Running Video-MME with a different number of frames

I tried to evaluate Qwen2.5-VL-3B-Instruct on the Video-MME data set but with a different number of frames. I added the following line to video_dataset_config.py:

`videomme_dataset = {
'Video-MME_8frame': partial(VideoMME, dataset='Video-MME', nframe=8),
'Video-MME_8frame_subs': partial(VideoMME, dataset='Video-MME', nframe=8, use_subtitle=True),
'Video-MME_1fps': partial(VideoMME, dataset='Video-MME', fps=1.0),
'Video-MME_0.5fps': partial(VideoMME, dataset='Video-MME', fps=0.5),
'Video-MME_0.5fps_subs': partial(VideoMME, dataset='Video-MME', fps=0.5, use_subtitle=True),
'Video-MME_768frame': partial(VideoMME, dataset='Video-MME', nframe=768), # <- added line
}```

I then ran the evaluation using:

python run.py --data Video-MME_768frame --model Qwen2.5-VL-3B-Instruct

I had two issues doing it this way. The first is that it is taking a lot longer than it should be. The other is that the amount of time for each video got a lot longer for longer videos, which should not be the case since the number of frames should be constant.
Is this the wrong way to do it?

Another thing I wanted to do is to evaluate the model using a constant frame rate of 2fps, but with a maximum number of frames capped at 768 frames for each video. Is this option supported? If so, how to do it?