Fix: Add fallback for InferenceClient API compatibility

app.py

@@ -193,20 +193,36 @@ class InferenceAPIBot:
     def generate_answer(self, prompt: str, **kwargs) -> str:
         """Generate answer using Inference API"""
         try:
-            # Convert prompt to chat format
-            messages = [{"role": "user", "content": prompt}]
+            # Use text generation API (more reliable than chat.completions)
+            # The InferenceClient supports both formats, but text_generation is more stable
+            max_tokens = kwargs.get('max_new_tokens', 512)
+            temperature = kwargs.get('temperature', 0.2)
+            top_p = kwargs.get('top_p', 0.9)
             
-            # Call Inference API
-            completion = self.client.chat.completions.create(
-                model=self.current_model,
-                messages=messages,
-                max_tokens=kwargs.get('max_new_tokens', 512),
-                temperature=kwargs.get('temperature', 0.2),
-                top_p=kwargs.get('top_p', 0.9),
-            )
-            
-            answer = completion.choices[0].message.content
-            return answer
+            # Try chat.completions first (newer API)
+            try:
+                messages = [{"role": "user", "content": prompt}]
+                completion = self.client.chat.completions.create(
+                    model=self.current_model,
+                    messages=messages,
+                    max_tokens=max_tokens,
+                    temperature=temperature,
+                    top_p=top_p,
+                )
+                answer = completion.choices[0].message.content
+                return answer
+            except (AttributeError, TypeError) as e:
+                # Fallback to text generation API if chat.completions not available
+                logger.warning(f"chat.completions not available, using text_generation: {e}")
+                response = self.client.text_generation(
+                    prompt,
+                    model=self.current_model,
+                    max_new_tokens=max_tokens,
+                    temperature=temperature,
+                    top_p=top_p,
+                    return_full_text=False,
+                )
+                return response
         except Exception as e:
             logger.error(f"Error calling Inference API: {e}", exc_info=True)
             return f"Error generating answer: {str(e)}"
@@ -264,14 +280,30 @@ class InferenceAPIBot:
             ]
             
             # Call Inference API
-            completion = self.client.chat.completions.create(
-                model=self.current_model,
-                messages=messages,
-                max_tokens=512 if target_level in ["college", "doctoral"] else 384,
-                temperature=0.4 if target_level in ["college", "doctoral"] else 0.3,
-            )
+            max_tokens = 512 if target_level in ["college", "doctoral"] else 384
+            temperature = 0.4 if target_level in ["college", "doctoral"] else 0.3
             
-            enhanced_answer = completion.choices[0].message.content
+            try:
+                # Try chat.completions first
+                completion = self.client.chat.completions.create(
+                    model=self.current_model,
+                    messages=messages,
+                    max_tokens=max_tokens,
+                    temperature=temperature,
+                )
+                enhanced_answer = completion.choices[0].message.content
+            except (AttributeError, TypeError) as e:
+                # Fallback to text generation
+                logger.warning(f"chat.completions not available for readability, using text_generation: {e}")
+                # Combine system and user messages for text generation
+                combined_prompt = f"{system_message}\n\n{user_message}"
+                enhanced_answer = self.client.text_generation(
+                    combined_prompt,
+                    model=self.current_model,
+                    max_new_tokens=max_tokens,
+                    temperature=temperature,
+                    return_full_text=False,
+                )
             # Clean the answer (same as bot.py)
             cleaned = self.bot._clean_readability_answer(enhanced_answer, target_level)