Update app.py

app.py

@@ -68,10 +68,11 @@ def model_inference(input_dict, history):
     yield "Thinking..."
     for new_text in streamer:
         buffer += new_text
+        # Remove <|im_end|> or similar tokens from the output
+        buffer = buffer.replace("<|im_end|>", "")
         time.sleep(0.01)
         yield buffer
 
-
 # Example inputs
 examples = [
     [{"text": "summarize the letter", "files": ["examples/1.png"]}],