deepseek-ai/flashmla

FlashMLA is an attention kernel library that supports DeepSeek-V4.1 inference on NVIDIA GPUs and Huawei Ascend NPUs.

stars

★ 13.2K

Forks

1.2K

Language

C++

License

MIT