Code

kernel: added simple MLA kernel#396

Code

Merged

guocuimi merged 10 commits intomainvectorch-ai/ScaleLLM:mainfrom mlavectorch-ai/ScaleLLM:mlaCopy head branch name to clipboardFeb 8, 2025

+1,064-4

Commits

Commits on Feb 5, 2025

• kernel: added MLA kernel

  guocuimi

  committed
  4d424c0

  Copy full SHA for 4d424c0

• added max dynamic smem size for different sm

  guocuimi

  committed
  f1c4b86

  Copy full SHA for f1c4b86

• added mla_ref implementation

  guocuimi

  committed
  b46120d

  Copy full SHA for b46120d

Commits on Feb 8, 2025

• added mla_ref implementation

  guocuimi

  committed
  613821e

  Copy full SHA for 613821e

• simple Q*K^T*V to profile huge head_dim perf

  guocuimi

  committed
  b0e5498

  Copy full SHA for b0e5498

• update

  guocuimi

  committed
  13140c4

  Copy full SHA for 13140c4

• added qk_rope part

  guocuimi

  committed
  619b092

  Copy full SHA for 619b092

• rename

  guocuimi

  committed
  624f55a

  Copy full SHA for 624f55a

• added softmax support

  guocuimi

  committed
  e729ea9

  Copy full SHA for e729ea9

• Merge branch 'main' into mla

  guocuimi

  authored
  3d31c73

  Copy full SHA for 3d31c73