Update app.py

app.py

@@ -54,7 +54,7 @@ def get_past_key_values(system_prompt):
     test_messages = [*start_messages, {'role': 'user', 'content': 'Hello World!'}]
     tokenized_test = tokenizer.apply_chat_template(test_messages, return_tensors='pt')    
     assert (tokenized_test[:, :tokenized.shape[1]] == tokenized).all().cpu().item()
-    return model(tokenized.to(model.device)).past_key_values
+    return model(tokenized.to(model.device)).past_key_values.cpu().detach()
 
 
 @spaces.GPU
@@ -64,7 +64,7 @@ def generate(text, past_key_values):
         {'role': 'user', 'content': text}
     ]
     response = pipe(messages, 
-                    past_key_values=past_key_values, 
+                    past_key_values=past_key_values.to(model.device), 
                     **generate_kwargs)[0]['generated_text']
     return response[-1]['content']