varunneal

/

flash-attention-3

like 12

Hello there 👋

We're in the process of migrating the kernels to a new repo type "kernel", which will provide better integration support from the rest of the Hugging Face Hub ecosystem. Here is an example kernel page: https://huggingface.co/kernels/sgl-project/sgl-flash-attn3. This kernel has the "model" repo type, hence, we're reaching out.

Support for this new repo type has been shipped to the kernels library and we have dropped support for loading the kernels with the "model" repo type in our latest release. Users, however, will be able to use this repo but with an older version of kernels.

To help you through this migration, we have prepared the following steps:

1. Please request access to the "kernel" repo type for your username/org (Settings -> Account)
2. Rebuild your library with the latest kernels version
3. Use the migration tool https://github.com/huggingface/migrate-to-kernel-repo-tool or hugging face cli to create a new kernel repo and upload to the new repo type
4. Ensure downstream users are using v0.14.0 (in pre release) at the moment uv add --reinstall --pre kernels to use the new repo type

Let us know if you have any questions or need further assistance! Thank you for being a early user of the kernels library! 🤗