Update app.py

app.py

@@ -114,7 +114,7 @@ class SinaReasonMedicalChat:
         }
 
         # Generate the full response
-        generated_ids = self.model.generate(**generation_kwargs)[0]
+        output = self.model.generate(**generation_kwargs)[0]
         
         # Decode the response
         full_response = self.tokenizer.decode(output[len(tokenized.input_ids) : (-1 if output[-1] == tokenizer.eos_token_id else len(output) ) ])