Update demo.

app.py

@@ -130,7 +130,7 @@ def save_video_to_local(video_path):
 
 
 @spaces.GPU(duration=120)
-def generate(image, video, first_run, state, state_, textbox_in, tensor, modals, dtype=torch.float16):
+def generate(image, video, first_run, state, state_, textbox_in, dtype=torch.float16):
     flag = 1
     if not textbox_in:
         if len(state_.messages) > 0:
@@ -144,11 +144,12 @@ def generate(image, video, first_run, state, state_, textbox_in, tensor, modals,
     video = video if video else "none"
     assert not (os.path.exists(image) and os.path.exists(video))
 
+    tensor = []
+    modals = []
+
     if type(state) is not Conversation:
         state = conv_templates[conv_mode].copy()
         state_ = conv_templates[conv_mode].copy()
-        tensor = []
-        modals = []
 
     first_run = False if len(state.messages) > 0 else True
 
@@ -192,29 +193,25 @@ def generate(image, video, first_run, state, state_, textbox_in, tensor, modals,
     state.append_message(state.roles[1], textbox_out)
 
     return (gr.update(value=image if os.path.exists(image) else None, interactive=True), gr.update(value=video if os.path.exists(video) else None, interactive=True), 
-            state.to_gradio_chatbot(), False, state, state_, gr.update(value=None, interactive=True), 
-            # tensor, modals
-            )
+            state.to_gradio_chatbot(), False, state, state_, gr.update(value=None, interactive=True))
 
 
-def regenerate(state, state_, textbox, tensor, modals):
+def regenerate(state, state_, textbox):
     state.messages.pop(-1)
     state_.messages.pop(-1)
-    tensor.pop(-1)
-    modals.pop(-1)
     textbox = gr.update(value=None, interactive=True)
     if len(state.messages) > 0:
-        return state, state_, textbox, state.to_gradio_chatbot(), False, tensor, modals
-    return state, state_, textbox, state.to_gradio_chatbot(), True, tensor, modals
+        return state, state_, textbox, state.to_gradio_chatbot(), False
+    return state, state_, textbox, state.to_gradio_chatbot(), True
 
 
-def clear_history(state, state_, tensor, modals):
+def clear_history(state, state_):
     state = conv_templates[conv_mode].copy()
     state_ = conv_templates[conv_mode].copy()
     return (gr.update(value=None, interactive=True),
             gr.update(value=None, interactive=True), \
             state.to_gradio_chatbot(), \
-            True, state, state_, gr.update(value=None, interactive=True), [], [])
+            True, state, state_, gr.update(value=None, interactive=True))
 
 
 conv_mode = "llama_2"
@@ -261,8 +258,8 @@ with gr.Blocks(title='VideoLLaMA2🚀', theme=gr.themes.Default(), css=block_css
     state = gr.State()
     state_ = gr.State()
     first_run = gr.State()
-    tensor = gr.State()
-    modals = gr.State()
+    # tensor = gr.State()
+    # modals = gr.State()
 
     with gr.Row():
         with gr.Column(scale=3):
@@ -306,15 +303,24 @@ with gr.Blocks(title='VideoLLaMA2🚀', theme=gr.themes.Default(), css=block_css
     gr.Markdown(tos_markdown)
     gr.Markdown(learn_more_markdown)
 
-    submit_btn.click(generate, [image, video, first_run, state, state_, textbox, tensor, modals],
-                    [image, video, chatbot, first_run, state, state_, textbox, 
+    submit_btn.click(
+        generate, 
+        [image, video, first_run, state, state_, textbox],
+        [image, video, chatbot, first_run, state, state_, textbox, 
                     #  tensor, modals
                      ])
 
-    regenerate_btn.click(regenerate, [state, state_, textbox, tensor, modals], [state, state_, textbox, chatbot, first_run, tensor, modals]).then(
-        generate, [image, video, first_run, state, state_, textbox, tensor, modals], [image, video, chatbot, first_run, state, state_, textbox, tensor, modals])
-
-    clear_btn.click(clear_history, [state, state_, tensor, modals],
-                    [image, video, chatbot, first_run, state, state_, textbox, tensor, modals])
+    regenerate_btn.click(
+        regenerate, 
+        [state, state_, textbox], 
+        [state, state_, textbox, chatbot, first_run]).then(
+        generate, 
+        [image, video, first_run, state, state_, textbox], 
+        [image, video, chatbot, first_run, state, state_, textbox])
+
+    clear_btn.click(
+        clear_history, 
+        [state, state_],
+        [image, video, chatbot, first_run, state, state_, textbox])
 
 demo.launch()