Revert to Qwen models while Llama access is pending.

Use Qwen2.5-7B for the agent and Qwen2-VL-7B for vision so the Space works without gated Meta model approval.

agent.py

@@ -114,7 +114,8 @@ class BasicAgent:
     """Full GAIA agent with search, web, and multimodal file tools."""
 
     def __init__(self):
-        model_id = os.getenv("HF_MODEL_ID", "meta-llama/Meta-Llama-3.1-8B-Instruct")
+        # Open models — no Meta/Llama gating. Override via HF_MODEL_ID Space secret.
+        model_id = os.getenv("HF_MODEL_ID", "Qwen/Qwen2.5-7B-Instruct")
         provider = os.getenv("HF_INFERENCE_PROVIDER", "auto")
         token = os.getenv("HF_TOKEN")
 

tools.py

@@ -25,7 +25,7 @@ def _hf_client(*, for_chat: bool = False) -> InferenceClient:
 
 
 def _vision_model() -> str:
-    return os.getenv("HF_VISION_MODEL", "meta-llama/Llama-3.2-11B-Vision-Instruct")
+    return os.getenv("HF_VISION_MODEL", "Qwen/Qwen2-VL-7B-Instruct")
 
 
 def _asr_model() -> str: