Missing la_train_info_ai_6k_type3.json training on Video-GoBench

[outerform]

Hi!

I am trying to reproduce the training process the go training data, but a configuration file: ./data/go_dataset_size9/la_train_info_ai_6k_type3.json is missin, which is used by LDM/ldm/datasets/go_dataset.py, line 96.

Could you please provide it?

[MaverickRen]

Apologies for the inconvenience caused by the bug. I've fixed the path dependency issue in the data loading and added supplementary instructions for VideoGoBench.

[outerform]

Thank you.

VideoWorld/VideoWorld/configs/go_battle_vs_human.py

Line 112 in 2e1efee

shards='data/go_dataset_size9/kataselfpaly_filterdup_ignore_wcap_10M_image_tar/{000000000..000001259}.tar',

It seems the train_dataloader use 1260 tar files. But there are only 120 tar files on huggingface. Are there some files missing?

[MaverickRen]

Apologies, our Hugging Face dataset was indeed not fully uploaded with all 1260 files because it is extremely large (several terabytes). The previous 100+ subsets were randomly selected from the 1260 tar files, and the data distribution and format are completely identical. If you still wish to download the complete dataset, I can continue uploading the remaining tar files (in fact, I have already uploaded the vast majority of them