Update app.py

app.py

@@ -57,12 +57,14 @@ start_messages = [
 
 # past_key_values = PastKV()
 
+
 def past_kv_to_device(past_kv, device):
-    return tuple((k.to(device).detach(), v.to(device).detach()) for k, v in past_kv)
+    return tuple((torch.tensor(k).to(device).detach(), torch.tensor(v).to(device).detach()) for k, v in past_kv)
 
 def detach_past_kv(past_kv):
     return tuple((k.cpu().detach().numpy().tolist(), v.cpu().detach().numpy().tolist()) for k, v in past_kv)
 
+
 @spaces.GPU
 def set_past_key_values():
     model, tokenizer = pipe.model, pipe.tokenizer
@@ -91,6 +93,6 @@ if __name__ == "__main__":
     with torch.no_grad():
         past_key_values = set_past_key_values()
         print(f'{past_key_values=}')
-        demo = gr.Interface(generate, 
-                            inputs=["textbox", gr.State(past_key_values)], outputs="textbox")
+        demo = gr.Interface(partial(generate, past_key_values=past_kv_to_device(past_key_values, pipe.model.device)), 
+                            inputs="textbox", outputs="textbox")
         demo.launch()