Update run.py

run.py

@@ -26,6 +26,8 @@ def load_hf_dataset(dataset_path, auth_token):
 
 def fast_caption(sys_prompt, usr_prompt, temp, top_p, max_tokens, model, key, endpoint, video_hf, video_hf_auth, parquet_index, video_od, video_od_auth, video_gd, video_gd_auth, frame_format, frame_limit):
     progress_info = []
+    processor = VideoProcessor(frame_format=frame_format, frame_limit=frame_limit)
+    api = AzureAPI(key=key, endpoint=endpoint, model=model, temp=temp, top_p=top_p, max_tokens=max_tokens)
     with tempfile.TemporaryDirectory() as temp_dir:
         # temp_dir = '/opt/run'
         csv_filename = os.path.join('/dev/shm', str(parquet_index).zfill(6) + '_gpt4o_caption.csv')
@@ -59,10 +61,8 @@ def fast_caption(sys_prompt, usr_prompt, temp, top_p, max_tokens, model, key, en
                             _chunk.append(_v.name)
                             md5 = hashlib.md5(binary).hexdigest()
                             
-                            processor = VideoProcessor(frame_format=frame_format, frame_limit=frame_limit)
                             frames = processor._decode(_v.name)
                             base64_list = processor.to_base64_list(frames)
-                            api = AzureAPI(key=key, endpoint=endpoint, model=model, temp=temp, top_p=top_p, max_tokens=max_tokens)
                             caption = api.get_caption(sys_prompt, usr_prompt, base64_list)
                             writer.writerow({'md5': md5, 'caption': caption})
                             # writer.writerow({'md5': md5, 'caption': 'caption'})