Commits

Breadcrumbs

History for

onmain

User selector

All users

Datepicker

All time

Commit History

Commits on Jul 11, 2023

• Merge branch 'main' into zhuoliny/flamingo-dev

  jaredcasper

  committed
  e5fdfd7

  Copy full SHA for e5fdfd7

Commits on Jun 25, 2023

• code refactor + packing text and img in to a single bin file

  kvareddy

  committed
  efd8f78

  Copy full SHA for efd8f78

Commits on May 31, 2023

• Convert bert to use config, some cleanup of module.

  jaredcasper

  committed
  02fffd2

  Copy full SHA for 02fffd2

Commits on Mar 23, 2023

• Move pipeline parallel functionality into core with associated changes.

  jaredcasper

  committed
  3c92fa9

  Copy full SHA for 3c92fa9

Commits on Sep 24, 2022

• Alias core.parallel_state as mpu and use it throughout code. RIP mpu.

  jaredcasper

  committed
  5942af9

  Copy full SHA for 5942af9

Commits on Sep 23, 2022

• Clean up licensing.

  Show description for fabad46
  jaredcasper

  committed
  fabad46

  Copy full SHA for fabad46

Commits on Aug 10, 2022

• Timing levels

  

  shoeybi

  authored and

  jaredcasper

  committed
  77efccc

  Copy full SHA for 77efccc

Commits on Jul 30, 2021

• Support for pipeline parallelism in T5 model

  Show description for 46c74b4
  deepakn94

  committed
  46c74b4

  Copy full SHA for 46c74b4

Commits on Apr 16, 2021

• Integrate code from t5_main into existing code.

  jaredcasper

  committed
  48a5e0d

  Copy full SHA for 48a5e0d

Commits on Mar 24, 2021

• pipeline code simplification

  

  kvareddy

  authored and

  jaredcasper

  committed
  3b91262

  Copy full SHA for 3b91262

Commits on Feb 9, 2021

• Interleaved pipeline execution and code refactoring

  Show description for dd88906
  deepakn94

  committed
  dd88906

  Copy full SHA for dd88906

Commits on Jan 25, 2021

• Adding option to remove the binary head for BERT

  

  shoeybi

  authored and

  jaredcasper

  committed
  dcff1ac

  Copy full SHA for dcff1ac

Commits on Dec 19, 2020

• Fix TensorBoard writes

  

  shoeybi

  authored and

  deepakn94

  committed
  b81cad6

  Copy full SHA for b81cad6

Commits on Nov 12, 2020

• Move division of loss tensor by number of microbatches to training.py

  deepakn94

  committed
  c671de3

  Copy full SHA for c671de3

• Refactor code according to Jared's comments: move pipelining and non-pipelining training loops into separate methods

  Show description for 1979c24
  deepakn94

  committed
  1979c24

  Copy full SHA for 1979c24

• Divide gradient by number of microbatches in minibatch

  deepakn94

  committed
  3d7194c

  Copy full SHA for 3d7194c

• Intra-layer MP -> Tensor MP, Inter-layer MP -> Pipeline MP

  deepakn94

  committed
  52a5f2f

  Copy full SHA for 52a5f2f

• Pipeline parallelism implementation with periodic full-pipeline syncs

  Show description for 7abd3e9
  deepakn94

  committed
  7abd3e9

  Copy full SHA for 7abd3e9

Commits on Jul 7, 2020

• Addressed Jared's comments

  kantneel

  committed
  8d7f508

  Copy full SHA for 8d7f508

Commits on Jun 26, 2020

• Merge branch 'master' into ict-merge

  kantneel

  committed
  ac967fa

  Copy full SHA for ac967fa

Commits on Jun 24, 2020

• More ict_merge changes and interactive testing

  kantneel

  committed
  3354081

  Copy full SHA for 3354081

Commits on Jun 5, 2020

• addressed jareds comments

  shoeybi

  committed
  197c132

  Copy full SHA for 197c132

• added fp16 lm cross entropy to bert

  shoeybi

  committed
  7802200

  Copy full SHA for 7802200

Commits on May 3, 2020

• Move get_train_val... to dataset_utils

  kantneel

  committed
  16a64c4

  Copy full SHA for 16a64c4

Commits on Apr 16, 2020

• changed licence 2019 to 2020

  shoeybi

  committed
  463d125

  Copy full SHA for 463d125

Commits on Apr 13, 2020

• data loading for BERT and GPT cleaned up

  shoeybi

  committed
  2e38461

  Copy full SHA for 2e38461

Commits on Mar 31, 2020

• arguments in the model refactored

  shoeybi

  committed
  deffcb6

  Copy full SHA for deffcb6

Commits on Mar 30, 2020

• removed building tokenizer from bert dataset

  shoeybi

  committed
  dedb2ef

  Copy full SHA for dedb2ef

• both bert and gpt2 tested and working

  shoeybi

  committed
  1788c91

  Copy full SHA for 1788c91

Commits on Mar 29, 2020

• refactored pretrain-bert

  shoeybi

  committed
  3366a5b

  Copy full SHA for 3366a5b

• refactored training

  shoeybi

  committed
  27e14f8

  Copy full SHA for 27e14f8

• utils partially refactored and tested

  shoeybi

  committed
  3f58649

  Copy full SHA for 3f58649

Commits on Mar 28, 2020

• working on utils

  shoeybi

  committed
  5050203

  Copy full SHA for 5050203

• arguments.py refactored

  shoeybi

  committed
  94e2ca5

  Copy full SHA for 94e2ca5

Commits on Mar 27, 2020

• working on bert

  shoeybi

  committed
  599e959

  Copy full SHA for 599e959

Pagination

Previous

Next