Code

Replace FlashAttention with xformers#70

Code

Merged

WoosukKwon merged 12 commits intomainvllm-project/vllm:mainfrom xformersvllm-project/vllm:xformersCopy head branch name to clipboardMay 5, 2023

+87-131

Commits

Commits on May 3, 2023

• Fix GPTAttention

  WoosukKwon

  committed
  c653294

  Copy full SHA for c653294

Commits on May 4, 2023

• Merge branch 'main' into xformers

  WoosukKwon

  committed
  49e406a

  Copy full SHA for 49e406a

• Remove OPT and LLaMA attention

  WoosukKwon

  committed
  00c6129

  Copy full SHA for 00c6129

• Replace flash attention with xformers

  WoosukKwon

  committed
  a8afbca

  Copy full SHA for a8afbca

• Remove flash attention in comments

  WoosukKwon

  committed
  4a47179

  Copy full SHA for 4a47179

• Fix attention kernel tests

  WoosukKwon

  committed
  76ad4dc

  Copy full SHA for 76ad4dc

• Minor

  WoosukKwon

  committed
  0330f83

  Copy full SHA for 0330f83

• Make tests robust to precision errors

  WoosukKwon

  committed
  e6bfe3a

  Copy full SHA for e6bfe3a

• Merge branch 'main' into xformers

  WoosukKwon

  committed
  36e9f96

  Copy full SHA for 36e9f96

• Add bfloat16 in kernel unit tests

  WoosukKwon

  committed
  f0c462b

  Copy full SHA for f0c462b

• Update README

  WoosukKwon

  committed
  02ad58b

  Copy full SHA for 02ad58b

• Head size

  WoosukKwon

  committed
  bd7657b

  Copy full SHA for bd7657b