update context

app.py

@@ -82,23 +82,24 @@ def analyze_video(prompt, video, progress=gr.Progress()):
     else:
         video_path = video.name
     
-    progress(10, desc="Encoding video")
+    progress(0.1, desc="Encoding video")
     encoded_video = encode_video(video_path)
     
-    progress(30, desc="Extracting audio")
+    progress(0.3, desc="Extracting audio")
     # Extract audio and transcribe
     audio_path = extract_audio(video_path)
     
-    progress(50, desc="Transcribing audio")
+    progress(0.5, desc="Transcribing audio")
     transcription = transcribe_audio(audio_path)
     
     # Clean up temporary audio file
     os.remove(audio_path)
     
-    progress(70, desc="Preparing context")
+    progress(0.7, desc="Preparing context")
     context = [
-        {"role": "user", "content": [prompt] + encoded_video},
-        {"role": "assistant", "content": f"Transcription of the video: {transcription}"}
+        {"role": "user", "content": encoded_video},
+        {"role": "assistant", "content": f"Transcription of the video: {transcription}"},
+        {"role": "user", "content": prompt}
     ]
     params = {
         'sampling': True,
@@ -112,16 +113,16 @@ def analyze_video(prompt, video, progress=gr.Progress()):
         "max_slice_nums": 1 if len(encoded_video) > 16 else 2
     }
     
-    progress(80, desc="Generating response")
+    progress(0.8, desc="Generating response")
     response = model.chat(image=None, msgs=context, tokenizer=tokenizer, **params)
     
-    progress(90, desc="Finalizing")
+    progress(0.9, desc="Finalizing")
     end_time = time.time()
     processing_time = end_time - start_time
     analysis_result = f"Analysis Result:\n{response}\n\n"
     processing_time = f"Processing Time: {processing_time:.2f} seconds"
     
-    progress(100, desc="Complete")
+    progress(1, desc="Complete")
     return analysis_result, processing_time
 
 with gr.Blocks() as demo: