change load device

app.py

@@ -30,7 +30,7 @@ src_vocab = len(vocab)
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 model = Transformer(len(vocab), len(vocab), d_model, N, heads).to(device)
 model.load_state_dict(torch.load(hf_hub_download(repo_id="nickgardner/chatbot",
-                                      filename="alpaca_train_380_epoch.pt")))
+                                      filename="alpaca_train_380_epoch.pt"), map_location=device))
 model.eval()
 
 def respond(custom_string):