soniajoseph / ViT-Prisma

ViT Prisma is a mechanistic interpretability library for Vision Transformers (ViTs).
Other
165 stars 18 forks source link

Make sure all the functions used in the Interactive Attention Head Tour are documented #86

Open soniajoseph opened 6 months ago

soniajoseph commented 6 months ago

Docstrings are not complete for functions. We can prioritize our public facing tutorials, and the functions they use, in terms of what we should document first.

Document all the vit_prisma functions imported into the Interactive Attention Head Tour: https://colab.research.google.com/drive/1P252fCvTHNL_yhqJDeDVOXKCzIgIuAz2#scrollTo=xrAzVMmb-DmG

Ensure that we have type and shape for all input and outputs for the functions. Follow best documentation practices.