You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am Sayak from the Kernels team at Hugging Face. I noticed that this project uses Flash Attention which includes a long build time. We ship pre-built binaries (which provide bit-exact outputs as the upstream) and thereby, we make it easy to use.
Using FA3 on a supported machine is as easy as:
# make sure `kernels` is installed: `pip install -U kernels`fromkernelsimportget_kernelkernel_module=get_kernel("kernels-community/flash-attn3")
flash_attn_func=kernel_module.flash_attn_funcflash_attn_func(...)
Let us know if you'd be interested in this and and we'd be happy to provide a draft of how it would look in your repo.
Hey,
I am Sayak from the Kernels team at Hugging Face. I noticed that this project uses Flash Attention which includes a long build time. We ship pre-built binaries (which provide bit-exact outputs as the upstream) and thereby, we make it easy to use.
Using FA3 on a supported machine is as easy as:
Let us know if you'd be interested in this and and we'd be happy to provide a draft of how it would look in your repo.