chunked/nltk_chunking/tasks_video_classification/chunk_43.txt

Also, if your dataset follows a similar structure (as shown above), then using the pytorchvideo.data.Ucf101() should work just fine.