You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While going through the repo and configs, I noticed a small inconsistency:
• In the paper (Table 1 and Appendix C), the Sparsh models are listed as ViT-B/14.
• However, the config files (config/model/*.yaml) specify patch_size: 16.
• The released checkpoints also appear to use 16×16 patch embeddings.
So it seems the models correspond to ViT-B/16 rather than ViT-B/14.
Just wanted to flag this in case it helps with reproducibility.
While going through the repo and configs, I noticed a small inconsistency:
• In the paper (Table 1 and Appendix C), the Sparsh models are listed as ViT-B/14.
• However, the config files (config/model/*.yaml) specify patch_size: 16.
• The released checkpoints also appear to use 16×16 patch embeddings.
So it seems the models correspond to ViT-B/16 rather than ViT-B/14.
Just wanted to flag this in case it helps with reproducibility.