Update app.py

app.py

@@ -53,7 +53,7 @@ def stream_chat(message: str, history: list, temperature: float, max_new_tokens:
     print(f"Conversation is -\n{conversation}")
     
     input_ids = tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True)
-    inputs = tokenizer(input_ids, return_tensors="pt").to(0)
+    inputs = tokenizer(input_ids, return_tensors="pt").to("cpu") #GPU 0, CPU 1
     
     streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)