Code

Several fixes to unblock CI#3047

Code

Merged

jeffra merged 34 commits intomastermicrosoft/DeepSpeed:masterfrom loadams/update-megatronmicrosoft/DeepSpeed:loadams/update-megatronCopy head branch name to clipboardMar 21, 2023

+74-72

Commits

Commits on Mar 16, 2023

• Update a few workflows to use torch 1.13.1 for now

  loadams

  committed
  c20f881

  Copy full SHA for c20f881

• Pin torch version on two other workflows

  loadams

  committed
  4d94c0b

  Copy full SHA for 4d94c0b

• Accelerate works when reverting to 1.13.1, checking if we use latest accelerate

  loadams

  committed
  107e25a

  Copy full SHA for 107e25a

• Update --torch-ver in the torch-latest-v100

  loadams

  committed
  be294c6

  Copy full SHA for be294c6

• Update fixes for pytorch lightning test

  loadams

  committed
  2efdb9f

  Copy full SHA for 2efdb9f

• Fix formatting

  loadams

  committed
  e6494a3

  Copy full SHA for e6494a3

• Bump CUDA version to 11.7 for torch latest

  loadams

  committed
  6c18aa6

  Copy full SHA for 6c18aa6

Commits on Mar 17, 2023

• Update Megatron version to latest

  loadams

  committed
  6274963

  Copy full SHA for 6274963

• Fix outdated flag

  loadams

  committed
  291edb5

  Copy full SHA for 291edb5

• switch to testing MDS main branch

  jeffra

  committed
  05ac2fa

  Copy full SHA for 05ac2fa

• Switch name from GPT2Model to GPTModel

  loadams

  committed
  34bc4b1

  Copy full SHA for 34bc4b1

• Merge branch 'loadams/update-megatron' of https://github.com/microsoft/DeepSpeed into loadams/update-megatron

  loadams

  committed
  12e12ba

  Copy full SHA for 12e12ba

• Formatting issue

  loadams

  committed
  0daed47

  Copy full SHA for 0daed47

• Merge branch 'master' into loadams/update-megatron

  loadams

  authored
  c335088

  Copy full SHA for c335088

• Pin to previous accelerate version

  loadams

  committed
  cec22fb

  Copy full SHA for cec22fb

• Merge branch 'loadams/update-megatron' of https://github.com/microsoft/DeepSpeed into loadams/update-megatron

  loadams

  committed
  7f75fc3

  Copy full SHA for 7f75fc3

• Merge convlits

  loadams

  committed
  df47f3a

  Copy full SHA for df47f3a

• Revert "Merge convlits"

  Show description for 8f94bec
  loadams

  committed
  8f94bec

  Copy full SHA for 8f94bec

• Add comment clarifying pinning pytorch-lightning and add microbatchsize

  loadams

  committed
  ec257b5

  Copy full SHA for ec257b5

• formatting

  loadams

  committed
  0f676a6

  Copy full SHA for 0f676a6

Commits on Mar 18, 2023

• add skip for megatron tests

  mrwyattii

  committed
  eb88ce5

  Copy full SHA for eb88ce5

• Remove changes for micro batch

  loadams

  committed
  ee8c168

  Copy full SHA for ee8c168

Commits on Mar 20, 2023

• skip more tests with latest torch

  mrwyattii

  committed
  ba2ee86

  Copy full SHA for ba2ee86

• Revert megatron-lm import change when changing version

  loadams

  committed
  9cf95a6

  Copy full SHA for 9cf95a6

• Update nv-accelerate workflow to use latest since they added cpuadam support

  loadams

  committed
  d1394e7

  Copy full SHA for d1394e7

• Test with older transformers version

  loadams

  committed
  364ab04

  Copy full SHA for 364ab04

• Remove fix to previous git commit as it fails with torch > 2.

  loadams

  committed
  f16bdcf

  Copy full SHA for f16bdcf

• Revert "Test with older transformers version"

  Show description for 09ef360
  loadams

  committed
  09ef360

  Copy full SHA for 09ef360

Commits on Mar 21, 2023

• revert change updating flag for new megatron-lm

  loadams

  committed
  b7c902a

  Copy full SHA for b7c902a

• Formatting

  loadams

  committed
  1c0b104

  Copy full SHA for 1c0b104

• revert changes to megatron_model file

  loadams

  committed
  71c4f3f

  Copy full SHA for 71c4f3f

• Merge branch 'master' into loadams/update-megatron

  jeffra

  authored
  72c5315

  Copy full SHA for 72c5315

• skip pp tests on p40, add pytest defaults

  jeffra

  committed
  7fd99ea

  Copy full SHA for 7fd99ea

• fix typos

  jeffra

  committed
  0f744d3

  Copy full SHA for 0f744d3