Upload patch_vllm.py with huggingface_hub

patch_vllm.py

@@ -24,15 +24,11 @@ if '"model.layers.": "language_model.model.layers."' not in c:
 else:
     print("kimi_k25.py already patched")
 
-# Patch 3: deepseek_mtp.py - add text_config extraction everywhere config is read
+# Patch 3: deepseek_mtp.py - extract text_config
 mtp_path = "/usr/local/lib/python3.12/dist-packages/vllm/model_executor/models/deepseek_mtp.py"
 with open(mtp_path) as f:
     c = f.read()
-
-# Replace ALL occurrences of "vllm_config.model_config.hf_config" in this file
-# to add text_config fallback
 if "text_config" not in c:
-    # Strategy: add a helper function at the top, then use it
     import_marker = "from .utils import maybe_prefix"
     helper = '''from .utils import maybe_prefix
 
@@ -40,8 +36,6 @@ def _get_text_config(hf_config):
     """Extract text_config from VLM configs (e.g. KimiK25Config)."""
     return getattr(hf_config, 'text_config', hf_config)'''
     c = c.replace(import_marker, helper)
-    
-    # Replace all config reads
     c = c.replace(
         "config = vllm_config.model_config.hf_config\n        self.mtp_start_layer_idx",
         "config = _get_text_config(vllm_config.model_config.hf_config)\n        self.mtp_start_layer_idx")
@@ -51,8 +45,20 @@ def _get_text_config(hf_config):
     c = c.replace(
         "config = vllm_config.speculative_config.draft_model_config.hf_config\n        self.config = config",
         "config = _get_text_config(vllm_config.speculative_config.draft_model_config.hf_config)\n        self.config = config")
-    
     with open(mtp_path, "w") as f: f.write(c)
-    print("deepseek_mtp.py PATCHED (all config reads)")
+    print("deepseek_mtp.py PATCHED")
 else:
     print("deepseek_mtp.py already patched")
+
+# Patch 4: eagle.py - handle KimiK25Config missing image_token_index
+eagle_path = "/usr/local/lib/python3.12/dist-packages/vllm/v1/spec_decode/eagle.py"
+with open(eagle_path) as f:
+    c = f.read()
+old_eagle = "                self.model.config.image_token_index = (\n                    target_model.config.image_token_index\n                )"
+new_eagle = "                self.model.config.image_token_index = getattr(\n                    target_model.config, 'image_token_index',\n                    getattr(target_model.config, 'media_placeholder_token_id', 0)\n                )"
+if "media_placeholder_token_id" not in c:
+    c = c.replace(old_eagle, new_eagle)
+    with open(eagle_path, "w") as f: f.write(c)
+    print("eagle.py PATCHED (KimiK25 image_token_index fallback)")
+else:
+    print("eagle.py already patched")