Update app.py

app.py

@@ -103,9 +103,9 @@ def respond(message, chat_history):
 
         tokenized_chat = tokenizer.apply_chat_template(context, tokenize=True, add_generation_prompt=True, return_tensors="pt")
 
-        outputs = model.generate(tokenized_chat, temp=0.5, top_k = 40, top_p = 1, max_tokens = max_new_tokens)
+        bot_message = model.generate(tokenized_chat, temp=0.5, top_k = 40, top_p = 1, max_tokens = max_new_tokens)
 
-        bot_message = tokenizer.decode(outputs[0]).split("<|assistant|>")[-1].replace("</s>","")
+        #bot_message = tokenizer.decode(outputs[0]).split("<|assistant|>")[-1].replace("</s>","")
 
         #bot_message = model.generate(prompt=prompt, temp=0.5, top_k = 40, top_p = 1, max_tokens = max_new_tokens, streaming=False)