Update config.json

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/home/azureuser/phi4/phi4-works/phi4-mini/output_bn_new-13/checkpoint-4000",
+  "_name_or_path": "Phi-4-multimodal-instruct",
   "architectures": [
     "Phi4MMForCausalLM"
   ],
@@ -50,7 +50,7 @@
   "auto_map": {
     "AutoConfig": "configuration_phi4mm.Phi4MMConfig",
     "AutoModelForCausalLM": "modeling_phi4mm.Phi4MMForCausalLM",
-    "AutoTokenizer": "microsoft/Phi-4-multimodal-instruct--Xenova/gpt-4o"
+    "AutoTokenizer": "Xenova/gpt-4o"
   },
   "bos_token_id": 199999,
   "embd_layer": {
@@ -80,11 +80,22 @@
   "full_attn_mod": 1,
   "hidden_act": "silu",
   "hidden_size": 3072,
-  "img_processor": null,
   "initializer_range": 0.02,
   "intermediate_size": 8192,
   "interpolate_factor": 1,
   "lm_head_bias": false,
+  "vision_lora": {
+    "dp": 0.0,
+    "layer": "layers.*((self_attn\\.(qkv_proj|o_proj))|(mlp\\.(gate_up|down)_proj))",
+    "lora_alpha": 512,
+    "r": 256
+  },
+  "speech_lora": {
+    "dp": 0.01,
+    "layer": "((layers.*self_attn\\.(qkv|o)_proj)|(layers.*mlp\\.(gate_up|down)_proj))",
+    "lora_alpha": 640,
+    "r": 320
+  },
   "max_position_embeddings": 131072,
   "mlp_bias": false,
   "model_type": "phi4mm",
@@ -201,21 +212,10 @@
   },
   "rope_theta": 10000.0,
   "sliding_window": 262144,
-  "speech_lora": {
-    "dp": 0.01,
-    "layer": "((layers.*self_attn\\.(qkv|o)_proj)|(layers.*mlp\\.(gate_up|down)_proj))",
-    "lora_alpha": 640,
-    "r": 320
-  },
   "tie_word_embeddings": true,
   "torch_dtype": "bfloat16",
-  "transformers_version": "4.48.3",
+  "transformers_version": "4.46.1",
   "use_cache": true,
-  "vision_lora": {
-    "dp": 0.0,
-    "layer": "layers.*((self_attn\\.(qkv_proj|o_proj))|(mlp\\.(gate_up|down)_proj))",
-    "lora_alpha": 512,
-    "r": 256
-  },
-  "vocab_size": 200064
-}
+  "vocab_size": 200064,
+  "_attn_implementation": "flash_attention_2"
+}