Longvideobench的评测是不是有bug？

    if video_llm:
        message.append(dict(type='video', value=osp.join(self.data_root, line['video_path'])))
    else:
        if not self.use_subtitle:
            with open(osp.join(self.data_root, "subtitles", line["subtitle_path"])) as f:
                subtitles = json.load(f)

            frame_message = insert_subtitles_into_frames(
                frames,
                [ind_ / fps for ind_ in indices],
                subtitles,
                line["starting_timestamp_for_subtitles"],
                line["duration"]
            )

            message += frame_message
        else:
            for im in frames:
                message.append(dict(type='image', value=im))

如果模型是video_llm是不是相当于无论怎么设置其实都是不加字幕的？但longvideobench应该是默认加字幕来评测的吧。
然后if not self.use_subtitle这块感觉命名有点疑惑，虽然这个相当于默认加字幕是合理的，但是这个命名感觉让人困惑。

Start from the shown Python subtitle-handling branch and trace the LongVideoBench evaluation entry point, including video_llm and use_subtitle. Verify whether video_llm bypasses subtitles and whether the default matches LongVideoBench’s stated evaluation behavior; done means the behavior and the confusing option naming are resolved consistently.