Test 3

app.py

@@ -78,12 +78,11 @@ def response(state: AppState, audio: tuple):
                 tuple(torch.tensor(vec).cuda() for vec in tup)
                 for tup in state.model_outs.past_key_values
             )
-            print(outs)
     prev_outs = state.model_outs
     start = False
     for resp, outs in diva_audio(
         (state.sampling_rate, state.stream),
-        prev_outs=(prev_out if state.model_outs is not None is not None else None),
+        prev_outs=(prev_outs if state.model_outs is not None else None),
     ):
         if not start:
             state.conversation.append({"role": "assistant", "content": resp})
@@ -99,7 +98,7 @@ def response(state: AppState, audio: tuple):
             tuple(vec.cpu().numpy() for vec in tup) for tup in outs.past_key_values
         )
     return (
-        AppState(conversation=state.conversation),
+        AppState(conversation=state.conversation, model_outs=outs),
         state.conversation,
     )
 
@@ -190,7 +189,7 @@ with gr.Blocks(theme=theme, js=js) as demo:
     respond = input_audio.stop_recording(
         response, [state, input_audio], [state, chatbot]
     )
-    restart = respond.then(start_recording_user, [state], [input_audio]).then(
+    restart = respond.success(start_recording_user, [state], [input_audio]).then(
         lambda state: state, state, state, js=js_reset
     )