Update demos.

app.py

@@ -92,7 +92,7 @@ class Chat:
 
     @spaces.GPU(duration=120)
     @torch.inference_mode()
-    def generate(self, tensor: list, modals: list, prompt: str, first_run: bool, state):
+    def generate(self, tensor: list, modals: list, prompt: str, first_run: bool, state, temperature, top_p, max_output_tokens):
         # TODO: support multiple turns of conversation.
         assert len(tensor) == len(modals)
 
@@ -117,8 +117,9 @@ class Chat:
                 images_or_videos=tensor,
                 modal_list=modals,
                 do_sample=True,
-                temperature=0.2,
-                max_new_tokens=1024,
+                temperature=temperature,
+                top_p=top_p,
+                max_new_tokens=max_output_tokens,
                 use_cache=True,
                 stopping_criteria=[stopping_criteria],
             )
@@ -129,7 +130,7 @@ class Chat:
 
 
 @spaces.GPU(duration=120)
-def generate(image, video, first_run, state, state_, textbox_in, dtype=torch.float16):
+def generate(image, video, first_run, state, state_, textbox_in, temperature, top_p, max_output_tokens, dtype=torch.float16):
     flag = 1
     if not textbox_in:
         if len(state_.messages) > 0:
@@ -174,7 +175,7 @@ def generate(image, video, first_run, state, state_, textbox_in, dtype=torch.flo
     if os.path.exists(image) and os.path.exists(video):
         text_en_in = text_en_in.replace(DEFAULT_MMODAL_TOKEN['VIDEO'], '').strip()
         text_en_in = DEFAULT_MMODAL_TOKEN['VIDEO'] + '\n' + text_en_in
-    text_en_out, state_ = handler.generate(tensor, modals, text_en_in, first_run=first_run, state=state_)
+    text_en_out, state_ = handler.generate(tensor, modals, text_en_in, first_run=first_run, state=state_, temperature=temperature, top_p=top_p, max_output_tokens=max_output_tokens)
     state_.messages[-1] = (state_.roles[1], text_en_out)
 
     text_en_out = text_en_out.split('#')[0]
@@ -236,12 +237,65 @@ with gr.Blocks(title='VideoLLaMA 2 🔥🚀🔥', theme=gr.themes.Default(primar
             image = gr.Image(label="Input Image", type="filepath")
             video = gr.Video(label="Input Video")
 
+            with gr.Accordion("Parameters", open=False) as parameter_row:
+                # num_beams = gr.Slider(
+                #     minimum=1,
+                #     maximum=10,
+                #     value=1,
+                #     step=1,
+                #     interactive=True,
+                #     label="beam search numbers",
+                # )
+
+                temperature = gr.Slider(
+                    minimum=0.1,
+                    maximum=1.0,
+                    value=0.2,
+                    step=0.1,
+                    interactive=True,
+                    label="Temperature",
+                )
+
+                top_p = gr.Slider(
+                        minimum=0.0,
+                        maximum=1.0,
+                        value=0.7,
+                        step=0.1,
+                        interactive=True,
+                        label="Top P",
+                )
+
+                max_output_tokens = gr.Slider(
+                    minimum=64,
+                    maximum=1024,
+                    value=512,
+                    step=64,
+                    interactive=True,
+                    label="Max output tokens",
+                )
+
+        with gr.Column(scale=7):
+            chatbot = gr.Chatbot(label="VideoLLaMA 2", bubble_full_width=True, height=750)
+            with gr.Row():
+                with gr.Column(scale=8):
+                    textbox.render()
+                with gr.Column(scale=1, min_width=50):
+                    submit_btn = gr.Button(value="Send", variant="primary", interactive=True)
+            with gr.Row(elem_id="buttons") as button_row:
+                upvote_btn     = gr.Button(value="👍  Upvote", interactive=True)
+                downvote_btn   = gr.Button(value="👎  Downvote", interactive=True)
+                # flag_btn     = gr.Button(value="⚠️  Flag", interactive=True)
+                # stop_btn     = gr.Button(value="⏹️  Stop Generation", interactive=False)
+                regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=True)
+                clear_btn      = gr.Button(value="🗑️  Clear history", interactive=True)
+
+        with gr.Column():
             cur_dir = os.path.dirname(os.path.abspath(__file__))
             gr.Examples(
                 examples=[
                     [
                         f"{cur_dir}/examples/extreme_ironing.jpg",
-                        "What is unusual about this image?",
+                        "What is the phone recording?",
                     ],
                     [
                         f"{cur_dir}/examples/waterview.jpg",
@@ -254,28 +308,30 @@ with gr.Blocks(title='VideoLLaMA 2 🔥🚀🔥', theme=gr.themes.Default(primar
                 ],
                 inputs=[image, textbox],
             )
-
-        with gr.Column(scale=7):
-            chatbot = gr.Chatbot(label="VideoLLaMA 2", bubble_full_width=True, height=750)
-            with gr.Row():
-                with gr.Column(scale=8):
-                    textbox.render()
-                with gr.Column(scale=1, min_width=50):
-                    submit_btn = gr.Button(value="Send", variant="primary", interactive=True)
-            with gr.Row(elem_id="buttons") as button_row:
-                upvote_btn     = gr.Button(value="👍  Upvote", interactive=True)
-                downvote_btn   = gr.Button(value="👎  Downvote", interactive=True)
-                # flag_btn     = gr.Button(value="⚠️  Flag", interactive=True)
-                # stop_btn     = gr.Button(value="⏹️  Stop Generation", interactive=False)
-                regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=True)
-                clear_btn      = gr.Button(value="🗑️  Clear history", interactive=True)
+            gr.Examples(
+                examples=[
+                    [
+                        f"{cur_dir}/examples/rap.mp4",
+                        "What happens in this video?",
+                    ],
+                    [
+                        f"{cur_dir}/examples/demo2.mp4",
+                        "Do you think it's morning or night in this video? Why?",
+                    ],
+                    [
+                        f"{cur_dir}/examples/demo3.mp4",
+                        "At the intersection, in which direction does the red car turn?",
+                    ],
+                ],
+                inputs=[video, textbox],
+            )
 
     gr.Markdown(tos_markdown)
     gr.Markdown(learn_more_markdown)
 
     submit_btn.click(
         generate, 
-        [image, video, first_run, state, state_, textbox],
+        [image, video, first_run, state, state_, textbox, temperature, top_p, max_output_tokens],
         [image, video, chatbot, first_run, state, state_, textbox])
 
     regenerate_btn.click(
@@ -283,7 +339,7 @@ with gr.Blocks(title='VideoLLaMA 2 🔥🚀🔥', theme=gr.themes.Default(primar
         [state, state_, textbox], 
         [state, state_, textbox, chatbot, first_run]).then(
         generate, 
-        [image, video, first_run, state, state_, textbox], 
+        [image, video, first_run, state, state_, textbox, temperature, top_p, max_output_tokens], 
         [image, video, chatbot, first_run, state, state_, textbox])
 
     clear_btn.click(

examples/{1034346401.mp4 → demo2.mp4}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08b62a634fe49edc0a19fc53f6ea5cfb345d9b2a6a7047811344c16832dc42b2
-size 1678095
+oid sha256:aec475bc837a1372f0b1c9ccea2c0c293f8d90f3f381f68f0691964d6d48fdca
+size 3292167

examples/{sample_demo_1.mp4 → demo3.mp4}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc6562a172eb9cb3c760a3c9992349c1faa2c793c112b7b9e50bd5cb17c2164d
-size 1549315
+oid sha256:a8783c215899aea0324b74bc4254b105bc2aa1759080dca0eb8166b2405e8cd5
+size 4527999

examples/{sample_demo_3.mp4 → rap.mp4}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da6126bce64c64a3d6f7ce889fbe15b5f1c2e3f978846351d8c7a79a950b429e
-size 463547
+oid sha256:cde3c54b5e9aba7cf77c6530684a4bba45f661dd1ab664043375bf9582196200
+size 13779546

examples/sample_demo_9.mp4

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f9702694f185e27ae016b85024b367e140cf93a4e3124d072816fd32f2ca0d96
-size 631864