Update app.py

app.py

@@ -106,7 +106,7 @@ class BasicModel:
         # This serialization might need adjustment based on how CodeAgent sends prompts.
         # It needs to handle both initial strings and potential chat histories (as a list of ChatMessage objects).
         if isinstance(messages, str):
-            return f"<|start_header_id|>user<|end_header_id|>\n\n{messages}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
+            return f"<|im_start|>user\n\n{messages}<|im_end|><|im_start|>assistant<|im_end|>\n\n"
             
         prompt = []
         for m in messages: ## <- For each ChatMessage object
@@ -120,12 +120,12 @@ class BasicModel:
                  content_text = "".join([c.get('text', '') for c in content if c.get('type') == 'text'])
             elif isinstance(content, str): ## <- If m.content is simply a string
                  content_text = content
-            prompt.append(f"<|start_header_id|>{role}<|end_header_id|>\n\n{content_text}<|eot_id|>") # Llama 3.1 format
+            prompt.append(f"<|im_start|>{role}\n\n{content_text}<|im_end|>") # Qwen3 format
 
         print(f"{role}: {content_text}") ## <- Print the last message in log
         
         # Add the assistant prompt start
-        prompt.append("<|start_header_id|>assistant<|end_header_id|>\n\n")
+        prompt.append("<|im_start|>assistant\n\n")
         
         return "".join(prompt)
         
@@ -137,7 +137,7 @@ class BasicModel:
         # 2. Get the response
         terminators = [
             self.pipe.tokenizer.eos_token_id,
-            self.pipe.tokenizer.convert_tokens_to_ids("<|eot_id|>")
+            self.pipe.tokenizer.convert_tokens_to_ids("<|endoftext|>")
         ]
         prompt_str = self._serialize_messages(prompt)
         outputs = self.pipe(prompt_str, eos_token_id=terminators, **gen_kwargs)