litgpt
litgpt copied to clipboard
Add FlashAttention v3 support
This adds FlashAttention v3 support for H100 and H800 GPUs.
We should wait until they release the full code with bf16 supported and it is tested on an H100 machine.
We should wait until they release the full code with bf16 supported and it is tested on an H100 machine.
Has it been done? Maybe it's better to keep it in a Draft mode till all the testing is completed?
Maybe it's better to keep it in a Draft mode till all the testing is completed?
Sure, I can move that to draft mode. But like I said in the original text,
We should wait until they release the full code with bf16 supported and it is tested on an H100 machine.