Upload patch_models.py

patch_models.py

@@ -15,10 +15,10 @@ AVAILABLE_MODELS.extend([
     {"id": "deepseek-ai/DeepSeek-V4-Pro", "label": "DeepSeek V4 Pro", "provider": "huggingface", "tier": "free", "recommended": True},
     {"id": "deepseek-ai/DeepSeek-V4-Flash", "label": "DeepSeek V4 Flash", "provider": "huggingface", "tier": "free"},
     {"id": "openai/deepseek/deepseek-v4-flash", "label": "DeepSeek V4 Flash (OR)", "provider": "openrouter", "tier": "free"},
-    {"id": "nvidia/nemotron-3-super-120b-a12b", "label": "Nemotron 3 Super 120B", "provider": "openrouter", "tier": "free"},
+    {"id": "openai/nvidia/nemotron-3-super-120b-a12b", "label": "Nemotron 3 Super 120B", "provider": "openrouter", "tier": "free"},
     {"id": "google/gemma-3-1b-it", "label": "Gemma 3 1B", "provider": "huggingface", "tier": "free", "recommended": True},
     {"id": "Qwen/Qwen3-Coder-Next", "label": "Qwen3 Coder Next", "provider": "huggingface", "tier": "free", "recommended": True},
-    {"id": "owl-alpha", "label": "Owl Alpha", "provider": "openrouter", "tier": "free", "recommended": True},
+    {"id": "openai/owl-alpha", "label": "Owl Alpha", "provider": "openrouter", "tier": "free", "recommended": True},
     {"id": "openai/google/gemini-2.0-flash-001", "label": "Gemini 2.0 Flash", "provider": "openrouter", "tier": "free"},
     {"id": "openai/google/gemma-4-31b-it", "label": "Gemma 4 31B", "provider": "openrouter", "tier": "free", "recommended": True},
     {"id": "openai/nvidia/nemotron-3-ultra-550b-a55b", "label": "Nemotron 3 Ultra 550B", "provider": "openrouter", "tier": "free", "recommended": True},
@@ -49,24 +49,6 @@ with open(AGENT_FILE, "w") as f:
 with open(LLM_PARAMS_FILE) as f:
     llm_content = f.read()
 
-# Add OpenRouter routing before the final HF Router fallback.
-# Models starting with "openai/" go to OpenRouter API, NOT HF Router.
-# Without this patch, _resolve_llm_params would double-prefix them
-# (openai/openai/...) and send them to router.huggingface.co which
-# doesn't serve OpenRouter-only models like Laguna/Mistral/Nvidia free.
-openrouter_patch = '''
-    # === PATCH: Route "openai/..." models to OpenRouter ===
-    if normalized_model.startswith("openai/"):
-        openrouter_key = api_key if api_key else ""
-        return {
-            "model": normalized_model,
-            "api_base": "https://openrouter.ai/api/v1",
-            "api_key": openrouter_key,
-        }
-    # === END PATCH ===
-'''
-
-# Insert after the local_model_provider check, before the HF Router fallback
 marker = """    if local_model_provider(normalized_model) is not None:
         return _resolve_local_model_params(normalized_model, reasoning_effort, strict)
 
@@ -76,12 +58,14 @@ replacement = """    if local_model_provider(normalized_model) is not None:
         return _resolve_local_model_params(normalized_model, reasoning_effort, strict)
 
     # === PATCH: Route "openai/..." models to OpenRouter ===
+    # These are OpenRouter-specific models (Laguna, Nemotron free, Owl Alpha, etc.)
+    # that aren't served by the HF Router. LiteLLM openai/ prefix + custom api_base
+    # makes it send the bare model name to OpenRouter.
     if normalized_model.startswith("openai/"):
-        openrouter_key = api_key if api_key else ""
         return {
             "model": normalized_model,
             "api_base": "https://openrouter.ai/api/v1",
-            "api_key": openrouter_key,
+            "api_key": api_key or "",
         }
     # === END PATCH ===
 
@@ -103,4 +87,8 @@ except SyntaxError as e:
 with open(LLM_PARAMS_FILE, "w") as f:
     f.write(llm_content)
 
-print("✅ Backend patched successfully")
+print("✅ Backend patched successfully")
+print("  - AVAILABLE_MODELS: 12 new models")
+print("  - Owl Alpha id: openai/owl-alpha (matches frontend modelPath)")
+print("  - _resolve_llm_params: openai/ prefix → OpenRouter routing")
+print("  - DEFAULT_FREE_MODEL_ID: deepseek-ai/DeepSeek-V4-Pro")