Code

Merging the train-olmo-large branch#599

Code

Merged

dirkgr merged 53 commits intomainallenai/OLMo:mainfrom train-olmo-largeallenai/OLMo:train-olmo-largeCopy head branch name to clipboardMay 30, 2024

Commits

Commits on May 6, 2024

• Llamaish7 config, finally

  dirkgr

  committed
  351f431

  Copy full SHA for 351f431

Commits on May 7, 2024

• Create way more detailed artifacts

  dirkgr

  committed
  4d3afa6

  Copy full SHA for 4d3afa6

Commits on May 8, 2024

• Plan C: backtrack, cut LR in half

  epwalsh

  committed
  732b397

  Copy full SHA for 732b397

• fix branch

  epwalsh

  committed
  59034d6

  Copy full SHA for 59034d6

• pile llamaish

  AkshitaB

  committed
  7015172

  Copy full SHA for 7015172

• try on jupiter

  AkshitaB

  committed
  26ba864

  Copy full SHA for 26ba864

• Resume the detailed llamaish run

  dirkgr

  committed
  453fed1

  Copy full SHA for 453fed1

• Merge branch 'train-olmo-large' of https://github.com/allenai/LLM into train-olmo-large

  dirkgr

  committed
  612934b

  Copy full SHA for 612934b

• fix filename

  AkshitaB

  committed
  3ce4ec9

  Copy full SHA for 3ce4ec9

• run on pluto

  AkshitaB

  committed
  05b8f39

  Copy full SHA for 05b8f39

• whitelist nodes

  epwalsh

  committed
  391a537

  Copy full SHA for 391a537

• lumi config

  AkshitaB

  committed
  c541af1

  Copy full SHA for c541af1

• lumi script

  AkshitaB

  committed
  4095258

  Copy full SHA for 4095258

Commits on May 9, 2024

• jupiter

  AkshitaB

  committed
  92496f4

  Copy full SHA for 92496f4

• increase sync timeout

  AkshitaB

  committed
  c1816db

  Copy full SHA for c1816db

• load from path

  AkshitaB

  committed
  d37125a

  Copy full SHA for d37125a

• fix

  AkshitaB

  committed
  45e407b

  Copy full SHA for 45e407b

Commits on May 10, 2024

• update config for restart

  epwalsh

  committed
  420d6dd

  Copy full SHA for 420d6dd

Commits on May 13, 2024

• update config

  epwalsh

  committed
  05ad705

  Copy full SHA for 05ad705

Commits on May 14, 2024

• Try stop wandb from flaking at cancel

  

  2015aroras

  authored and

  dirkgr

  committed
  06fc94f

  Copy full SHA for 06fc94f

Commits on May 16, 2024

• Configure mitchish7 to continue (with the datafix)

  dirkgr

  committed
  1e603a8

  Copy full SHA for 1e603a8

• I am normal.

  dirkgr

  committed
  52cfe84

  Copy full SHA for 52cfe84

• This is supposed to run on Jupiter.

  dirkgr

  committed
  5c8b115

  Copy full SHA for 5c8b115

Commits on May 18, 2024

• Config for a restart with constant LR

  dirkgr

  committed
  45acb65

  Copy full SHA for 45acb65

• Fix OLMo-core

  dirkgr

  committed
  b394ac2

  Copy full SHA for b394ac2

Commits on May 20, 2024

• Need to specify nodes

  dirkgr

  committed
  5f14392

  Copy full SHA for 5f14392

• Fix OLMo-core

  dirkgr

  committed
  b8b4368

  Copy full SHA for b8b4368

• New nodes

  dirkgr

  committed
  7a6d39d

  Copy full SHA for 7a6d39d

Commits on May 21, 2024

• Continue running the 70B on Jupiter

  dirkgr

  committed
  a03bbf2

  Copy full SHA for a03bbf2

• Run in the proper workspace

  dirkgr

  committed
  3f5f019

  Copy full SHA for 3f5f019

• Run as me

  dirkgr

  committed
  8fc56c1

  Copy full SHA for 8fc56c1

• Run preemptible

  dirkgr

  committed
  8d74d6c

  Copy full SHA for 8d74d6c

• Multiple clusters

  dirkgr

  committed
  6f8a909

  Copy full SHA for 6f8a909

Commits on May 29, 2024

• Formatting

  dirkgr

  committed
  96b1bcb

  Copy full SHA for 96b1bcb

• Improve the script that dry-runs the dataloader

  dirkgr

  committed
  a7e6588

  Copy full SHA for a7e6588

• Off by one

  dirkgr

  committed
  44f3c2d

  Copy full SHA for 44f3c2d

• Save in torch format

  dirkgr

  committed
  613715c

  Copy full SHA for 613715c

• Revert "Save in torch format"

  Show description for 1c22396
  dirkgr

  committed
  1c22396

  Copy full SHA for 1c22396

• Reapply "Save in torch format"

  Show description for 055eff7
  dirkgr

  committed
  055eff7

  Copy full SHA for 055eff7

• Torch doesn't have an uint16 type.

  dirkgr

  committed
  446e90b

  Copy full SHA for 446e90b

• Merge remote-tracking branch 'origin/main' into train-olmo-large

  Show description for 9c809e6
  dirkgr

  committed
  9c809e6

  Copy full SHA for 9c809e6

• Silence warning

  dirkgr

  committed
  99c71f7

  Copy full SHA for 99c71f7

• Silence another warning

  dirkgr

  committed
  bed4324

  Copy full SHA for bed4324

• isort

  dirkgr

  committed
  4beb980

  Copy full SHA for 4beb980

• Run preemptible

  dirkgr

  committed
  60d90f4

  Copy full SHA for 60d90f4

• More clusters

  dirkgr

  committed
  d5912f9

  Copy full SHA for d5912f9

• preemptible

  dirkgr

  committed
  16fad9d

  Copy full SHA for 16fad9d

Commits on May 30, 2024

• Remove usages of Auto* methods in hf_olmo tests

  2015aroras

  committed
  9383d29

  Copy full SHA for 9383d29

• Make test_config_save save to temp file

  2015aroras

  committed
  86f7f97

  Copy full SHA for 86f7f97

• Run ruff

  2015aroras

  committed
  5b98858

  Copy full SHA for 5b98858

• This should respect the default.

  dirkgr

  committed
  f7f0f88

  Copy full SHA for f7f0f88

• Merge remote-tracking branch 'origin/shanea/fix-hf-olmo-tests' into train-olmo-large

  dirkgr

  committed
  20f5213

  Copy full SHA for 20f5213

• Merge branch 'main' into train-olmo-large

  dirkgr

  authored
  cb4af89

  Copy full SHA for cb4af89