Update app.py

app.py

@@ -55,7 +55,8 @@ latex_delimiters_set = [{
 def predict(message, history, system_prompt, temperature, max_new_tokens, top_k, repetition_penalty, top_p):
     # Format history with a given chat template
     
-    #stop_tokens = [tokenizer.eos_token_id]
+    stop_tokens = [tokenizer.eos_token_id]
+    pad_token_id=[tokenizer.eos_token_id] #151643
     instruction = system_prompt + "\n\n"
     for user, assistant in history:
         instruction += f"<｜User｜>{user}<｜Assistant｜>{assistant}\n"
@@ -87,8 +88,8 @@ def predict(message, history, system_prompt, temperature, max_new_tokens, top_k,
     outputs = []
     for new_token in streamer:
         outputs.append(new_token)
-        #if new_token in stop_tokens:
-            #break
+        if new_token in stop_tokens:
+            break
         yield "".join(outputs)