Upload stage2_graph.py

agents/stage2_graph.py

@@ -150,12 +150,22 @@ async def call_llm(
         if not hf_token:
             raise ValueError("HF_TOKEN not set")
         
-        # Initialize client with provider
-        # Provider is set at client level, not per-call
-        client = InferenceClient(
-            token=hf_token,
-            provider=provider,
-        )
+        # New HuggingFace router URL (api-inference is deprecated)
+        router_url = "https://router.huggingface.co"
+        
+        # Initialize client with provider and new router
+        try:
+            client = InferenceClient(
+                token=hf_token,
+                provider=provider,
+                base_url=router_url,
+            )
+        except TypeError:
+            # Fallback for older versions
+            client = InferenceClient(
+                token=hf_token,
+                provider=provider,
+            )
         
         # Call without provider argument (it's set at client level)
         response = client.chat_completion(
@@ -202,7 +212,12 @@ async def call_llm(
             from huggingface_hub import InferenceClient
             
             hf_token = os.environ.get("HF_TOKEN")
-            client = InferenceClient(token=hf_token)
+            router_url = "https://router.huggingface.co"
+            
+            try:
+                client = InferenceClient(token=hf_token, base_url=router_url)
+            except TypeError:
+                client = InferenceClient(token=hf_token)
             
             # Try appending provider to model name
             model_with_provider = f"{model}:{provider}"