Use inference providers router with Llama 3.1 default model.

Fix model_not_supported by routing through HF Inference Providers (auto) instead of assuming Qwen on hf-inference.

agent.py

@@ -114,10 +114,15 @@ class BasicAgent:
     """Full GAIA agent with search, web, and multimodal file tools."""
 
     def __init__(self):
-        model_id = os.getenv("HF_MODEL_ID", "Qwen/Qwen2.5-7B-Instruct")
+        model_id = os.getenv("HF_MODEL_ID", "meta-llama/Meta-Llama-3.1-8B-Instruct")
+        provider = os.getenv("HF_INFERENCE_PROVIDER", "auto")
         token = os.getenv("HF_TOKEN")
 
-        model = InferenceClientModel(model_id=model_id, token=token)
+        model = InferenceClientModel(
+            model_id=model_id,
+            provider=provider,
+            token=token,
+        )
         wiki_tool = WikipediaSearchTool(
             user_agent="HF-Agents-Course-Student (https://huggingface.co/agents-course)",
             language="en",
@@ -139,7 +144,7 @@ class BasicAgent:
             additional_authorized_imports=["re", "json", "math"],
         )
         tool_names = [getattr(t, "name", str(t)) for t in toolset]
-        print(f"BasicAgent initialized (model={model_id}, tools={tool_names}).")
+        print(f"BasicAgent initialized (model={model_id}, provider={provider}, tools={tool_names}).")
 
     def __call__(
         self,

tools.py

@@ -15,13 +15,17 @@ from smolagents import tool
 from youtube_transcript_api import YouTubeTranscriptApi
 
 
-def _hf_client() -> InferenceClient:
+def _hf_client(*, for_chat: bool = False) -> InferenceClient:
     token = os.getenv("HF_TOKEN")
-    return InferenceClient(token=token)
+    if for_chat:
+        provider = os.getenv("HF_INFERENCE_PROVIDER", "auto")
+    else:
+        provider = os.getenv("HF_INFERENCE_PROVIDER", "hf-inference")
+    return InferenceClient(token=token, provider=provider)
 
 
 def _vision_model() -> str:
-    return os.getenv("HF_VISION_MODEL", "Qwen/Qwen2-VL-7B-Instruct")
+    return os.getenv("HF_VISION_MODEL", "meta-llama/Llama-3.2-11B-Vision-Instruct")
 
 
 def _asr_model() -> str:
@@ -132,7 +136,7 @@ def analyze_image(file_path: str, question: str) -> str:
     }.get(path.suffix.lower(), "image/png")
 
     encoded = base64.b64encode(path.read_bytes()).decode("ascii")
-    client = _hf_client()
+    client = _hf_client(for_chat=True)
     response = client.chat_completion(
         model=_vision_model(),
         messages=[