Refactor SmolAgent to use chat completion format

app.py

@@ -27,17 +27,26 @@ class SmolAgent:
         print("SmolAgent initialized with direct inference client.")
 
     def __call__(self, question: str) -> str:
-        prompt = f"{SYSTEM_PROMPT}\n\nQuestion: {question}\n\nAnswer:"
         print(f"\n🪐 Running on question:\n{question}\n")
         try:
-            response = self.client.text_generation(
-                prompt,
-                max_new_tokens=100,
+            # Use chat completion format which is more reliable
+            messages = [
+                {"role": "system", "content": SYSTEM_PROMPT},
+                {"role": "user", "content": question}
+            ]
+            
+            response = self.client.chat_completion(
+                messages=messages,
+                max_tokens=100,
                 temperature=0.1,
                 stop=["\n"],
             )
-            cleaned_response = response.strip()
-            print(f"✅ Raw model response:\n{response}\n")
+            
+            # Extract the assistant's response
+            assistant_message = response.choices[0].message.content
+            cleaned_response = assistant_message.strip()
+            
+            print(f"✅ Raw model response:\n{assistant_message}\n")
             print(f"✅ Cleaned response to submit:\n{cleaned_response}\n")
             
             # Parse the response to extract the final answer if it follows the template
@@ -56,7 +65,23 @@ class SmolAgent:
             import traceback
             traceback.print_exc()
             print(f"❌ AGENT ERROR: {e}")
-            return f"AGENT ERROR: {e}"
+            # Fallback: try text_generation with different parameters
+            try:
+                print("🔄 Trying fallback text_generation method...")
+                prompt = f"{SYSTEM_PROMPT}\n\nQuestion: {question}\n\nAnswer:"
+                response = self.client.text_generation(
+                    prompt,
+                    max_new_tokens=50,
+                    temperature=0.1,
+                    do_sample=True,
+                    return_full_text=False,
+                )
+                cleaned_response = response.strip()
+                print(f"✅ Fallback response: {cleaned_response}")
+                return cleaned_response
+            except Exception as fallback_error:
+                print(f"❌ Fallback also failed: {fallback_error}")
+                return f"AGENT ERROR: {e}"
 
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):