Upload Phi3ForCausalLM

config.json

@@ -29,32 +29,82 @@
   "pad_token_id": 199999,
   "partial_rotary_factor": 0.75,
   "quantization_config": {
-    "modules_to_not_convert": null,
+    "include_embedding": true,
+    "modules_to_not_convert": [],
     "quant_method": "torchao",
     "quant_type": {
       "default": {
         "_data": {
-          "act_mapping_type": {
-            "_data": "ASYMMETRIC",
-            "_type": "MappingType"
-          },
-          "group_size": 32,
-          "layout": {
-            "_data": {},
-            "_type": "PlainLayout",
-            "_version": 1
-          },
-          "mapping_type": {
-            "_data": "SYMMETRIC",
-            "_type": "MappingType"
-          },
-          "set_inductor_config": true
+          "module_fqn_to_config": {
+            "_default": {
+              "_data": {
+                "act_mapping_type": {
+                  "_data": "ASYMMETRIC",
+                  "_type": "MappingType"
+                },
+                "layout": {
+                  "_data": {},
+                  "_type": "QDQLayout",
+                  "_version": 1
+                },
+                "weight_dtype": {
+                  "_data": "int4",
+                  "_type": "torch.dtype"
+                },
+                "weight_granularity": {
+                  "_data": {
+                    "group_size": 32
+                  },
+                  "_type": "PerGroup",
+                  "_version": 1
+                },
+                "weight_mapping_type": {
+                  "_data": "SYMMETRIC",
+                  "_type": "MappingType"
+                },
+                "weight_scale_dtype": {
+                  "_data": "bfloat16",
+                  "_type": "torch.dtype"
+                }
+              },
+              "_type": "Int8DynamicActivationIntxWeightConfig",
+              "_version": 1
+            },
+            "model.embed_tokens": {
+              "_data": {
+                "granularity": {
+                  "_data": {
+                    "axis": 0
+                  },
+                  "_type": "PerAxis",
+                  "_version": 1
+                },
+                "layout": {
+                  "_data": {},
+                  "_type": "QDQLayout",
+                  "_version": 1
+                },
+                "mapping_type": {
+                  "_data": "SYMMETRIC",
+                  "_type": "MappingType"
+                },
+                "scale_dtype": null,
+                "weight_dtype": {
+                  "_data": "int8",
+                  "_type": "torch.dtype"
+                }
+              },
+              "_type": "IntxWeightOnlyConfig",
+              "_version": 1
+            }
+          }
         },
-        "_type": "Int8DynamicActivationInt4WeightConfig",
+        "_type": "AOPerModuleConfig",
         "_version": 1
       }
     },
-    "quant_type_kwargs": {}
+    "quant_type_kwargs": {},
+    "untie_embedding_weights": true
   },
   "resid_pdrop": 0.0,
   "rms_norm_eps": 1e-05,
@@ -163,8 +213,8 @@
   },
   "rope_theta": 10000.0,
   "sliding_window": 262144,
-  "tie_word_embeddings": true,
-  "torch_dtype": "bfloat16",
+  "tie_word_embeddings": false,
+  "torch_dtype": "float32",
   "transformers_version": "4.52.0.dev0",
   "use_cache": true,
   "vocab_size": 200064

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:968c41e72dfc2fb39ed55f2a85e207979a062db7064c601004fabf647c2614c6
+size 4812014346