fix

app.py

@@ -163,7 +163,7 @@ def inference(image, task_type, instruction):
     return response
 
 
-inputs = [gr.inputs.Image(type='pil'), gr.Audio(source="upload", type="filepath"), gr.Video(source="upload", type="filepath"), gr.inputs.Radio(choices=['Image Captioning',  "Visual Question Answering",], type="value", default="Image Captioning", label="Task"), gr.inputs.Textbox(lines=1, label="Instruction")]
+inputs = [gr.inputs.Image(type='pil'), gr.inputs.Radio(choices=['Image Captioning',  "Visual Question Answering",], type="value", default="Image Captioning", label="Task"), gr.inputs.Textbox(lines=1, label="Instruction")]
 outputs = ['text']
 examples = [
     ['examples/images/soccer.jpg',  'Image Captioning', None],