Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Dec 9, 2021

• Add fused mixed precision lamb optimizer. (#1237)

  Show description for 3c8f516
  kevinstephano

  authored
  3c8f516

  Copy full SHA for 3c8f516

Commits on Jun 15, 2021

• update fusedadam docstring

  FDecaYed

  committed
  ab520f8

  Copy full SHA for ab520f8

Commits on Apr 15, 2021

• Add unit tests for Fused NovoGrad (#1065)

  Show description for 59d2f7a
  sudhakarsingh27

  authored
  59d2f7a

  Copy full SHA for 59d2f7a

Commits on Aug 6, 2020

• sgd supports zero-grad (#926)

  

  ngimel

  authored
  85b1783

  Copy full SHA for 85b1783

Commits on Aug 5, 2020

• set device guard for multi tensor optimizer implementations (#927)

  Show description for 274cc06
  

  ngimel

  authored
  274cc06

  Copy full SHA for 274cc06

Commits on May 29, 2020

• Merge pull request #851 from kexinyu/master

  Show description for 36c9e90
  kexinyu

  authored
  36c9e90

  Copy full SHA for 36c9e90

Commits on May 28, 2020

• Fixed a typo (#856)

  xeonvs

  authored
  87aca22

  Copy full SHA for 87aca22

Commits on May 21, 2020

• make fused LAMB async

  kexinyu

  committed
  f54cc1c

  Copy full SHA for f54cc1c

Commits on May 19, 2020

• Merge pull request #819 from kexinyu/master

  Show description for 8abb690
  kexinyu

  authored
  8abb690

  Copy full SHA for 8abb690

Commits on May 14, 2020

• Add FusedAdagrad (#822)

  ajtulloch

  authored
  3bae8c8

  Copy full SHA for 3bae8c8

Commits on May 13, 2020

• Fixes flake8 --select W605 test warnings (#829)

  Show description for 9165b27
  asears

  authored
  9165b27

  Copy full SHA for 9165b27

Commits on May 2, 2020

• initialize on device

  kexinyu

  committed
  bd6e66d

  Copy full SHA for bd6e66d

• initialize with tensor

  kexinyu

  committed
  9033ad5

  Copy full SHA for 9033ad5

• save a sync when calculating global gradient norm

  kexinyu

  committed
  f560bd0

  Copy full SHA for f560bd0

Commits on May 1, 2020

• make use_nvlamb a class attribute for FusedLAMB

  kexinyu

  committed
  85e4af7

  Copy full SHA for 85e4af7

• add import

  kexinyu

  committed
  3fd3e2c

  Copy full SHA for 3fd3e2c

Commits on Apr 28, 2020

• LAMB: global grad clipping & more flexibility in adaptive lr

  kexinyu

  committed
  5b30011

  Copy full SHA for 5b30011

Commits on Dec 18, 2019

• fix beta2 => beta1 (#661)

  

  bonlime

  authored and

  mcarilli

  committed
  0ce8ad3

  Copy full SHA for 0ce8ad3

Commits on Oct 4, 2019

• move previous fused_adam and fp16_optimizer to contrib (#517)

  Show description for 1904e48
  

  FDecaYed

  authored and

  mcarilli

  committed
  1904e48

  Copy full SHA for 1904e48

Commits on Sep 3, 2019

• Fix issues in fused_dam (#469)

  Show description for 7fa7492
  

  FDecaYed

  authored and

  mcarilli

  committed
  7fa7492

  Copy full SHA for 7fa7492

Commits on Aug 30, 2019

• [novograd] move exp_avg_sq to param device in load_state_dict (#459)

  

  FDecaYed

  authored and

  mcarilli

  committed
  53eae19

  Copy full SHA for 53eae19

Commits on Aug 27, 2019

• Docstring updates

  definitelynotmcarilli

  committed
  b9e5d37

  Copy full SHA for b9e5d37

• Docstring updates

  definitelynotmcarilli

  committed
  17e8a55

  Copy full SHA for 17e8a55

• Updating docstrings for fused optimizers

  definitelynotmcarilli

  committed
  427e82c

  Copy full SHA for 427e82c

Commits on Aug 16, 2019

• clean up variance options support by all fused optimizers:

  Show description for 18062b6
  FDecaYed

  committed
  18062b6

  Copy full SHA for 18062b6

• fix novograd init overflow

  FDecaYed

  committed
  7a219aa

  Copy full SHA for 7a219aa

• add fused lamb, put lamb kernels into one file

  FDecaYed

  committed
  c8f9cce

  Copy full SHA for c8f9cce

Commits on Aug 13, 2019

• Reverse to Fused* naming, clean up accordingly:

  Show description for 007c594
  FDecaYed

  committed
  007c594

  Copy full SHA for 007c594

• add multi-precision support for novograd, clean import

  FDecaYed

  committed
  37a1c12

  Copy full SHA for 37a1c12

Commits on Aug 12, 2019

• converge fused_sgd and sgd code(dtype support, fused kernel, wd_after)

  FDecaYed

  committed
  8599b85

  Copy full SHA for 8599b85

• keep old fused* name and rename new optimizers without prefix

  FDecaYed

  committed
  adad599

  Copy full SHA for adad599

• Merge branch 'multi_tensor_sgd' into deyuf/fused_optimizer_v2

  FDecaYed

  committed
  4d6ed50

  Copy full SHA for 4d6ed50

Commits on Aug 8, 2019

• initial commit to make fused optimizers compatible with AMP

  FDecaYed

  committed
  690b1f7

  Copy full SHA for 690b1f7

Commits on May 27, 2019

• FusedSGD tests passing for all opt_levels

  definitelynotmcarilli

  committed
  848c777

  Copy full SHA for 848c777

Commits on May 23, 2019

• Merging in latest master changes

  definitelynotmcarilli

  committed
  c142714

  Copy full SHA for c142714