Kernels:
Trusted publisher
| library_name: kernels | |
| license: bsd-3-clause | |
| > [!WARNING] | |
| > These rotary kernels are based on the old [CUDA implementations](https://github.com/Dao-AILab/flash-attention/tree/2715c53932c28e81c15ad4d1690639b77ddda6c1/csrc/rotary) in the [FlashAttention repository](https://github.com/Dao-AILab/flash-attention). These kernels are no longer maintained upstream. | |
| This is the repository card of kernels-community/rotary that has been pushed on the Hub. It was built to be used with the [`kernels` library](https://github.com/huggingface/kernels). This card was automatically generated. | |
| ## How to use | |
| ```python | |
| # make sure `kernels` is installed: `pip install -U kernels` | |
| from kernels import get_kernel | |
| kernel_module = get_kernel("kernels-community/rotary", version=3) | |
| apply_rotary = kernel_module.apply_rotary | |
| apply_rotary(...) | |
| ``` | |
| ## Available functions | |
| - `apply_rotary` | |
| - `apply_rotary_transformers` | |
| ## Available layers | |
| - `apply_rotary_transformers` | |
| ## Benchmarks | |
| Benchmarking script is available for this kernel. Run `kernels benchmark kernels-community/rotary --version 3`. | |