Spaces:

---

kernels-community

/

README

Like 9

Running

Hi Im DucViet,

I'm making encoder - decoder model like bart and florence2 faster on CPU using conitnous batching and paged attention . I find out vLLM CPU kernels is great but to vendor it in my repo, I want to make kernel as public for other can use. CPU flash-attn2 kernels in kernel-community is only build for amx cpu.

https://github.com/ducviet00/vllm-cpu-attn
https://github.com/ducviet00/fast-encdec