Update llm_node.py

llm_node.py

@@ -1,16 +1,14 @@
-from langchain_community.llms import HuggingFaceHub
+import os
+from huggingface_hub import InferenceClient
 
 def llm_node(question, search_result):
-    # Initialize Hugging Face model (free)
-    llm = HuggingFaceHub(
-        repo_id="HuggingFaceH4/zephyr-7b-beta",  # You can swap with mistral or mixtral if needed
-        model_kwargs={
-            "temperature": 0.1,
-            "max_new_tokens": 500
-        }
+    # Initialize the Hugging Face Inference Client
+    client = InferenceClient(
+        repo_id="HuggingFaceH4/zephyr-7b-beta",  # ✅ You can swap in another, e.g., mistralai/Mistral-7B-Instruct-v0.2
+        token=os.getenv("HUGGINGFACEHUB_API_TOKEN")
     )
 
-    # Build prompt combining search + question
+    # Craft the prompt carefully
     prompt = f"""You are solving a GAIA benchmark evaluation question.
 
 Here’s the question:
@@ -26,5 +24,14 @@ Here’s retrieved information:
 
 Your answer:"""
 
-    response = llm.invoke(prompt)
+    # Call the model
+    response = client.text_generation(
+        prompt,
+        max_new_tokens=500,
+        temperature=0.1,
+        top_p=0.95,
+        repetition_penalty=1.1
+    )
+
+    # Clean up response text
     return response.strip()