qparams": { + "group_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": 0.056334882974624634, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05596393719315529, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03328239545226097, + "qparams": { + "group_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": 0.02857457660138607, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028458960354328156, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02609657123684883, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025848455727100372, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.017071355134248734, + "qparams": { + "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": 0.016645628958940506, + "qparams": { + "group_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": 0.016497327014803886, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010162098333239555, + "qparams": { + "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": 0.03328239545226097, + "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": 0.03328239545226097, + "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": 0.20212531089782715, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15428388118743896, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.13332070410251617, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10805689543485641, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0901348665356636, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06790741533041, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11774501949548721, + "qparams": { + "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": 0.10639883577823639, + "qparams": { + "group_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": 0.09674671292304993, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0632292777299881, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06114540621638298, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06015940383076668, + "qparams": { + "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": 0.051662709563970566, + "qparams": { + "group_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": 0.04432656615972519, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.042435240000486374, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03064669296145439, + "qparams": { + "group_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": 0.024523328989744186, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023891661316156387, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.019763261079788208, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0184615571051836, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0175717081874609, + "qparams": { + "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": 0.017962681129574776, + "qparams": { + "group_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": 0.015448618680238724, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013758032582700253, + "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": 0.051662709563970566, + "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": 0.051662709563970566, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.1865144520998001, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17519983649253845, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17168734967708588, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15604013204574585, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0881408303976059, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08403903990983963, + "qparams": { + "group_size": { + "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": 0.09799911081790924, + "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": 0.09060346335172653, + "qparams": { + "group_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": 0.08940853178501129, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07931079715490341, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07537841796875, + "qparams": { + "group_size": { + "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": 0.050005823373794556, + "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": 0.04344213753938675, + "qparams": { + "group_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": 0.0423298105597496, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0420604906976223, + "qparams": { + "group_size": { + "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": 0.02501353807747364, + "qparams": { + "group_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": 0.021858643740415573, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021785805001854897, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02006523869931698, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01990031637251377, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013174479827284813, + "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": 0.013320069760084152, + "qparams": { + "group_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": 0.012808924540877342, + "qparams": { + "group_size": { + "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": 0.008990042842924595, + "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": 0.050005823373794556, + "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": 0.050005823373794556, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.24225746095180511, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22762833535671234, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.22305059432983398, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.2028089314699173, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11432751268148422, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10901398956775665, + "qparams": { + "group_size": { + "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": 0.12711584568023682, + "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": 0.11740466207265854, + "qparams": { + "group_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": 0.11592991650104523, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1028800830245018, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09777314215898514, + "qparams": { + "group_size": { + "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": 0.06477245688438416, + "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": 0.056194696575403214, + "qparams": { + "group_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": 0.054779939353466034, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05444467440247536, + "qparams": { + "group_size": { + "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": 0.03239421546459198, + "qparams": { + "group_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": 0.02803296595811844, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02794082649052143, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02567501924932003, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025463320314884186, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.016977515071630478, + "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": 0.016685284674167633, + "qparams": { + "group_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": 0.01651618629693985, + "qparams": { + "group_size": { + "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": 0.010757612995803356, + "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": 0.05444467440247536, + "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": 0.05444467440247536, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.25357258319854736, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22409449517726898, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20822051167488098, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18493837118148804, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.116892509162426, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10287424176931381, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14711782336235046, + "qparams": { + "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": 0.1345691829919815, + "qparams": { + "group_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": 0.1217912957072258, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10004478693008423, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09523063898086548, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0749678909778595, + "qparams": { + "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": 0.06479877978563309, + "qparams": { + "group_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": 0.05672057718038559, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05465269833803177, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03822150453925133, + "qparams": { + "group_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": 0.030045051127672195, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02918216772377491, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026750605553388596, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02540215477347374, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.02131503075361252, + "qparams": { + "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": 0.02041296660900116, + "qparams": { + "group_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": 0.018866753205657005, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014220139011740685, + "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": 0.05465269833803177, + "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": 0.05465269833803177, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.13500314950942993, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12641020119190216, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12321814894676208, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1116960346698761, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06342918425798416, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06006183847784996, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07160333544015884, + "qparams": { + "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": 0.06627707928419113, + "qparams": { + "group_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": 0.06436308473348618, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05689907819032669, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05400749295949936, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03635343164205551, + "qparams": { + "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": 0.031662799417972565, + "qparams": { + "group_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": 0.030384572222828865, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03007880598306656, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.018164489418268204, + "qparams": { + "group_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": 0.01553618349134922, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015417827293276787, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.014192561618983746, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013993761502206326, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009429218247532845, + "qparams": { + "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": 0.009318123571574688, + "qparams": { + "group_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": 0.00900785718113184, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005921858828514814, + "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": 0.05400749295949936, + "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": 0.05400749295949936, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1114712730050087, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10441858321428299, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10161212831735611, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0921381339430809, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.052435193210840225, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04957348108291626, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05961206927895546, + "qparams": { + "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": 0.05497767776250839, + "qparams": { + "group_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": 0.053194012492895126, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04704703390598297, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04477105289697647, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.030290402472019196, + "qparams": { + "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": 0.02630559168756008, + "qparams": { + "group_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": 0.025134164839982986, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.024858297780156136, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.015155903063714504, + "qparams": { + "group_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": 0.012874887324869633, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012761999852955341, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011773727834224701, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011595051735639572, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007895849645137787, + "qparams": { + "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": 0.007777981925755739, + "qparams": { + "group_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": 0.007510788273066282, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004984239116311073, + "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": 0.053194012492895126, + "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": 0.053194012492895126, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "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": 0.25507840514183044, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23936741054058075, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23434457182884216, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.21272744238376617, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.12012507021427155, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11438749730587006, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13384899497032166, + "qparams": { + "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": 0.12366174906492233, + "qparams": { + "group_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": 0.12185925245285034, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10793570429086685, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.1024417132139206, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06807402521371841, + "qparams": { + "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": 0.0590333417057991, + "qparams": { + "group_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": 0.05742618069052696, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05704144015908241, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03395070135593414, + "qparams": { + "group_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": 0.029140904545783997, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.029023248702287674, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02659691870212555, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026354391127824783, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01743757724761963, + "qparams": { + "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": 0.01699434593319893, + "qparams": { + "group_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": 0.016838762909173965, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010409349575638771, + "qparams": { + "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": 0.03395070135593414, + "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": 0.03395070135593414, + "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": 0.20100750029087067, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1584112048149109, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.140045627951622, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10185927152633667, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09178560227155685, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07257908582687378, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11450641602277756, + "qparams": { + "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": 0.10545103251934052, + "qparams": { + "group_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": 0.09632375836372375, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06192340701818466, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05768468976020813, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.058573558926582336, + "qparams": { + "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": 0.05058145895600319, + "qparams": { + "group_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": 0.04432491585612297, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04274572432041168, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.029297899454832077, + "qparams": { + "group_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": 0.023232821375131607, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02261143922805786, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.017485208809375763, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.016255106776952744, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.015520954504609108, + "qparams": { + "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": 0.015500332228839397, + "qparams": { + "group_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": 0.013501848094165325, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010442765429615974, + "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": 0.05058145895600319, + "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": 0.05058145895600319, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.1851232796907425, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1737861931324005, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1702754944562912, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15485885739326477, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08752301335334778, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08340796828269958, + "qparams": { + "group_size": { + "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": 0.09726325422525406, + "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": 0.08994676917791367, + "qparams": { + "group_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": 0.08876819163560867, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07874441146850586, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07476042211055756, + "qparams": { + "group_size": { + "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": 0.049655672162771225, + "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": 0.04313257709145546, + "qparams": { + "group_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": 0.04203573241829872, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04177020117640495, + "qparams": { + "group_size": { + "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": 0.02483641356229782, + "qparams": { + "group_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": 0.021735183894634247, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02165888249874115, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.019959820434451103, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01979554072022438, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013100621290504932, + "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": 0.013293526135385036, + "qparams": { + "group_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": 0.012740974314510822, + "qparams": { + "group_size": { + "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": 0.009023342281579971, + "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": 0.049655672162771225, + "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": 0.049655672162771225, + "qparams": { + "group_size": { + "4": 128 + }, + "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": 0.2336680144071579, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21959882974624634, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21512989699840546, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1956963837146759, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11032310128211975, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10516936331987381, + "qparams": { + "group_size": { + "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": 0.12261122465133667, + "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": 0.11328200250864029, + "qparams": { + "group_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": 0.11187960207462311, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09927037358283997, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09433505684137344, + "qparams": { + "group_size": { + "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": 0.0624840073287487, + "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": 0.05422452837228775, + "qparams": { + "group_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": 0.05286569148302078, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0525447241961956, + "qparams": { + "group_size": { + "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": 0.031232766807079315, + "qparams": { + "group_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": 0.0270499549806118, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026960166171193123, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024773811921477318, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024564286693930626, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01631850376725197, + "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": 0.01608794741332531, + "qparams": { + "group_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": 0.01586741767823696, + "qparams": { + "group_size": { + "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": 0.010356280952692032, + "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": 0.05422452837228775, + "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": 0.05422452837228775, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.24639302492141724, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21627177298069, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2004500925540924, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17787164449691772, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11298568546772003, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0990217998623848, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.14184719324111938, + "qparams": { + "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": 0.13054805994033813, + "qparams": { + "group_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": 0.11824695765972137, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09636791050434113, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09183522313833237, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07257580012083054, + "qparams": { + "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": 0.06269146502017975, + "qparams": { + "group_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": 0.05473531037569046, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05270995944738388, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03647303953766823, + "qparams": { + "group_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": 0.02887524664402008, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028061464428901672, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025609323754906654, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024286895990371704, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01957225240767002, + "qparams": { + "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": 0.019497958943247795, + "qparams": { + "group_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": 0.017025554552674294, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013406232930719852, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05473531037569046, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05473531037569046, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.12978681921958923, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.12161970138549805, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.118490070104599, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1073964536190033, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.061012279242277145, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05775934085249901, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06886183470487595, + "qparams": { + "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": 0.06375843286514282, + "qparams": { + "group_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": 0.06191844120621681, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05475067347288132, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05189663544297218, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03495951369404793, + "qparams": { + "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": 0.030473966151475906, + "qparams": { + "group_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": 0.02922344207763672, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.028934385627508163, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.01747288927435875, + "qparams": { + "group_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": 0.014953340403735638, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014834815636277199, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013656032271683216, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013466965407133102, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.009068065322935581, + "qparams": { + "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": 0.008973993360996246, + "qparams": { + "group_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": 0.008655265904963017, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005708745680749416, + "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": 0.05475067347288132, + "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": 0.05475067347288132, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10691038519144058, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10015635937452316, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09753835201263428, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08841794729232788, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05026116594672203, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.047545190900564194, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05705142766237259, + "qparams": { + "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": 0.05267232283949852, + "qparams": { + "group_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": 0.05100183188915253, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.045118656009435654, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04285828024148941, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.028985919430851936, + "qparams": { + "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": 0.025177154690027237, + "qparams": { + "group_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": 0.024089252576231956, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02382829412817955, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014503410086035728, + "qparams": { + "group_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": 0.012341577559709549, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01223598513752222, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011284848675131798, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01111628394573927, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007556579075753689, + "qparams": { + "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": 0.007449626922607422, + "qparams": { + "group_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": 0.007196090649813414, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004773928318172693, + "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": 0.05267232283949852, + "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": 0.05267232283949852, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.25193965435028076, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23642905056476593, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23150715231895447, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.21014562249183655, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11868292093276978, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11301469057798386, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13221831619739532, + "qparams": { + "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": 0.1221534013748169, + "qparams": { + "group_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": 0.1203790009021759, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10660892724990845, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10111594945192337, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06719815731048584, + "qparams": { + "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": 0.05831942707300186, + "qparams": { + "group_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": 0.056736089289188385, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05637175217270851, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03353354334831238, + "qparams": { + "group_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": 0.028803730383515358, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02869066782295704, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.026289433240890503, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026045192033052444, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01722557656466961, + "qparams": { + "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": 0.016796231269836426, + "qparams": { + "group_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": 0.01664738543331623, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010294945910573006, + "qparams": { + "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": 0.03353354334831238, + "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": 0.03353354334831238, + "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": 0.18910948932170868, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.15426668524742126, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1427278220653534, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.107673779129982, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08626700937747955, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07322805374860764, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.10219847410917282, + "qparams": { + "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": 0.09392040222883224, + "qparams": { + "group_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": 0.09086072444915771, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06265033036470413, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05298590660095215, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05246550962328911, + "qparams": { + "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": 0.04515444487333298, + "qparams": { + "group_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": 0.041717518121004105, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04086800292134285, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02628408931195736, + "qparams": { + "group_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": 0.021950814872980118, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021762173622846603, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01738950051367283, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01678440533578396, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.014142057858407497, + "qparams": { + "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": 0.01434599980711937, + "qparams": { + "group_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": 0.01305735670030117, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01010963972657919, + "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": 0.05298590660095215, + "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": 0.05298590660095215, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "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": 0.18159006536006927, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.17049771547317505, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.16709928214550018, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.15182580053806305, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08584462106227875, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08179554343223572, + "qparams": { + "group_size": { + "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": 0.09537789225578308, + "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": 0.08824357390403748, + "qparams": { + "group_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": 0.08707057684659958, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07721275836229324, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07333603501319885, + "qparams": { + "group_size": { + "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": 0.04861083999276161, + "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": 0.04230395704507828, + "qparams": { + "group_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": 0.04121158644556999, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.040952879935503006, + "qparams": { + "group_size": { + "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": 0.024313513189554214, + "qparams": { + "group_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": 0.021264778450131416, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02118920348584652, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.019507525488734245, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0193425752222538, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.0127568319439888, + "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": 0.012931302189826965, + "qparams": { + "group_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": 0.012399299070239067, + "qparams": { + "group_size": { + "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": 0.008690346963703632, + "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": 0.04861083999276161, + "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": 0.04861083999276161, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.22503063082695007, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.21136672794818878, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.2070915251970291, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18827345967292786, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10620058327913284, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10123144835233688, + "qparams": { + "group_size": { + "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": 0.11797142028808594, + "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": 0.10910235345363617, + "qparams": { + "group_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": 0.10769202560186386, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09554963558912277, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09074532985687256, + "qparams": { + "group_size": { + "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": 0.060077037662267685, + "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": 0.052178919315338135, + "qparams": { + "group_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": 0.050856854766607285, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05054426193237305, + "qparams": { + "group_size": { + "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": 0.029985258355736732, + "qparams": { + "group_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": 0.02593141794204712, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025845441967248917, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02372170053422451, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02352234721183777, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.015535577200353146, + "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": 0.015299766324460506, + "qparams": { + "group_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": 0.015092256478965282, + "qparams": { + "group_size": { + "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": 0.00966828316450119, + "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": 0.052178919315338135, + "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": 0.052178919315338135, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.2450784593820572, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21337071061134338, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19761835038661957, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17555460333824158, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.111907958984375, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0976763442158699, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1402783840894699, + "qparams": { + "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": 0.12901677191257477, + "qparams": { + "group_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": 0.11767901480197906, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09503783285617828, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09044807404279709, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07160777598619461, + "qparams": { + "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": 0.061918411403894424, + "qparams": { + "group_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": 0.054130930453538895, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.052156608551740646, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03595048189163208, + "qparams": { + "group_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": 0.028411537408828735, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02765350416302681, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.025073198601603508, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023760782554745674, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019148195162415504, + "qparams": { + "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": 0.01900000125169754, + "qparams": { + "group_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": 0.01662786491215229, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.012835848145186901, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.054130930453538895, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.054130930453538895, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.12426650524139404, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11632942408323288, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11313794553279877, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1024094820022583, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.058338068425655365, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.055116601288318634, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06623777002096176, + "qparams": { + "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": 0.06124714016914368, + "qparams": { + "group_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": 0.059233229607343674, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05228511616587639, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.049604382365942, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03363870829343796, + "qparams": { + "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": 0.029296398162841797, + "qparams": { + "group_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": 0.027977511286735535, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02765829674899578, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.016827134415507317, + "qparams": { + "group_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": 0.014323463663458824, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014197682030498981, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.013075805269181728, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012877333909273148, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008756079711019993, + "qparams": { + "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": 0.008648599497973919, + "qparams": { + "group_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": 0.008333658799529076, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0055319457314908504, + "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": 0.05228511616587639, + "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": 0.05228511616587639, + "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.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10042297095060349, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09396423399448395, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09129446744918823, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08264312148094177, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.047148752957582474, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04448153078556061, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.053759392350912094, + "qparams": { + "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": 0.04970964789390564, + "qparams": { + "group_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": 0.04786670207977295, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.042265135794878006, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04010393097996712, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.027317119762301445, + "qparams": { + "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": 0.023766187950968742, + "qparams": { + "group_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": 0.022602463141083717, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02232944965362549, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013662879355251789, + "qparams": { + "group_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": 0.011593589559197426, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011474218219518661, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010589340701699257, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010411737486720085, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007123380433768034, + "qparams": { + "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": 0.0070371259935200214, + "qparams": { + "group_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": 0.006748386193066835, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.00451927725225687, + "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": 0.053759392350912094, + "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": 0.053759392350912094, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.24612073600292206, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.23086240887641907, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22598636150360107, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.20495471358299255, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11594755202531815, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.11037133634090424, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.1294177621603012, + "qparams": { + "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": 0.11943136900663376, + "qparams": { + "group_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": 0.1175815537571907, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10411378741264343, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09873665869235992, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06578734517097473, + "qparams": { + "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": 0.05706486105918884, + "qparams": { + "group_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": 0.05547954887151718, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05510098859667778, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.032850172370672226, + "qparams": { + "group_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": 0.02820519357919693, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.028088973835110664, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02573813870549202, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02549411915242672, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016906440258026123, + "qparams": { + "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": 0.016535142436623573, + "qparams": { + "group_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": 0.01630586013197899, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010253218933939934, + "qparams": { + "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": 0.032850172370672226, + "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": 0.032850172370672226, + "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": 0.19214580953121185, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.14985357224941254, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.13471443951129913, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1048320010304451, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08662783354520798, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07018184661865234, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.10617172718048096, + "qparams": { + "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": 0.0970250591635704, + "qparams": { + "group_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": 0.09283030778169632, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05997705087065697, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05507296696305275, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.05478796362876892, + "qparams": { + "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": 0.046848565340042114, + "qparams": { + "group_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": 0.04209277406334877, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.040901925414800644, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.02759074606001377, + "qparams": { + "group_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": 0.02239232137799263, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022111376747488976, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01734054833650589, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.016474906355142593, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01513600256294012, + "qparams": { + "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": 0.015137174166738987, + "qparams": { + "group_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": 0.013631814159452915, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010875038802623749, + "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": 0.046848565340042114, + "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": 0.046848565340042114, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.1745283603668213, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.16378897428512573, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1604149043560028, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.14577540755271912, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.08244377374649048, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07854782789945602, + "qparams": { + "group_size": { + "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": 0.09169553965330124, + "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": 0.08478281646966934, + "qparams": { + "group_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": 0.08363648504018784, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07412289083003998, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07039950042963028, + "qparams": { + "group_size": { + "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": 0.04678172990679741, + "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": 0.04064832627773285, + "qparams": { + "group_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": 0.039583414793014526, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0393301360309124, + "qparams": { + "group_size": { + "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": 0.023405160754919052, + "qparams": { + "group_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": 0.020441122353076935, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02037338726222515, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018749693408608437, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018589461222290993, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.01231966633349657, + "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": 0.012454258278012276, + "qparams": { + "group_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": 0.011963156051933765, + "qparams": { + "group_size": { + "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": 0.008396491408348083, + "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": 0.04678172990679741, + "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": 0.04678172990679741, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.21063460409641266, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1977650374174118, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.19384199380874634, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.17611965537071228, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09940403699874878, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09471984207630157, + "qparams": { + "group_size": { + "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": 0.11047637462615967, + "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": 0.10215459018945694, + "qparams": { + "group_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": 0.10082191228866577, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08939098566770554, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08484892547130585, + "qparams": { + "group_size": { + "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": 0.056267231702804565, + "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": 0.04887017607688904, + "qparams": { + "group_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": 0.04761030152440071, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04731340706348419, + "qparams": { + "group_size": { + "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": 0.028112728148698807, + "qparams": { + "group_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": 0.024295413866639137, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024211976677179337, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02222316525876522, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022032098844647408, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.014587941579520702, + "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": 0.014367789030075073, + "qparams": { + "group_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": 0.014163749292492867, + "qparams": { + "group_size": { + "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": 0.009119085967540741, + "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": 0.04887017607688904, + "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": 0.04887017607688904, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.23902730643749237, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20601825416088104, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19028373062610626, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.16866767406463623, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10907351225614548, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0944751650094986, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13741274178028107, + "qparams": { + "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": 0.1255195438861847, + "qparams": { + "group_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": 0.11491430550813675, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09164709597826004, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08728577196598053, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.07005054503679276, + "qparams": { + "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": 0.060461241751909256, + "qparams": { + "group_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": 0.05301694944500923, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.05112165957689285, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.035296741873025894, + "qparams": { + "group_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": 0.028292197734117508, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027596773579716682, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024933932349085808, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023699015378952026, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.019089579582214355, + "qparams": { + "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": 0.01942998543381691, + "qparams": { + "group_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": 0.0166791919618845, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013852816075086594, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05301694944500923, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05301694944500923, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.11632457375526428, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10871539264917374, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10560736060142517, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.09554188698530197, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05450642481446266, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.051443301141262054, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06209699809551239, + "qparams": { + "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": 0.05739590525627136, + "qparams": { + "group_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": 0.0553552582859993, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04881913959980011, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04635470360517502, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.031565140932798386, + "qparams": { + "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": 0.02746131457388401, + "qparams": { + "group_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": 0.02615450508892536, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.025840241461992264, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.015812376514077187, + "qparams": { + "group_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": 0.01346517726778984, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013334332033991814, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01230614259839058, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01210844237357378, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008296365849673748, + "qparams": { + "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": 0.008256747387349606, + "qparams": { + "group_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": 0.007873591035604477, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005421676207333803, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05450642481446266, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05450642481446266, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.097303606569767, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09098616987466812, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08825007826089859, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0798759013414383, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.045631472021341324, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04296761751174927, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05232999846339226, + "qparams": { + "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": 0.048320744186639786, + "qparams": { + "group_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": 0.04633355513215065, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04088874161243439, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03884571045637131, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.02659161388874054, + "qparams": { + "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": 0.023120233789086342, + "qparams": { + "group_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": 0.021894771605730057, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.021602561697363853, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.013315677642822266, + "qparams": { + "group_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": 0.011265466921031475, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.011137488298118114, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.010298103094100952, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010111449286341667, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0069824401289224625, + "qparams": { + "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": 0.006906646303832531, + "qparams": { + "group_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": 0.006586348172277212, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004501062445342541, + "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": 0.05232999846339226, + "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": 0.05232999846339226, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "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": 0.23166899383068085, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21701611578464508, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.21235837042331696, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19244815409183502, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.1090206652879715, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10364758223295212, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12186139822006226, + "qparams": { + "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": 0.1124558225274086, + "qparams": { + "group_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": 0.11061392724514008, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09774213284254074, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0926860049366951, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.062031567096710205, + "qparams": { + "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": 0.053708191961050034, + "qparams": { + "group_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": 0.05213341861963272, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.051770955324172974, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03096150793135166, + "qparams": { + "group_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": 0.026547331362962723, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.026432260870933533, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024204196408391, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023966729640960693, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.015994703397154808, + "qparams": { + "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": 0.015623378567397594, + "qparams": { + "group_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": 0.015423833392560482, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.009753920137882233, + "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": 0.053708191961050034, + "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": 0.053708191961050034, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.1268371194601059, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10906285047531128, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10147316753864288, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08199247717857361, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05840824171900749, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.050632014870643616, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07175596058368683, + "qparams": { + "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": 0.0655333623290062, + "qparams": { + "group_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": 0.060457248240709305, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04589106887578964, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0414704903960228, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03684043511748314, + "qparams": { + "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": 0.03239237517118454, + "qparams": { + "group_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": 0.02926461026072502, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.028485700488090515, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.018929410725831985, + "qparams": { + "group_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": 0.017048029229044914, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.0167566929012537, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.014974917285144329, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014520342461764812, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.011175052262842655, + "qparams": { + "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": 0.013037577271461487, + "qparams": { + "group_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": 0.010279898531734943, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010990195907652378, + "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": 0.04589106887578964, + "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": 0.04589106887578964, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.16509677469730377, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.15492938458919525, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1517820656299591, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13785837590694427, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07803893834352493, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07433126121759415, + "qparams": { + "group_size": { + "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": 0.0868610143661499, + "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": 0.08027161657810211, + "qparams": { + "group_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": 0.07917295396327972, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07014486938714981, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06660936772823334, + "qparams": { + "group_size": { + "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": 0.0444331131875515, + "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": 0.03854657709598541, + "qparams": { + "group_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": 0.037510160356760025, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03726529702544212, + "qparams": { + "group_size": { + "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": 0.022249558940529823, + "qparams": { + "group_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": 0.01947014592587948, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.019395878538489342, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01787531189620495, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017722776159644127, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.011833666823804379, + "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": 0.012004556134343147, + "qparams": { + "group_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": 0.011495253071188927, + "qparams": { + "group_size": { + "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": 0.00826193392276764, + "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": 0.0444331131875515, + "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": 0.0444331131875515, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.19481924176216125, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.18288414180278778, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1791849285364151, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1627713441848755, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.09199671447277069, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08763565868139267, + "qparams": { + "group_size": { + "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": 0.10243906080722809, + "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": 0.0945950374007225, + "qparams": { + "group_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": 0.09330187737941742, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0826895460486412, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07851604372262955, + "qparams": { + "group_size": { + "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": 0.05228288844227791, + "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": 0.045292291790246964, + "qparams": { + "group_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": 0.04410218447446823, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04382219538092613, + "qparams": { + "group_size": { + "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": 0.02614443004131317, + "qparams": { + "group_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": 0.02261471003293991, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.022530989721417427, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.02070402354001999, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02052231691777706, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.013732850551605225, + "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": 0.013531148433685303, + "qparams": { + "group_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": 0.013337541371583939, + "qparams": { + "group_size": { + "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": 0.008812898769974709, + "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": 0.05228288844227791, + "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": 0.05228288844227791, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.23073355853557587, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.19726991653442383, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.18249544501304626, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1617794930934906, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.10519197583198547, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09087247401475906, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.13184188306331635, + "qparams": { + "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": 0.11985337734222412, + "qparams": { + "group_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": 0.11100364476442337, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08770550787448883, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08353111892938614, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.06759659945964813, + "qparams": { + "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": 0.05792728811502457, + "qparams": { + "group_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": 0.051277291029691696, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04960354417562485, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03426530957221985, + "qparams": { + "group_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": 0.02761526219546795, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027043016627430916, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024289172142744064, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.023209311068058014, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.01907171681523323, + "qparams": { + "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": 0.019153546541929245, + "qparams": { + "group_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": 0.017011582851409912, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.014076866209506989, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051277291029691696, + "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": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051277291029691696, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.11390718072652817, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10664789378643036, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10363856703042984, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.0938592180609703, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05347590520977974, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05049048364162445, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.061222609132528305, + "qparams": { + "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": 0.05635339766740799, + "qparams": { + "group_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": 0.05427417904138565, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.047960519790649414, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.045692116022109985, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.031101495027542114, + "qparams": { + "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": 0.02697388455271721, + "qparams": { + "group_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": 0.025675809010863304, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.025364110246300697, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.015593478456139565, + "qparams": { + "group_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": 0.013216391205787659, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013083534315228462, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.012091172859072685, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01189351174980402, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.008201775141060352, + "qparams": { + "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": 0.008094086311757565, + "qparams": { + "group_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": 0.007788390852510929, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.005298457108438015, + "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": 0.05427417904138565, + "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": 0.05427417904138565, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "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": 0.09314120560884476, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08721049129962921, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.0845344215631485, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.07656361907720566, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04371613636612892, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.041166841983795166, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05027330294251442, + "qparams": { + "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": 0.0463990792632103, + "qparams": { + "group_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": 0.04436969757080078, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03920934349298477, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.037261296063661575, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.025538790971040726, + "qparams": { + "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": 0.022197332233190536, + "qparams": { + "group_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": 0.020990479737520218, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.020693551748991013, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.012800738215446472, + "qparams": { + "group_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": 0.010811842046678066, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.010680251754820347, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.009888979606330395, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009704572148621082, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006734426598995924, + "qparams": { + "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": 0.006657654419541359, + "qparams": { + "group_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": 0.006343858782202005, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.004351133015006781, + "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": 0.05027330294251442, + "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": 0.05027330294251442, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "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": 0.2370854914188385, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.22230878472328186, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2175922393798828, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.19744133949279785, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.11167850345373154, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10629180818796158, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.12455813586711884, + "qparams": { + "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": 0.11497578769922256, + "qparams": { + "group_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": 0.11327361315488815, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.10023508965969086, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09505995362997055, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0633767619729042, + "qparams": { + "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": 0.05494635924696922, + "qparams": { + "group_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": 0.0534273162484169, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.053077321499586105, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03162660449743271, + "qparams": { + "group_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": 0.0272025465965271, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.027095574885606766, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.024825798347592354, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.024590272456407547, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.016288310289382935, + "qparams": { + "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": 0.01600135676562786, + "qparams": { + "group_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": 0.01572839356958866, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.009992783889174461, + "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": 0.0534273162484169, + "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": 0.0534273162484169, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.13841873407363892, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.11570130288600922, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.10594229400157928, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08539952337741852, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.0626288503408432, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.054136086255311966, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.07903434336185455, + "qparams": { + "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": 0.07208842784166336, + "qparams": { + "group_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": 0.06644050776958466, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04849828779697418, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04425659775733948, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.040812231600284576, + "qparams": { + "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": 0.03525404632091522, + "qparams": { + "group_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": 0.03105543553829193, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.029991120100021362, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.020852604880928993, + "qparams": { + "group_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": 0.01751929335296154, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017171282321214676, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.015044543892145157, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01437782496213913, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.012063405476510525, + "qparams": { + "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": 0.013023202307522297, + "qparams": { + "group_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": 0.01086634024977684, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010429542511701584, + "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": 0.04849828779697418, + "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": 0.04849828779697418, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.1367710679769516, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1283247321844101, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.1256686896085739, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11433199793100357, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.06491903215646744, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06182669475674629, + "qparams": { + "group_size": { + "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": 0.0723060891032219, + "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": 0.06677402555942535, + "qparams": { + "group_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": 0.06585133820772171, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05841827020049095, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.055551450699567795, + "qparams": { + "group_size": { + "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": 0.03735007345676422, + "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": 0.03267006203532219, + "qparams": { + "group_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": 0.03183482214808464, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.031628891825675964, + "qparams": { + "group_size": { + "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": 0.01887856051325798, + "qparams": { + "group_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": 0.017550485208630562, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.017496950924396515, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01636020839214325, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.016245119273662567, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.010727289132773876, + "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": 0.012173784896731377, + "qparams": { + "group_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": 0.010471759364008904, + "qparams": { + "group_size": { + "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": 0.009830644354224205, + "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": 0.03735007345676422, + "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": 0.03735007345676422, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.15605248510837555, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.14653337001800537, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.14353758096694946, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.13047651946544647, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07413624972105026, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.07063597440719604, + "qparams": { + "group_size": { + "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": 0.08269276469945908, + "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": 0.0762183740735054, + "qparams": { + "group_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": 0.07515043020248413, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0666942447423935, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0636029914021492, + "qparams": { + "group_size": { + "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": 0.04268437623977661, + "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": 0.03729197382926941, + "qparams": { + "group_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": 0.03635038435459137, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.03612504154443741, + "qparams": { + "group_size": { + "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": 0.021637430414557457, + "qparams": { + "group_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": 0.020060783252120018, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.020003098994493484, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.018703226000070572, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.018569862470030785, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.012395600788295269, + "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": 0.013925235718488693, + "qparams": { + "group_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": 0.012117286212742329, + "qparams": { + "group_size": { + "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": 0.011255688965320587, + "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": 0.04268437623977661, + "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": 0.04268437623977661, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.17267166078090668, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.1460842788219452, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.13483066856861115, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.11910782009363174, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.07864741235971451, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06763291358947754, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.09968121349811554, + "qparams": { + "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": 0.08959450572729111, + "qparams": { + "group_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": 0.08290993422269821, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06519171595573425, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.06227472051978111, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.0512198768556118, + "qparams": { + "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": 0.04404333978891373, + "qparams": { + "group_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": 0.038997501134872437, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.0377948135137558, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.026433885097503662, + "qparams": { + "group_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": 0.022161215543746948, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.021716784685850143, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01978216879069805, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01898730918765068, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.015668250620365143, + "qparams": { + "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": 0.016510596498847008, + "qparams": { + "group_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": 0.014062440022826195, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.013298225589096546, + "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": 0.0512198768556118, + "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": 0.0512198768556118, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "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": 0.10647717118263245, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09962192922830582, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.09666207432746887, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.08760009706020355, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.04984170198440552, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0470077283680439, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.05749214068055153, + "qparams": { + "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": 0.05280415713787079, + "qparams": { + "group_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": 0.05062507838010788, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0447663851082325, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0427057147026062, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.029247144237160683, + "qparams": { + "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": 0.025274109095335007, + "qparams": { + "group_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": 0.023920098319649696, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02359386347234249, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.014655137434601784, + "qparams": { + "group_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": 0.012321601621806622, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012179904617369175, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.011285853572189808, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01108016911894083, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.007712588645517826, + "qparams": { + "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": 0.007580077741295099, + "qparams": { + "group_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": 0.007274019997566938, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0049592568539083, + "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": 0.05280415713787079, + "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": 0.05280415713787079, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "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": 0.08295907080173492, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07762238383293152, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07516022771596909, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.06803616136312485, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.03885145112872124, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03652311488986015, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.04489041492342949, + "qparams": { + "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": 0.04144199565052986, + "qparams": { + "group_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": 0.039443351328372955, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03486987575888634, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03315615653991699, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.022841984406113625, + "qparams": { + "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": 0.01984555460512638, + "qparams": { + "group_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": 0.018659112975001335, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.018368778750300407, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.011449846439063549, + "qparams": { + "group_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": 0.009639938361942768, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009513136930763721, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.008835289627313614, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.008653643541038036, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.006052718032151461, + "qparams": { + "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": 0.006003360729664564, + "qparams": { + "group_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": 0.005671882536262274, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.003976107109338045, + "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": 0.04489041492342949, + "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": 0.04489041492342949, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "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": 0.21998122334480286, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.20621077716350555, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.2018376588821411, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.18319061398506165, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.103658027946949, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09858453273773193, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.11742367595434189, + "qparams": { + "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": 0.10693882405757904, + "qparams": { + "group_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": 0.10518189519643784, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.09316474199295044, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.08887773752212524, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.059924401342868805, + "qparams": { + "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": 0.051406633108854294, + "qparams": { + "group_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": 0.04980707913637161, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.04939712584018707, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.03035220503807068, + "qparams": { + "group_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": 0.02564351260662079, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.025504928082227707, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.023500952869653702, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.02324090525507927, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0159232746809721, + "qparams": { + "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": 0.01549350656569004, + "qparams": { + "group_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": 0.015278337523341179, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.010112738236784935, + "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": 0.051406633108854294, + "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": 0.051406633108854294, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "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": 0.1187150627374649, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.08899453282356262, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.07796429097652435, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.06847966462373734, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05282076820731163, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04054154455661774, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.06876188516616821, + "qparams": { + "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": 0.06001362204551697, + "qparams": { + "group_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": 0.05700906366109848, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.037307292222976685, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.03707827255129814, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.03580031171441078, + "qparams": { + "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": 0.030052850022912025, + "qparams": { + "group_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": 0.026788000017404556, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.025982504710555077, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.018805181607604027, + "qparams": { + "group_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": 0.015865182504057884, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.015686243772506714, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.01339490246027708, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012904570437967777, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.011506873182952404, + "qparams": { + "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": 0.012380117550492287, + "qparams": { + "group_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": 0.0105619290843606, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.01052588690072298, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05282076820731163, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05282076820731163, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "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": 0.12078049033880234, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.11327625066041946, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.11093422770500183, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.10083207488059998, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.05686429142951965, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.0541689470410347, + "qparams": { + "group_size": { + "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": 0.06344915926456451, + "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": 0.05849582329392433, + "qparams": { + "group_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": 0.05769415944814682, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05113987997174263, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04863195866346359, + "qparams": { + "group_size": { + "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": 0.03235001862049103, + "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": 0.028047453612089157, + "qparams": { + "group_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": 0.027309291064739227, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.027132907882332802, + "qparams": { + "group_size": { + "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": 0.01620340906083584, + "qparams": { + "group_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": 0.014142408035695553, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.014094793237745762, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.012987232767045498, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.012876327149569988, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.008574196137487888, + "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": 0.008682365529239178, + "qparams": { + "group_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": 0.008325297385454178, + "qparams": { + "group_size": { + "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": 0.00593169778585434, + "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": 0.05113987997174263, + "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": 0.05113987997174263, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.12090599536895752, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.11344221234321594, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "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": 0.11107149720191956, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.1009664312005043, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.057304322719573975, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.054614145308732986, + "qparams": { + "group_size": { + "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": 0.06400565803050995, + "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": 0.058934468775987625, + "qparams": { + "group_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": 0.05813908949494362, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.05158195272088051, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.04918115213513374, + "qparams": { + "group_size": { + "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": 0.03303203731775284, + "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": 0.028667347505688667, + "qparams": { + "group_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": 0.027930378913879395, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.02776370383799076, + "qparams": { + "group_size": { + "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": 0.016685638576745987, + "qparams": { + "group_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": 0.015149732120335102, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.01510312408208847, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.014062019996345043, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.013965530321002007, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 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": 0.009375439956784248, + "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": 0.010206717997789383, + "qparams": { + "group_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": 0.009139719419181347, + "qparams": { + "group_size": { + "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": 0.007983655668795109, + "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": 0.05158195272088051, + "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": 0.05158195272088051, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "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": 0.07108021527528763, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.059651851654052734, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "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": 0.053933389484882355, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_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": 0.04752644896507263, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 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": 0.03246389701962471, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.027235122397542, + "qparams": { + "group_size": { + "6": 32, + "3": 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": 0.04336948320269585, + "qparams": { + "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": 0.038284026086330414, + "qparams": { + "group_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": 0.03429454192519188, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.026852024719119072, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "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": 0.025827206671237946, + "qparams": { + "group_size": { + "4": 64, + "3": 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": 0.022361746057868004, + "qparams": { + "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": 0.01915614679455757, + "qparams": { + "group_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": 0.016562992706894875, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "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": 0.01587746851146221, + "qparams": { + "group_size": { + "6": 32, + "4": 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": 0.011678855866193771, + "qparams": { + "group_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": 0.009994877502322197, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.009750327095389366, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 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": 0.009096876718103886, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_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": 0.008714219555258751, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 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": 0.0070499517023563385, + "qparams": { + "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": 0.00803468283265829, + "qparams": { + "group_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": 0.006260349880903959, + "qparams": { + "group_size": { + "8": 128, + "6": 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": 0.0068732029758393764, + "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": 0.053933389484882355, + "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": 0.053933389484882355, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + } + } + ], + "last_module_idx": 82, + "base_perplexity": 8.616362429400148 +} \ No newline at end of file