support vllm

config.json

@@ -3,11 +3,11 @@
     "Ernie4_5_VLMoeForConditionalGeneration"
   ],
   "auto_map": {
-    "AutoConfig": "configuration_ernie_45t_vl.Ernie4_5_VLMoEConfig",
-    "AutoModel": "modeling_ernie_45t_vl.Ernie4_5_VLMoeForConditionalGeneration",
-    "AutoModelForCausalLM": "modeling_ernie_45t_vl.Ernie4_5_VLMoeForConditionalGeneration",
-    "AutoProcessor": "processing_ernie_45t_vl.Ernie_45T_VLProcessor",
-    "AutoImageProcessor": "processing_ernie_45t_vl.Ernie_45T_VLImageProcessor"
+    "AutoConfig": "configuration_ernie4_5_vl.Ernie4_5_VLMoEConfig",
+    "AutoModel": "modeling_ernie4_5_vl.Ernie4_5_VLMoeForConditionalGeneration",
+    "AutoModelForCausalLM": "modeling_ernie4_5_vl.Ernie4_5_VLMoeForConditionalGeneration",
+    "AutoProcessor": "processing_ernie4_5_vl.Ernie4_5_VLProcessor",
+    "AutoImageProcessor": "processing_ernie4_5_vl.Ernie4_5_VLImageProcessor"
   },
   "pad_token_id": 0,
   "bos_token_id": 1,
@@ -17,6 +17,8 @@
   "hidden_size": 2560,
   "intermediate_size": 12288,
   "im_patch_id": 100295,
+  "video_start_token_id": 101306,
+  "video_end_token_id": 101307,
   "max_position_embeddings": 131072,
   "model_type": "ernie4_5_moe_vl",
   "moe_capacity": [128, 128, 128],
@@ -44,6 +46,14 @@
   "use_cache": true,
   "use_rmsnorm": true,
   "use_bias": false,
+  "rope_scaling": {
+    "type": "default",
+    "mrope_section": [
+      22,
+      22,
+      20
+    ]
+  },
   "vision_config": {
     "attn_implementation": "eager",
     "depth": 32,