Upload patch_models.py

patch_models.py

@@ -102,9 +102,9 @@ new_func = '''def _available_models() -> list[dict[str, Any]]:
             "tier": "free",
         },
         {
-            "id": "deepseek/deepseek-v4-flash",
+            "id": "openai/deepseek/deepseek-v4-flash:free",
             "label": "DeepSeek V4 Flash (OR)",
-            "provider": "huggingface",
+            "provider": "openrouter",
             "tier": "free",
         },
         {
@@ -122,14 +122,14 @@ new_func = '''def _available_models() -> list[dict[str, Any]]:
             "recommended": True,
         },
         {
-            "id": "tencent/hy3-preview",
+            "id": "openai/tencent/hy3-preview:free",
             "label": "Hy3 Preview",
             "provider": "openrouter",
             "tier": "free",
             "recommended": True,
         },
         {
-            "id": "cohere/north-mini-code:free",
+            "id": "openai/cohere/north-mini-code:free",
             "label": "Riverflow V2.5 Pro",
             "provider": "openrouter",
             "tier": "free",
@@ -182,11 +182,11 @@ else:
             '    {"id": "Qwen/Qwen3-Coder-Next", "label": "Qwen3 Coder Next", "provider": "huggingface", "tier": "free", "recommended": True},\n'
             '    {"id": "google/gemma-3-1b-it", "label": "Gemma 3 1B", "provider": "huggingface", "tier": "free", "recommended": True},\n'
             '    {"id": "openai/google/gemini-2.0-flash-001", "label": "Gemini 2.0 Flash", "provider": "openrouter", "tier": "free"},\n'
-            '    {"id": "deepseek/deepseek-v4-flash", "label": "DeepSeek V4 Flash (OR)", "provider": "huggingface", "tier": "free"},\n'
+            '    {"id": "openai/deepseek/deepseek-v4-flash:free", "label": "DeepSeek V4 Flash (OR)", "provider": "openrouter", "tier": "free"},\n'
             '    {"id": "openai/google/gemma-4-9b-it:free", "label": "Gemma 4 9B", "provider": "openrouter", "tier": "free", "recommended": True},\n'
             '    {"id": "openai/google/gemma-4-31b-it:free", "label": "Gemma 4 31B", "provider": "openrouter", "tier": "free", "recommended": True},\n'
-            '    {"id": "tencent/hy3-preview", "label": "Hy3 Preview", "provider": "openrouter", "tier": "free", "recommended": True},\n'
-            '    {"id": "cohere/north-mini-code:free", "label": "Riverflow V2.5 Pro", "provider": "openrouter", "tier": "free", "recommended": True},\n'
+            '    {"id": "openai/tencent/hy3-preview:free", "label": "Hy3 Preview", "provider": "openrouter", "tier": "free", "recommended": True},\n'
+            '    {"id": "openai/cohere/north-mini-code:free", "label": "Riverflow V2.5 Pro", "provider": "openrouter", "tier": "free", "recommended": True},\n'
             '    {"id": "openai/nvidia/nemotron-3-super-120b-a12b:free", "label": "Nemotron 3 Super 120B", "provider": "openrouter", "tier": "free"},\n'
             '    {"id": "openai/nvidia/nemotron-3-ultra-550b-a55b:free", "label": "Nemotron 3 Ultra 550B", "provider": "openrouter", "tier": "free", "recommended": True},\n'
             '    {"id": "openai/poolside/laguna-m.1:free", "label": "Laguna M.1", "provider": "openrouter", "tier": "free", "recommended": True},\n'
@@ -216,6 +216,7 @@ with open(AGENT_FILE, "w") as f:
 with open(LLM_PARAMS_FILE) as f:
     llm_content = f.read()
 
+# First, try to replace the full block (original version)
 old_block = """    hf_model = normalized_model
     api_key = _resolve_hf_router_token(session_hf_token)
     params = {
@@ -230,7 +231,7 @@ new_block = """    hf_model = normalized_model
     # === PATCH: Route models to OpenRouter via LiteLLM ===
     # Models with openai/<provider>/ prefix go to OpenRouter.
     # Keep the "openai/" prefix — LiteLLM needs it for provider detection.
-    _or_prefixes = ("openai/google/", "openai/deepseek/", "openai/nvidia/", "openai/poolside/", "openai/sourceful/", "openai/openrouter/")
+    _or_prefixes = ("openai/google/", "openai/deepseek/", "openai/nvidia/", "openai/poolside/", "openai/sourceful/", "openai/tencent/", "openai/cohere/", "openai/openrouter/")
     if normalized_model.startswith(_or_prefixes):
         return {
             "model": normalized_model,
@@ -239,32 +240,76 @@ new_block = """    hf_model = normalized_model
         }
     # === END PATCH ===
 
-    params = {
-        "model": f"openai/{hf_model}",
-        "api_base": HF_ROUTER_BASE_URL,
-        "api_key": api_key,
-    }"""
+    # Don't double-prefix: if model already starts with openai/, use as-is for HF Router
+    if normalized_model.startswith("openai/"):
+        params = {
+            "model": normalized_model,
+            "api_base": HF_ROUTER_BASE_URL,
+            "api_key": api_key,
+        }
+    else:
+        params = {
+            "model": f"openai/{hf_model}",
+            "api_base": HF_ROUTER_BASE_URL,
+            "api_key": api_key,
+        }"""
 
 if old_block in llm_content:
     llm_content = llm_content.replace(old_block, new_block)
-    print("OK: Patched _resolve_llm_params")
-elif "openai/openrouter/" in llm_content:
-    print("OK: Already patched")
+    print("OK: Patched _resolve_llm_params (full block replace)")
+elif "openai/tencent/" in llm_content:
+    print("OK: Already patched with tencent/cohere prefixes")
 else:
-    idx = llm_content.find("hf_model = normalized_model")
+    # Try updating just the _or_prefixes tuple if already patched
+    # First check if already patched
+    idx = llm_content.find('_or_prefixes = (')
     if idx >= 0:
-        print(f"WARN: Target block not found at idx {idx}")
-    import sys
-    sys.exit(1)
+        # Already has the OR patch, need to update prefixes
+        old_prefixes = llm_content[idx:llm_content.find(')', idx)+1]
+        new_prefixes = '_or_prefixes = ("openai/google/", "openai/deepseek/", "openai/nvidia/", "openai/poolside/", "openai/sourceful/", "openai/tencent/", "openai/cohere/", "openai/openrouter/")'
+        if old_prefixes != new_prefixes:
+            llm_content = llm_content.replace(old_prefixes, new_prefixes)
+            print("OK: Updated _or_prefixes to include tencent + cohere")
+        else:
+            print("OK: _or_prefixes already up to date")
+        
+        # Also fix the HF Router routing to handle openai/ prefix properly
+        old_hf_route = '''    params = {
+        "model": f"openai/{hf_model}",
+        "api_base": HF_ROUTER_BASE_URL,
+        "api_key": api_key,
+    }'''
+        new_hf_route = '''    # Don't double-prefix: if model already starts with openai/, use as-is for HF Router
+    if normalized_model.startswith("openai/"):
+        params = {
+            "model": normalized_model,
+            "api_base": HF_ROUTER_BASE_URL,
+            "api_key": api_key,
+        }
+    else:
+        params = {
+            "model": f"openai/{hf_model}",
+            "api_base": HF_ROUTER_BASE_URL,
+            "api_key": api_key,
+        }'''
+        if old_hf_route in llm_content:
+            llm_content = llm_content.replace(old_hf_route, new_hf_route)
+            print("OK: Fixed HF Router routing to avoid double openai/ prefix")
+        else:
+            print("OK: HF Router routing already handles openai/ prefix")
+    else:
+        print("FAIL: Could not find _or_prefixes or original block!")
+        import sys
+        sys.exit(1)
 
 try:
     ast.parse(llm_content)
     print("OK: llm_params.py syntax OK")
 except SyntaxError as e:
-    print(f"FAIL: Syntax error: {e}")
+    print(f"FAIL: Syntax error in llm_params.py: {e}")
     raise
 
 with open(LLM_PARAMS_FILE, "w") as f:
     f.write(llm_content)
 
-print("OK: Backend patched - 16 models, owl-alpha default, OpenRouter routing (keep openai/ prefix)")
+print("OK: Backend patched - 16 models, OR routing with tencent/cohere, HF Router fix for openai/ prefix")