The OSS you use are built for the community, but above all *BY* the community. 2 amazing recent contributions that I had the pleasure of reviewing.
📦 Packing with flash attn kwargs to avoid cross-contamination by @iamgrigorev
⚡ Faster FFD packing (x3.5!) by @MarioSasko
💫 huggingface_hub v0.14 💫
We introduce a new Filesystem API for typical ops like cp, mv, ls, glob, etc! Our implementation follows fsspec, meaning out-of-the-box support for popular libraries like Pandas or DuckDB 🐍
Keep reading to discover what we baked in this release 🧵
Woohoo we won #EMNLP2022 best demo! 🤗 That’s three in a row for @huggingface 🤯🔥
2020 - Transformers
2021 - Datasets
2022 - Evaluate & Evaluation on the Hub
Now, ImageNet-1K can be downloaded in just two lines using the @huggingface datasets library!
Thanks to @MarioSasko, manual download is no longer needed.
Announcing 🤗Datasets v2.0 for dataset preparation and sharing in NLP/Vision/Audio/etc. 🔥
📚 3500+ datasets available @ https://t.co/7zbZHdO0SE
⌨️ Load in one line of 🐍Python
↔️ Share with your team and the community
And with a ✨ New Documentation !
👉https://t.co/TwJEi6NdPf
I recently noticed that our recent dataset RedCaps (https://t.co/jMGzVuztQX) has an updated dataset page on Hugging Face (https://t.co/gbw3i87aIE). It also has a nice dataset explorer!
All credits to @MarioSasko, thanks for making our work more accessible @huggingface! 🤗