Update app.py

app.py

@@ -23,7 +23,7 @@ system_prompt = "<|im_start|>system\nYou are Dolphin, a helpful AI assistant.<|i
 def predict(message, history):
     history_transformer_format = history + [[message, ""]]
     messages = system_prompt + "".join(["".join(["\n<|im_start|>user\n" + item[0], "<|im_end|>\n<|im_start|>assistant\n" + item[1]]) for item in history_transformer_format])
-    input_ids = tokenizer([messages], return_tensors="pt").to('cuda').input_ids
+    input_ids = tokenizer([messages], return_tensors="pt").input_ids
     streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
     generate_kwargs = {
         'input_ids': input_ids,