Question about LongVA-TPO-10k data format and DATA_PATH for TPO training #4
Description
I downloaded LongVA-TPO-10k from Hugging Face (https://huggingface.co/datasets/ruili0/LongVA-TPO-10k). It contains CSV files and a videos/ folder, but no JSON. However, the training loader seems to read JSON (or YAML pointing to JSON).
Could you confirm what DATA_PATH should point to in the training script?
If a JSON/JSONL version is expected, is there an official file or a recommended CSV→JSON conversion/schema?
Any guidance would be greatly appreciated. Thank you for releasing the code and dataset!