No CUDA?

app.py

@@ -75,7 +75,7 @@ def response(state: AppState, audio: tuple):
     if spaces.config.Config.zero_gpu:
         if state.model_outs is not None:
             state.model_outs.past_key_values = tuple(
-                tuple(vec.cuda() for vec in tup)
+                tuple(torch.tensor(vec).cuda() for vec in tup)
                 for tup in state.model_outs.past_key_values
             )
     prev_outs = state.model_outs
@@ -91,12 +91,12 @@ def response(state: AppState, audio: tuple):
             state.conversation[-1]["content"] = resp
         # yield state, state.conversation
 
-    # del outs.logits
-    # del outs.hidden_states
-    # if spaces.config.Config.zero_gpu:
-    #     outs.past_key_values = tuple(
-    #         tuple(vec.cpu() for vec in tup) for tup in outs.past_key_values
-    #     )
+    del outs.logits
+    del outs.hidden_states
+    if spaces.config.Config.zero_gpu:
+        outs.past_key_values = tuple(
+            tuple(vec.cpu().numpy for vec in tup) for tup in outs.past_key_values
+        )
     return (
         AppState(conversation=state.conversation),
         state.conversation,