Commits

Breadcrumbs

History for

onmaster

User selector

All users

Datepicker

All time

Commit History

Commits on Oct 27, 2021

• FastLayerNorm compat with autocast (NVIDIA#1203)

  Show description for ae75763
  

  crcrpar

  and

  yjk21

  authored
  ae75763

  Copy full SHA for ae75763

Commits on Oct 2, 2021

• transformer utils (NVIDIA#1181)

  Show description for 365fdc1
  

  5 peopleauthored
  365fdc1

  Copy full SHA for 365fdc1

Commits on Sep 8, 2021

• enable ninja (NVIDIA#1164)

  Show description for 9ce0a10
  crcrpar

  authored
  9ce0a10

  Copy full SHA for 9ce0a10

Commits on Sep 1, 2021

• Seryilmaz/fuse norm into scale (NVIDIA#1149)

  Show description for 4d190db
  

  seryilmaz

  and

  Sukru Eryilmaz

  authored
  4d190db

  Copy full SHA for 4d190db

• Seryilmaz/more cublas lt (NVIDIA#1147)

  Show description for 6af09dd
  

  seryilmaz

  and

  Sukru Eryilmaz

  authored
  6af09dd

  Copy full SHA for 6af09dd

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 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 Apr 16, 2021

• adds fmhalib (NVIDIA#1074)

  yjk21

  authored
  5c9b21d

  Copy full SHA for 5c9b21d

Commits on Mar 24, 2021

• 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 Dec 1, 2020

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

  Show description for 6b7e77b
  

  3 peopleauthored
  6b7e77b

  Copy full SHA for 6b7e77b

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 1, 2020

• Add sm80 for CUDA >= 11 (NVIDIA#925)

  ptrblck

  authored
  5b53121

  Copy full SHA for 5b53121

Commits on Jun 1, 2020

• Add Pyprof removal warnings that point to new repo (NVIDIA#862)

  Show description for 097238f
  

  mcarilli

  and

  definitelynotmcarilli

  authored
  097238f

  Copy full SHA for 097238f

Commits on May 30, 2020

• Make separate apex option for distributed lamb

  thorjohnsen

  committed
  45388d4

  Copy full SHA for 45388d4

• Distributed LAMB optimizer

  thorjohnsen

  committed
  19892f1

  Copy full SHA for 19892f1

Commits on May 29, 2020

• Fuses dropout and softmax in backward pass, add bias support to CPP MHA, add additive mask support, separate Q/K/V parameters (NVIDIA#854)

  Show description for 6c2babf
  

  seryilmaz

  and

  Sukru Eryilmaz

  authored
  6c2babf

  Copy full SHA for 6c2babf

Commits on May 14, 2020

• Add FusedAdagrad (NVIDIA#822)

  ajtulloch

  authored
  3bae8c8

  Copy full SHA for 3bae8c8

Commits on Apr 23, 2020

• CUDAGenerator fix for #36026 (NVIDIA#801)

  Show description for 1f2aa91
  ptrblck

  authored
  1f2aa91

  Copy full SHA for 1f2aa91

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 Mar 23, 2020

• add l2norm source for FusedLAMB

  

  Kexin Yu

  committed
  a3ffb8a

  Copy full SHA for a3ffb8a

Commits on Mar 20, 2020

• extension name fix

  

  Kexin Yu

  committed
  b4c3201

  Copy full SHA for b4c3201

• apex.contrib.optimizers.FuseLamb first commit

  

  Kexin Yu

  committed
  b222ed2

  Copy full SHA for b222ed2

Commits on Mar 11, 2020

• Fix deprecated calls in multihead_attn and ninja build failure (NVIDIA#746)

  Show description for 80b90b9
  ptrblck

  authored
  80b90b9

  Copy full SHA for 80b90b9

Commits on Mar 2, 2020

• Revert "remove gencode from multihead_attn build (NVIDIA#731)"

  Show description for 5633f6d
  ptrblck

  committed
  5633f6d

  Copy full SHA for 5633f6d

Commits on Feb 27, 2020

• NHWC support for multi tensor apply (NVIDIA#732)

  Show description for de6378f
  mcarilli

  authored
  de6378f

  Copy full SHA for de6378f

Commits on Feb 25, 2020

• remove gencode from multihead_attn build (NVIDIA#731)

  ptrblck

  authored
  92b3b9a

  Copy full SHA for 92b3b9a

• remove duplicated multihead_attn install (NVIDIA#729)

  ptrblck

  authored
  5f6b9b0

  Copy full SHA for 5f6b9b0

Commits on Feb 24, 2020

• Change to Multihead Attention to allow Batched GEMMs larger than 64K. (NVIDIA#728)

  Show description for 1733946
  kevinstephano

  authored
  1733946

  Copy full SHA for 1733946

Commits on Feb 15, 2020

• change include_dirs to abs path (NVIDIA#719)

  FDecaYed

  authored
  50338df

  Copy full SHA for 50338df

Commits on Feb 6, 2020

• Add Fast Multihead Attention to APEX Contrib (NVIDIA#697)

  Show description for 3f94528
  kevinstephano

  authored
  3f94528

  Copy full SHA for 3f94528

Commits on Jan 21, 2020

• removing build target sm_70 from bnp (NVIDIA#683)

  

  jjsjann123

  authored and

  ptrblck

  committed
  b66ffc1

  Copy full SHA for b66ffc1

Commits on Jan 8, 2020

• add WAR for pip>=19.3.1 (NVIDIA#652)

  Show description for b5a7c5f
  

  ptrblck

  authored and

  mcarilli

  committed
  b5a7c5f

  Copy full SHA for b5a7c5f

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

Pagination

Previous

Next