Code

Refactor dispatch#38

Code

Open

tenpercent wants to merge 27 commits intoROCm:developROCm/xformers:developfrom tenpercent:refactor-dispatchtenpercent/xformers:refactor-dispatchCopy head branch name to clipboard

+1,449-1,412

Commits

Commits on Dec 4, 2024

• compilable poc

  tenpercent

  committed
  ff203ce

  Copy full SHA for ff203ce

• modify codegen and regenerate the instances to encode dispatch cases as types

  tenpercent

  committed
  6e9c4d3

  Copy full SHA for 6e9c4d3

• encode head dimension dispatch switch as type

  tenpercent

  committed
  adfcf82

  Copy full SHA for adfcf82

Commits on Dec 5, 2024

• enable offload compression

  tenpercent

  committed
  f28928a

  Copy full SHA for f28928a

• clang-format

  tenpercent

  committed
  52c4042

  Copy full SHA for 52c4042

• fix python lints

  tenpercent

  committed
  102a5e3

  Copy full SHA for 102a5e3

• run black

  tenpercent

  committed
  bf52044

  Copy full SHA for bf52044

• run isort

  tenpercent

  committed
  a26f1c1

  Copy full SHA for a26f1c1

• move out common struct

  tenpercent

  committed
  85dc1c0

  Copy full SHA for 85dc1c0

• add missing header

  tenpercent

  committed
  ec724b8

  Copy full SHA for ec724b8

• adjust number of compilation workers to avoid ooms in CI

  tenpercent

  committed
  2f2dca8

  Copy full SHA for 2f2dca8

• roll back run-clang-format

  tenpercent

  committed
  aa2657a

  Copy full SHA for aa2657a

Commits on Dec 6, 2024

• sync wheels_build with facebookresearch

  tenpercent

  committed
  7222125

  Copy full SHA for 7222125

• add missing script for new wheel build

  tenpercent

  committed
  47d80c9

  Copy full SHA for 47d80c9

• use separate runner for wheels and ci

  tenpercent

  committed
  de21a1d

  Copy full SHA for de21a1d

• lower max_jobs for ci build to 80

  tenpercent

  committed
  a2ba3e7

  Copy full SHA for a2ba3e7

• sync .github/actions with facebookresearch

  tenpercent

  committed
  69a92fe

  Copy full SHA for 69a92fe

• add rocm 6.2 for wheel build

  tenpercent

  committed
  3dad112

  Copy full SHA for 3dad112

• bump pytorch to 2.5.1

  tenpercent

  committed
  8cf45ca

  Copy full SHA for 8cf45ca

• do not offload compress prior to rocm 6.2

  tenpercent

  committed
  d576b53

  Copy full SHA for d576b53

• amend version check

  tenpercent

  committed
  b58b5de

  Copy full SHA for b58b5de

• try limiting container memory to avoid oom in ci

  tenpercent

  committed
  1a25871

  Copy full SHA for 1a25871

Commits on Dec 9, 2024

• tweak memory limit for rocm_ci

  tenpercent

  committed
  f16b48a

  Copy full SHA for f16b48a

• Merge remote-tracking branch 'origin/develop' into refactor-dispatch

  tenpercent

  committed
  698506d

  Copy full SHA for 698506d

Commits on Dec 10, 2024

• modify MAX_JOBS

  tenpercent

  committed
  32ad0c0

  Copy full SHA for 32ad0c0

• bump torch wheel to rocm+6.2 stable

  tenpercent

  committed
  1d36a2a

  Copy full SHA for 1d36a2a

Commits on Dec 12, 2024

• Merge remote-tracking branch 'origin/develop' into refactor-dispatch

  tenpercent

  committed
  070897c

  Copy full SHA for 070897c