Update app.py

app.py

@@ -117,7 +117,7 @@ class SinaReasonMedicalChat:
         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) ) ])
+        full_response = self.tokenizer.decode(output[len(tokenized.input_ids) : (-1 if output[-1] == self.tokenizer.eos_token_id else len(output) ) ])
 
         # Extract thinking and clinical summary
         thinking, response = self.extract_thinking_and_response(full_response)