Code

Add flash-attn#41

Code

Merged

RaymondLi0 merged 6 commits intomulti-query-attentionbigcode-project/Megatron-LM:multi-query-attentionfrom flash-attentionbigcode-project/Megatron-LM:flash-attentionCopy head branch name to clipboardMar 24, 2023

+113-5

Commits

Commits on Mar 22, 2023

• add flash-attn

  RaymondLi0

  committed
  7d5154f

  Copy full SHA for 7d5154f

Commits on Mar 23, 2023

• flash-attn: assert that alibi is not used

  RaymondLi0

  committed
  118f0a8

  Copy full SHA for 118f0a8

• fix import

  RaymondLi0

  committed
  d50a89b

  Copy full SHA for d50a89b

• update readme

  RaymondLi0

  committed
  61fe86d

  Copy full SHA for 61fe86d

Commits on Mar 24, 2023

• raise if using flash-attn with selective recomputation, swap if/else

  RaymondLi0

  committed
  f5019c8

  Copy full SHA for f5019c8

• change back to warning

  RaymondLi0

  committed
  0ff5746

  Copy full SHA for 0ff5746