This is a project that attempts to port Huggingface's Diffusers and Transformers libraries to Tinygrad, a more cross-vendor compatible ML framework.
- no instructions yet :c
The way the internals work is by providing a layer (tg_adapter) that helps more easily convert pytorch code into tinygrad code. The primary focus as of now is to make this layer more comprehensive enough that it can be a drop-in replacement pytorch, at least for all the functionality used in diffusers and transformers.
Further on it may be a priority to convert all the stuff to tinygrad code, but this initial approach will be much easier to create and validate!