fix

app.py

@@ -1,11 +1,13 @@
 import gradio as gr
+import spaces
 
 from process import inference
 
 @spaces.GPU
-def clickit(video):
+def clickit(video, prompt):
     return inference(
-        video
+        video,
+        prompt
     )
 
 with gr.Blocks() as blok:
@@ -14,13 +16,17 @@ with gr.Blocks() as blok:
             video = gr.Video(
                 label="video input",
             )
+            prompt = gr.Text(
+                label="Prompt",
+                value="Please describe this video in detail."
+            )
         with gr.Column():
             button = gr.Button("Caption it", variant="primary")
             text = gr.Text(label="Output")
         
         button.click(
             fn=clickit,
-            inputs=[video],
+            inputs=[video, prompt],
             outputs=[text]
         )
 

process.py

@@ -96,8 +96,7 @@ def predict(prompt, video_data, temperature):
         return response
 
 
-def inference(video):
-    prompt = "Please describe this video in detail."
+def inference(video, prompt):
     temperature = 0.1
     video_data = open(video, 'rb').read()
     response = predict(prompt, video_data, temperature)

requirements.txt

@@ -1,6 +1,7 @@
 argparse
 numpy
 decord
+spaces
 transformers==4.41.2
 einops==0.8.0 
 torchvision==0.15.1