Code

[usability] Accelerate Support#936

Code

Draft

wheresmyhair wants to merge 15 commits intomainOptimalScale/LMFlow:mainfrom lmflow-nightlyOptimalScale/LMFlow:lmflow-nightlyCopy head branch name to clipboard

+1,853-1,528

Commits

Commits on Feb 22, 2025

• [usability] accelerate support initial commit

  wheresmyhair

  committed
  beba6ef

  Copy full SHA for beba6ef

Commits on Feb 27, 2025

• [usability] accelerate support - scripts update

  wheresmyhair

  committed
  ef083b6

  Copy full SHA for ef083b6

Commits on Mar 4, 2025

• change ruff target ver

  wheresmyhair

  committed
  f7ab42b

  Copy full SHA for f7ab42b

• separate lisa and custom optims

  wheresmyhair

  committed
  19586d1

  Copy full SHA for 19586d1

• add decorator for base methods

  wheresmyhair

  committed
  e12e25a

  Copy full SHA for e12e25a

• encoder_decoder model temporarily disabled

  wheresmyhair

  committed
  edc9236

  Copy full SHA for edc9236

• future support for trainer

  wheresmyhair

  committed
  05e26a9

  Copy full SHA for 05e26a9

• fix fsdp cfg warp policy

  wheresmyhair

  committed
  593f44b

  Copy full SHA for 593f44b

• add note for single gpu

  wheresmyhair

  committed
  16eefe0

  Copy full SHA for 16eefe0

• add accelerate singlegpu config

  wheresmyhair

  committed
  29f10d8

  Copy full SHA for 29f10d8

Commits on Mar 6, 2025

• fix qlora script

  wheresmyhair

  committed
  8194d36

  Copy full SHA for 8194d36

• [usability] simplify shell scripts

  wheresmyhair

  committed
  0faa453

  Copy full SHA for 0faa453

• [usability] add hint for flash_attn

  wheresmyhair

  committed
  43ff307

  Copy full SHA for 43ff307

• [fix] fix create_customized_optimizer arg

  wheresmyhair

  committed
  0ba9b60

  Copy full SHA for 0ba9b60

• [usability] simplify dpov2 script

  wheresmyhair

  committed
  16d7af6

  Copy full SHA for 16d7af6