Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Jul 17, 2021

• Added more fusion and vectorized kernel for transducer (NVIDIA#1125)

  Show description for 0c2c6ee
  nanz-nv

  authored
  0c2c6ee

  Copy full SHA for 0c2c6ee

• Adds small-batch kernels (NVIDIA#1126)

  yjk21

  authored
  ed71996

  Copy full SHA for ed71996

Commits on Jul 16, 2021

• local_rank fix (NVIDIA#1129)

  Show description for c1378e6
  xwang233

  authored
  c1378e6

  Copy full SHA for c1378e6

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 May 26, 2021

• Distributed LAMB: Clip grads before reduce_scatter/all_reduce (NVIDIA#1099)

  Show description for ebcd7f0
  kexinyu

  authored
  ebcd7f0

  Copy full SHA for ebcd7f0

Commits on May 17, 2021

• compile cublasLt code only for cublas >= 11.0 (NVIDIA#1108)

  Show description for 00c1e56
  

  seryilmaz

  and

  Sukru Eryilmaz

  authored
  00c1e56

  Copy full SHA for 00c1e56

Commits on Apr 19, 2021

• Fix cublasLt context create/destroy overhead in MLP extension (NVIDIA#1083)

  Show description for 082f999
  seryilmaz

  authored
  082f999

  Copy full SHA for 082f999

Commits on Apr 17, 2021

• initial cublaslt support for MLP (NVIDIA#1080)

  Show description for b8be1bc
  

  seryilmaz

  and

  ptrblck

  authored
  b8be1bc

  Copy full SHA for b8be1bc

• remove MIT license (NVIDIA#1081)

  ptrblck

  authored
  b5eb38d

  Copy full SHA for b5eb38d

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

  Show description for 705cba9
  

  FDecaYed

  and

  ptrblck

  authored
  705cba9

  Copy full SHA for 705cba9

Commits on Apr 16, 2021

• adds fmhalib (NVIDIA#1074)

  yjk21

  authored
  5c9b21d

  Copy full SHA for 5c9b21d

Commits on Apr 15, 2021

• Update README.md (NVIDIA#1067)

  Show description for e5f2f67
  jayrodge

  authored
  e5f2f67

  Copy full SHA for e5f2f67

• DistributedFusedLAMB: enable no_copy and add barrier for SHARP (NVIDIA#1075)

  Show description for bb79158
  

  kexinyu

  and

  Kexin Yu

  authored
  bb79158

  Copy full SHA for bb79158

• Add unit tests for Fused NovoGrad (NVIDIA#1065)

  Show description for 59d2f7a
  sudhakarsingh27

  authored
  59d2f7a

  Copy full SHA for 59d2f7a

Commits on Mar 24, 2021

• sync-free Distributed LAMB + parameter reordering (NVIDIA#1055)

  Show description for a651e2c
  

  kexinyu

  and

  Kexin Yu

  authored
  a651e2c

  Copy full SHA for a651e2c

• Initial check-in of the transducer extensions (NVIDIA#1069)

  Show description for d86d1b0
  nanz-nv

  authored
  d86d1b0

  Copy full SHA for d86d1b0

Commits on Feb 23, 2021

• fast layer norm (NVIDIA#1037)

  yjk21

  authored
  e2083df

  Copy full SHA for e2083df

Commits on Feb 10, 2021

• fix import container_abcs issue (NVIDIA#1049)

  Show description for a78ccf0
  ShoufaChen

  authored
  a78ccf0

  Copy full SHA for a78ccf0

Commits on Jan 20, 2021

• cuda rng changes for graph capture with apex MHA (NVIDIA#1025)

  Show description for eefb1ba
  

  seryilmaz

  and

  Sukru Eryilmaz

  authored
  eefb1ba

  Copy full SHA for eefb1ba

Commits on Dec 17, 2020

• Merge pull request NVIDIA#1015 from jpool-nv/patch-1

  Show description for 154c633
  thorjohnsen

  authored
  154c633

  Copy full SHA for 154c633

• Update ASP README to highlight default recipe

  Show description for 56914d4
  jpool-nv

  authored
  56914d4

  Copy full SHA for 56914d4

Commits on Dec 4, 2020

• remove noise pip-version-check noise that hides the outcome of the build (NVIDIA#998)

  stas00

  authored
  8cf5ae6

  Copy full SHA for 8cf5ae6

• Distributed LAMB fixes (NVIDIA#1007)

  Show description for 8a80d47
  

  3 peopleauthored
  8a80d47

  Copy full SHA for 8a80d47

• Seryilmaz/fused dropout softmax (NVIDIA#985)

  Show description for 3fe10b5
  seryilmaz

  authored
  3fe10b5

  Copy full SHA for 3fe10b5

Commits on Dec 2, 2020

• Fix lack of proper loading of best_prec1 from the checkpoint (NVIDIA#1000)

  Show description for 6c186b3
  JanuszL

  authored
  6c186b3

  Copy full SHA for 6c186b3

Commits on Dec 1, 2020

• DistributedFusedAdam Model Parallelism Support (Megatron) (NVIDIA#981)

  Show description for 6b7e77b
  

  3 peopleauthored
  6b7e77b

  Copy full SHA for 6b7e77b

Commits on Oct 19, 2020

• Optimize the sync batchnorm by batching the communication (NVIDIA#980)

  Show description for 8a1ed9e
  lly-zero-one

  authored
  8a1ed9e

  Copy full SHA for 8a1ed9e

Commits on Sep 29, 2020

• use reshape instead of view (NVIDIA#971)

  ptrblck

  authored
  a109f85

  Copy full SHA for a109f85

Commits on Sep 15, 2020

• Merge pull request NVIDIA#959 from a-maci/update-ASP-readme

  Show description for 4a1fa2c
  thorjohnsen

  authored
  4a1fa2c

  Copy full SHA for 4a1fa2c

Commits on Sep 14, 2020

• Merge pull request NVIDIA#5 from a-maci/a-maci-patch-update-asp-readme

  Show description for 48fc613
  a-maci

  authored
  48fc613

  Copy full SHA for 48fc613

• Update README for ASP

  Show description for e3794f4
  a-maci

  authored
  e3794f4

  Copy full SHA for e3794f4

Commits on Aug 15, 2020

• Should pass stricter stride/size checks in pytorch (NVIDIA#942)

  mcarilli

  authored
  4ef930c

  Copy full SHA for 4ef930c

Commits on Aug 10, 2020

• move sm80 code inside MHA (NVIDIA#937)

  Show description for 5d9b5cb
  ptrblck

  authored
  5d9b5cb

  Copy full SHA for 5d9b5cb

Commits on Aug 6, 2020

• sgd supports zero-grad (NVIDIA#926)

  

  ngimel

  authored
  85b1783

  Copy full SHA for 85b1783

Pagination

Previous

Next