Commits

Branch selector

 master

User selector

FDecaYed

Datepicker

All time

Commit History

Commits on Jun 15, 2021

• Merge pull request NVIDIA#1118 from FDecaYed/deyuf/update_adam_string

  Show description for d06404f
  FDecaYed

  authored
  d06404f

  Copy full SHA for d06404f

• update fusedadam docstring

  FDecaYed

  committed
  ab520f8

  Copy full SHA for ab520f8

Commits on Apr 17, 2021

• Adding fast bottleneck implementation into contrib (NVIDIA#1079)

  Show description for 705cba9
  

  FDecaYed

  and

  ptrblck

  authored
  705cba9

  Copy full SHA for 705cba9

Commits on May 1, 2020

• Changes to make xentropysoftmax load/store vectorized when possible: (NVIDIA#725)

  Show description for cf50dc7
  FDecaYed

  authored
  cf50dc7

  Copy full SHA for cf50dc7

Commits on Apr 30, 2020

• enable wider load/store for multi_tensor_apply kernels (NVIDIA#763)

  Show description for 17ee854
  FDecaYed

  authored
  17ee854

  Copy full SHA for 17ee854

• Improvements to apex.mlp (NVIDIA#804)

  Show description for 31aceea
  FDecaYed

  authored
  31aceea

  Copy full SHA for 31aceea

Commits on Apr 22, 2020

• initial commit to add Multilayer Perceptron (MLP) extension (NVIDIA#790)

  FDecaYed

  authored
  71511fa

  Copy full SHA for 71511fa

Commits on Feb 15, 2020

• change include_dirs to abs path (NVIDIA#719)

  FDecaYed

  authored
  50338df

  Copy full SHA for 50338df

Commits on Oct 4, 2019

• move previous fused_adam and fp16_optimizer to contrib (NVIDIA#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 (NVIDIA#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 (NVIDIA#459)

  

  FDecaYed

  authored and

  mcarilli

  committed
  53eae19

  Copy full SHA for 53eae19

Commits on Aug 20, 2019

• add back lamb stage1/2 to amp_C python

  FDecaYed

  committed
  b9f0995

  Copy full SHA for b9f0995

Commits on Aug 17, 2019

• disable breaking test until switch to test against upstream v1.2.0

  FDecaYed

  committed
  f855f85

  Copy full SHA for f855f85

• add back legacy lamb code for backward comptibility now

  FDecaYed

  committed
  2bc766c

  Copy full SHA for 2bc766c

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

• Merge branch 'master' into deyuf/fused_optimizer_v2

  FDecaYed

  committed
  aee5aff

  Copy full SHA for aee5aff

• 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 Mar 8, 2019

• Merge branch 'master' into deyuf/update_norm

  FDecaYed

  authored
  2f0bf59

  Copy full SHA for 2f0bf59

Commits on Mar 7, 2019

• make norm compatible with pytorch version <= 1.0.1

  FDecaYed

  committed
  9949537

  Copy full SHA for 9949537

Commits on Mar 6, 2019

• change to use now great torch.norm

  FDecaYed

  committed
  9466741

  Copy full SHA for 9466741

Commits on Dec 15, 2018

• add unit tests for optimizers/fp16_optimizer

  FDecaYed

  committed
  afc8d1b

  Copy full SHA for afc8d1b

Commits on Dec 11, 2018

• address comment

  FDecaYed

  committed
  b6188fc

  Copy full SHA for b6188fc

• improve backward compatibility

  FDecaYed

  committed
  d665ab9

  Copy full SHA for d665ab9

• better print

  FDecaYed

  committed
  f06fece

  Copy full SHA for f06fece

• remove temporarily added lr warmup

  Show description for c8ca4bf
  FDecaYed

  committed
  c8ca4bf

  Copy full SHA for c8ca4bf

Commits on Dec 7, 2018

• address comments

  FDecaYed

  committed
  4d9dcb5

  Copy full SHA for 4d9dcb5

Commits on Dec 6, 2018

• WIP: improve fused adam

  FDecaYed

  committed
  be42aad

  Copy full SHA for be42aad