Update llm_node.py

llm_node.py

@@ -2,13 +2,10 @@ import os
 from huggingface_hub import InferenceClient
 
 def llm_node(question, search_result):
-    # Initialize the client (no repo_id here!)
     client = InferenceClient(token=os.getenv("HUGGINGFACEHUB_API_TOKEN"))
 
-    # Define the model you want to use
-    model_id = "HuggingFaceH4/zephyr-7b-beta"  # You can swap this with e.g., mistralai/Mistral-7B-Instruct-v0.2
+    model_id = "google/flan-t5-small"  # ✅ Small free-tier model
 
-    # Craft the prompt carefully
     prompt = f"""You are solving a GAIA benchmark evaluation question.
 
 Here’s the question:
@@ -24,15 +21,12 @@ Here’s retrieved information:
 
 Your answer:"""
 
-    # Call the model (pass model ID here)
     response = client.text_generation(
         model=model_id,
         prompt=prompt,
-        max_new_tokens=500,
+        max_new_tokens=200,  # smaller due to model limits
         temperature=0.1,
-        top_p=0.95,
-        repetition_penalty=1.1
+        top_p=0.9
     )
 
-    # Clean up response text
     return response.strip()