Add correct device

app.py

@@ -56,7 +56,7 @@ def respond(
     messages.append({"role": "user", "content": message})
 
 
-    tokenized = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True) #, tokenize=False) #
+    tokenized = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to("cuda") #, tokenize=False) #
     #print(tokenized)
     #tokenized = tokenizer(tokenized, return_tensors="pt")["input_ids"]
     print(tokenizer.batch_decode(tokenized)[0])