RuntimeError: Error reading datasets/videollava_pt/valley/023651_023700/1014646391.mp4

Hello, I encountered an issue during the pretraining phase where the video could not be read. After checking, it turns out that the video path in the JSON file is pointing to a video that is not actually stored.


I downloaded the valley dataset for the pretraining phase **strictly** according to the path from Google Drive mentioned in your README, but this issue has occurred many times (about 1 in 20 videos are missing). Have you encountered this issue during your training? Thank you very much!

Start with the Valley dataset download path in the README and compare it with the video paths referenced by the pretraining JSON files, including the missing MP4 path shown here. Check whether the reported files are absent from the downloaded dataset and document the expected resolution; the issue is done when the dataset references and available files agree or the project explains how missing videos should be handled.

Written by the indexing model from the issue text.