It would be nice to provide some pre-converted models to make things easier for new users, especially new users with low memory trying to use SDXL. There are some models with licenses that allow redistribution, but a full ONNX model has multiple folders and files, so they will need to be kept in HF repositories or some kind of archive file.
It should be easy to host some converted models in an R2/S3 bucket as ZIP archives and extract them as the "conversion" step. That can be set up as a new conversion pipeline and new file format in the extras schema.
It would be nice to provide some pre-converted models to make things easier for new users, especially new users with low memory trying to use SDXL. There are some models with licenses that allow redistribution, but a full ONNX model has multiple folders and files, so they will need to be kept in HF repositories or some kind of archive file.
It should be easy to host some converted models in an R2/S3 bucket as ZIP archives and extract them as the "conversion" step. That can be set up as a new conversion pipeline and new file format in the extras schema.