Modern ML engineering is 90% knowledge of git
The famous @ycombinator dropped a very cool article to showcase their 25+ companies who have trained their own AI models, rather than just use someone else's closed model through an API like a black box.
While llamafile cares deeply about helping the GPU poor, it offers a first-class experience to the 1% too.
Most of the main features of 🤗 http://huggingface.co are working with JavaScript disabled 🤯
You can still login or download model files.
A lot of sites just doesn't work without js nowadays.
The more than 500k models on huggingface provide vast opportunities for creating merged models. SakanaAILabs's Evolutionary Model Merge algorithm automatically finds optimal solutions in the space of possible model merges.
Organizations on huggingface can now set a default Repo visibility.
Your org's Models/datasets can be:
- public by default
- private by default
- private only (only admins can switch them to public)
Better AI governance for your company 🔥🔥
Very cool to see a bunch of new smaller datasets trending on http://hf.co/datasets!
All datasets are worth sharing:
- big or small
- in French, in Burmese, in Sanskrit or any language
- in text, image, audio, video, 3d, time-series, multi-modal,...
- with restrictive or open-licenses
They all contribute to making the field more inclusive, transparent & empower everyone to build AI!
AI building is becoming more collaborative within organizations! That's why we're super excited to introduce access control & resource groups to manage individual team member access with read and write for models, datasets and apps in private or public mode.
Enabled by default on Enterprise Hub: https://huggingface.co/enterprise