Code

Better cuBLAS handle management#1389

Code

Open

ptrendx wants to merge 5 commits intoNVIDIA:mainNVIDIA/TransformerEngine:mainfrom ptrendx:pr_cublas_handle_mgmtptrendx/TransformerEngine:pr_cublas_handle_mgmtCopy head branch name to clipboard

+84-34

Commits

Commits on Jan 2, 2025

• Do not create multiple cublas handle

  Show description for 999879e
  ptrendx

  committed
  999879e

  Copy full SHA for 999879e

• [pre-commit.ci] auto fixes from pre-commit.com hooks

  Show description for 7b868b0
  pre-commit-ci[bot]

  committed
  7b868b0

  Copy full SHA for 7b868b0

Commits on Jan 6, 2025

• Fix for multiple GPUs per thread

  Show description for 529aec2
  ptrendx

  committed
  529aec2

  Copy full SHA for 529aec2

• [pre-commit.ci] auto fixes from pre-commit.com hooks

  Show description for 44c7d70
  pre-commit-ci[bot]

  committed
  44c7d70

  Copy full SHA for 44c7d70

Commits on Jan 7, 2025

• Fix multithreaded execution

  Show description for e62a128
  ptrendx

  committed
  e62a128

  Copy full SHA for e62a128