token check

app.py

@@ -46,7 +46,7 @@ def predict(history, prompt, max_length, top_p, temperature):
 
     model_inputs = tokenizer.build_chat_input(query, history=messages, role='user').input_ids.to(
         next(model.parameters()).device)
-    streamer = TextIteratorStreamer(tokenizer, timeout=600, skip_prompt=True, skip_special_tokens=True)
+    streamer = TextIteratorStreamer(tokenizer, timeout=600, skip_prompt=True)
     eos_token_id = [tokenizer.eos_token_id, tokenizer.get_command("<|user|>"),
                     tokenizer.get_command("<|observation|>")]
     generate_kwargs = {
@@ -63,9 +63,7 @@ def predict(history, prompt, max_length, top_p, temperature):
     t = Thread(target=model.generate, kwargs=generate_kwargs)
     t.start()
     for new_token in streamer:
-        if new_token == '<|user|>':
-            continue
-        elif new_token:
+        if new_token and '<|user|>' not in new_token:
             history[-1][1] += new_token
         yield history