update

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d5eaab4c2f3b7c01a1cb63ac17ebc6e46451ec8c02cda8883bb336efd9ea977
-size 3358077296
+oid sha256:272dd7c6f22b5f01672cd8a4c062232f9ed95df64a49c7fbced931540cd30f66
+size 3358077336

quant_strategy.json

@@ -1,8 +1,8 @@
 {
     "measurement": {
         "model.layers.0": {
-            "accuracy": 0.9288430213928223,
-            "total_bits": 365124672,
+            "accuracy": 0.9242749214172363,
+            "total_bits": 360997440,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -17,10 +17,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -89,8 +89,8 @@
             }
         },
         "model.layers.1": {
-            "accuracy": 0.9289391040802002,
-            "total_bits": 449761824,
+            "accuracy": 0.9216856956481934,
+            "total_bits": 486917664,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -105,10 +105,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -129,10 +129,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -177,14 +177,14 @@
             }
         },
         "model.layers.2": {
-            "accuracy": 0.936192512512207,
-            "total_bits": 519946272,
+            "accuracy": 0.8546795845031738,
+            "total_bits": 360997440,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -193,10 +193,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -217,10 +217,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -253,10 +253,10 @@
             },
             "down_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -265,14 +265,14 @@
             }
         },
         "model.layers.3": {
-            "accuracy": 0.9241403341293335,
-            "total_bits": 515819040,
+            "accuracy": 0.9084997177124023,
+            "total_bits": 449761824,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -305,10 +305,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -353,14 +353,14 @@
             }
         },
         "model.layers.4": {
-            "accuracy": 0.924299955368042,
-            "total_bits": 519946272,
+            "accuracy": 0.8643641471862793,
+            "total_bits": 394026048,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -369,10 +369,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -429,10 +429,10 @@
             },
             "down_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -441,14 +441,14 @@
             }
         },
         "model.layers.5": {
-            "accuracy": 0.9709324836730957,
-            "total_bits": 697480224,
+            "accuracy": 0.8657441139221191,
+            "total_bits": 365124672,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -481,10 +481,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -493,10 +493,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -505,10 +505,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -517,10 +517,10 @@
             },
             "down_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -529,14 +529,14 @@
             }
         },
         "model.layers.6": {
-            "accuracy": 0.9690049886703491,
-            "total_bits": 697480224,
+            "accuracy": 0.877474308013916,
+            "total_bits": 449761824,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -545,10 +545,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -569,10 +569,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -581,10 +581,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -593,10 +593,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -617,14 +617,14 @@
             }
         },
         "model.layers.7": {
-            "accuracy": 0.9691200256347656,
-            "total_bits": 697480224,
+            "accuracy": 0.8887453079223633,
+            "total_bits": 482790432,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -633,10 +633,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -669,10 +669,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -681,10 +681,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -705,8 +705,8 @@
             }
         },
         "model.layers.8": {
-            "accuracy": 0.9361705780029297,
-            "total_bits": 486917664,
+            "accuracy": 0.9228010177612305,
+            "total_bits": 449761824,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -721,10 +721,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -745,10 +745,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -793,14 +793,14 @@
             }
         },
         "model.layers.9": {
-            "accuracy": 0.959143877029419,
-            "total_bits": 394026048,
+            "accuracy": 0.9577234387397766,
+            "total_bits": 365124672,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -809,10 +809,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -881,8 +881,8 @@
             }
         },
         "model.layers.10": {
-            "accuracy": 0.9522266387939453,
-            "total_bits": 394026048,
+            "accuracy": 0.9458887577056885,
+            "total_bits": 360997440,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -921,10 +921,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -969,14 +969,14 @@
             }
         },
         "model.layers.11": {
-            "accuracy": 0.9406924247741699,
-            "total_bits": 394026048,
+            "accuracy": 0.9322950839996338,
+            "total_bits": 360997440,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1057,8 +1057,8 @@
             }
         },
         "model.layers.12": {
-            "accuracy": 0.9434242248535156,
-            "total_bits": 394026048,
+            "accuracy": 0.9404451847076416,
+            "total_bits": 365124672,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -1073,10 +1073,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1097,10 +1097,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1145,8 +1145,8 @@
             }
         },
         "model.layers.13": {
-            "accuracy": 0.9376596212387085,
-            "total_bits": 394026048,
+            "accuracy": 0.9363645315170288,
+            "total_bits": 365124672,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -1161,10 +1161,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1185,10 +1185,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1233,8 +1233,8 @@
             }
         },
         "model.layers.14": {
-            "accuracy": 0.9380558729171753,
-            "total_bits": 394026048,
+            "accuracy": 0.9359749555587769,
+            "total_bits": 365124672,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -1249,10 +1249,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1273,10 +1273,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1321,8 +1321,8 @@
             }
         },
         "model.layers.15": {
-            "accuracy": 0.9555038213729858,
-            "total_bits": 486917664,
+            "accuracy": 0.9322938919067383,
+            "total_bits": 365124672,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -1361,10 +1361,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1397,10 +1397,10 @@
             },
             "down_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1409,8 +1409,8 @@
             }
         },
         "model.layers.16": {
-            "accuracy": 0.9523999691009521,
-            "total_bits": 486917664,
+            "accuracy": 0.939303994178772,
+            "total_bits": 449761824,
             "q_proj": {
                 "group_size": {
                     "2": 64
@@ -1425,10 +1425,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1449,10 +1449,10 @@
             },
             "o_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1497,14 +1497,14 @@
             }
         },
         "model.layers.17": {
-            "accuracy": 0.9459496736526489,
-            "total_bits": 515819040,
+            "accuracy": 0.9451323747634888,
+            "total_bits": 486917664,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1513,10 +1513,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1585,8 +1585,8 @@
             }
         },
         "model.layers.18": {
-            "accuracy": 0.9424376487731934,
-            "total_bits": 515819040,
+            "accuracy": 0.9493275880813599,
+            "total_bits": 519946272,
             "q_proj": {
                 "group_size": {
                     "4": 128
@@ -1601,10 +1601,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1673,14 +1673,14 @@
             }
         },
         "model.layers.19": {
-            "accuracy": 0.9421288967132568,
-            "total_bits": 515819040,
+            "accuracy": 0.9514966011047363,
+            "total_bits": 571557408,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1725,10 +1725,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1761,8 +1761,8 @@
             }
         },
         "model.layers.20": {
-            "accuracy": 0.9422779083251953,
-            "total_bits": 519946272,
+            "accuracy": 0.955375075340271,
+            "total_bits": 604586016,
             "q_proj": {
                 "group_size": {
                     "4": 128
@@ -1777,10 +1777,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1813,10 +1813,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1849,14 +1849,14 @@
             }
         },
         "model.layers.21": {
-            "accuracy": 0.9373316764831543,
-            "total_bits": 519946272,
+            "accuracy": 0.9731628894805908,
+            "total_bits": 664451616,
             "q_proj": {
                 "group_size": {
-                    "4": 128
+                    "2": 64
                 },
                 "bits": [
-                    4
+                    2
                 ],
                 "bits_prop": [
                     1
@@ -1901,10 +1901,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1913,10 +1913,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -1937,8 +1937,8 @@
             }
         },
         "model.layers.22": {
-            "accuracy": 0.9449453353881836,
-            "total_bits": 604586016,
+            "accuracy": 0.9785275459289551,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
                     "4": 128
@@ -1953,10 +1953,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2001,10 +2001,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2025,8 +2025,8 @@
             }
         },
         "model.layers.23": {
-            "accuracy": 0.9457166194915771,
-            "total_bits": 604586016,
+            "accuracy": 0.9788622856140137,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
                     "4": 128
@@ -2041,10 +2041,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2089,10 +2089,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2113,8 +2113,8 @@
             }
         },
         "model.layers.24": {
-            "accuracy": 0.9458975791931152,
-            "total_bits": 604586016,
+            "accuracy": 0.9794007539749146,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
                     "4": 128
@@ -2129,10 +2129,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2177,10 +2177,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2201,8 +2201,8 @@
             }
         },
         "model.layers.25": {
-            "accuracy": 0.9496898651123047,
-            "total_bits": 604586016,
+            "accuracy": 0.9806145429611206,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
                     "4": 128
@@ -2217,10 +2217,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2265,10 +2265,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2289,14 +2289,14 @@
             }
         },
         "model.layers.26": {
-            "accuracy": 0.9475545883178711,
-            "total_bits": 571557408,
+            "accuracy": 0.9806764125823975,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2305,10 +2305,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2341,10 +2341,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2377,14 +2377,14 @@
             }
         },
         "model.layers.27": {
-            "accuracy": 0.9472908973693848,
-            "total_bits": 571557408,
+            "accuracy": 0.9815640449523926,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2393,10 +2393,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2441,10 +2441,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2465,8 +2465,8 @@
             }
         },
         "model.layers.28": {
-            "accuracy": 0.9392316341400146,
-            "total_bits": 519946272,
+            "accuracy": 0.9820178747177124,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
                     "4": 128
@@ -2517,10 +2517,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2529,10 +2529,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2553,14 +2553,14 @@
             }
         },
         "model.layers.29": {
-            "accuracy": 0.9373917579650879,
-            "total_bits": 482790432,
+            "accuracy": 0.9836413264274597,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2569,10 +2569,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2605,10 +2605,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2617,10 +2617,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2641,14 +2641,14 @@
             }
         },
         "model.layers.30": {
-            "accuracy": 0.9403772354125977,
-            "total_bits": 482790432,
+            "accuracy": 0.9838729500770569,
+            "total_bits": 697480224,
             "q_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2657,10 +2657,10 @@
             },
             "k_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2693,10 +2693,10 @@
             },
             "up_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1
@@ -2705,10 +2705,10 @@
             },
             "gate_proj": {
                 "group_size": {
-                    "2": 64
+                    "4": 128
                 },
                 "bits": [
-                    2
+                    4
                 ],
                 "bits_prop": [
                     1