Swapped naive dot product attention for flash attention - #24

Open

Matt Smith (usryokousha) wants to merge 7 commits into

microsoft:mainfrom

usryokousha:efficient

Commits on Mar 31, 2023

incorporated fast attention into attention
Matthew Smith
committed
37b64d4

Commits on Apr 24, 2023

Commits on May 9, 2023

Masks are now optional, and not created. Fixes required to support FlashAttention (e.g. no mask, fp/bf16)
Mike Ranzinger (mranzinger)
committed
29c6ead

Commits on May 17, 2023

Merge pull request #1 from mranzinger/efficient
Matt Smith (usryokousha)
authored
dd69dcb