Refactor SmolAgent to utilize Hugging Face's InferenceClient for direct model inference, simplifying response handling and improving error reporting. Update requirements to remove smolagents package.

app.py

@@ -3,7 +3,7 @@ import gradio as gr
 import requests
 import inspect
 import pandas as pd
-from smolagents import CodeAgent, InferenceClientModel
+from huggingface_hub import InferenceClient
 
 # (Keep Constants as is)
 # --- Constants ---
@@ -20,44 +20,42 @@ class SmolAgent:
         if not hf_token:
             raise ValueError("Hugging Face token not found. Please set HF_TOKEN environment variable in HF Spaces settings.")
         
-        # 3. Initialize the SmolLM model
-        model = InferenceClientModel(
-            model_id="HuggingFaceTB/SmolLM-1.7B-Instruct",
+        self.client = InferenceClient(
+            model="HuggingFaceTB/SmolLM-1.7B-Instruct",
             token=hf_token,
         )
-        
-        # 4. Replace your current BasicAgent with a smolagents.CodeAgent
-        self._agent = CodeAgent(
-            tools=[],
-            model=model,
-            instructions=SYSTEM_PROMPT,
-        )
-        print("SmolAgent initialized.")
+        print("SmolAgent initialized with direct inference client.")
 
     def __call__(self, question: str) -> str:
-        print(f"Agent received question (first 50 chars): {question[:50]}...")
+        prompt = f"{SYSTEM_PROMPT}\n\nQuestion: {question}\n\nAnswer:"
+        print(f"\n🪐 Running on question:\n{question}\n")
         try:
-            # The system prompt is already set, just pass the question.
-            full_response = self._agent.run(question)
-            print(f"Agent full response: {full_response}")
-
-            # Parse the response to extract the final answer
-            if "FINAL ANSWER:" in full_response:
-                # Split and get the part after "FINAL ANSWER:"
-                final_answer_part = full_response.split("FINAL ANSWER:")[1]
-                # Remove brackets if they exist and strip whitespace
+            response = self.client.text_generation(
+                prompt,
+                max_new_tokens=100,
+                temperature=0.1,
+                stop=["\n"],
+            )
+            cleaned_response = response.strip()
+            print(f"✅ Raw model response:\n{response}\n")
+            print(f"✅ Cleaned response to submit:\n{cleaned_response}\n")
+            
+            # Parse the response to extract the final answer if it follows the template
+            if "FINAL ANSWER:" in cleaned_response:
+                final_answer_part = cleaned_response.split("FINAL ANSWER:")[1]
                 final_answer = final_answer_part.strip()
                 if final_answer.startswith('[') and final_answer.endswith(']'):
                     final_answer = final_answer[1:-1]
-                
-                print(f"Agent returning parsed answer: {final_answer}")
+                print(f"✅ Extracted final answer: {final_answer}")
                 return final_answer
             else:
-                print("Warning: 'FINAL ANSWER:' not found in response. Returning full response.")
-                return full_response
-
+                print(f"⚠️ No 'FINAL ANSWER:' found, returning cleaned response")
+                return cleaned_response
+                
         except Exception as e:
-            print(f"Error running agent: {e}")
+            import traceback
+            traceback.print_exc()
+            print(f"❌ AGENT ERROR: {e}")
             return f"AGENT ERROR: {e}"
 
 
@@ -189,14 +187,14 @@ with gr.Blocks() as demo:
         """
         **Instructions:**
 
-        1.  This space uses SmolLM-1.7B-Instruct model with smolagents for question answering.
+        1.  This space uses SmolLM-1.7B-Instruct model with direct inference for question answering.
         2.  Log in to your Hugging Face account using the button below. This uses your HF username for submission.
         3.  Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
 
         ---
         **Model Information:**
         - Using: HuggingFaceTB/SmolLM-1.7B-Instruct
-        - Framework: smolagents CodeAgent
+        - Framework: Direct InferenceClient (optimized for single-line answers)
         - No additional tools (pure reasoning)
 
         **Disclaimers:**

requirements.txt

@@ -1,6 +1,4 @@
-smolagents
 huggingface_hub
 gradio
 requests
-pandas
-duckduckgo-search
+pandas