Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Jul 16, 2019

• remove cuda version check from setup.pu

  

  crazyfrogspb

  committed
  339c18e

  Copy full SHA for 339c18e

Commits on Jul 12, 2019

• [sbn update] (#384)

  Show description for 574fe24
  

  jjsjann123

  authored and

  mcarilli

  committed
  574fe24

  Copy full SHA for 574fe24

• Add missing semicolon. (#390)

  Show description for 80e0143
  

  ezyang

  authored and

  mcarilli

  committed
  80e0143

  Copy full SHA for 80e0143

Commits on Jul 3, 2019

• Pulling in deprecation warning changes

  definitelynotmcarilli

  committed
  665b2dd

  Copy full SHA for 665b2dd

• Bumping container version for test

  definitelynotmcarilli

  committed
  d352d44

  Copy full SHA for d352d44

• Remove deprecated Type.h

  definitelynotmcarilli

  committed
  816813f

  Copy full SHA for 816813f

• Fix for #181 (add dot to CAST list)

  Show description for d627172
  

  mcarilli

  authored
  d627172

  Copy full SHA for d627172

Commits on Jul 2, 2019

• Update scaler.py

  

  mcarilli

  authored
  14fe2b6

  Copy full SHA for 14fe2b6

Commits on Jun 28, 2019

• Merge pull request #383 from NVIDIA/lamb_add_fp16_support_update_term

  Show description for ff6b8bb
  thorjohnsen

  authored
  ff6b8bb

  Copy full SHA for ff6b8bb

• Add support for fp16 update term (new UPD_T typename in template)

  thorjohnsen

  committed
  3aeea0d

  Copy full SHA for 3aeea0d

Commits on Jun 27, 2019

• Clarifying documentation on gradient accumulation

  definitelynotmcarilli

  committed
  18f2eae

  Copy full SHA for 18f2eae

Commits on Jun 25, 2019

• grid_sample should be fp32 for now

  definitelynotmcarilli

  committed
  656d14b

  Copy full SHA for 656d14b

Commits on Jun 24, 2019

• Merge branch 'master' of https://github.com/NVIDIA/apex

  definitelynotmcarilli

  committed
  9ce8017

  Copy full SHA for 9ce8017

• Docstring for multiple losses

  definitelynotmcarilli

  committed
  f855756

  Copy full SHA for f855756

• Update README.md

  

  mcarilli

  authored
  f17cd95

  Copy full SHA for f17cd95

• Updating gradient accumulation guidance

  definitelynotmcarilli

  committed
  ca35aa7

  Copy full SHA for ca35aa7

Commits on Jun 21, 2019

• Make main_amp.py more profiling-friendly

  definitelynotmcarilli

  committed
  f29b3f8

  Copy full SHA for f29b3f8

• Don't need to blacklist mean for pytorch >= 1.1

  definitelynotmcarilli

  committed
  4b9858e

  Copy full SHA for 4b9858e

Commits on Jun 20, 2019

• Fix end-of-epoch with record_stream

  definitelynotmcarilli

  committed
  90e5b05

  Copy full SHA for 90e5b05

Commits on Jun 19, 2019

• Merge branch 'master' of https://github.com/NVIDIA/apex

  definitelynotmcarilli

  committed
  1ccaaf4

  Copy full SHA for 1ccaaf4

• Fix for https://github.com/NVIDIA/apex/issues/361

  definitelynotmcarilli

  committed
  68c850d

  Copy full SHA for 68c850d

Commits on Jun 18, 2019

• Give custom to method a higher priority (#364)

  

  mcarilli

  authored
  cd6e46c

  Copy full SHA for cd6e46c

• Fix rare caching allocator race condition in imagenet prefetcher

  definitelynotmcarilli

  committed
  d5e2bb4

  Copy full SHA for d5e2bb4

Commits on Jun 17, 2019

• More helpful message for unexpected opt_level

  definitelynotmcarilli

  committed
  c3bcf18

  Copy full SHA for c3bcf18

Commits on Jun 14, 2019

• Merge branch 'master' of https://github.com/NVIDIA/apex

  definitelynotmcarilli

  committed
  a9f5f71

  Copy full SHA for a9f5f71

• Removing gradient_average_split_factor

  definitelynotmcarilli

  committed
  41c9851

  Copy full SHA for 41c9851

• Separate LDG/STG from compute loop (#359)

  

  thorjohnsen

  authored and

  mcarilli

  committed
  121a250

  Copy full SHA for 121a250

• Adding delay_overflow_check=False ninja control point

  definitelynotmcarilli

  committed
  4a9c2a5

  Copy full SHA for 4a9c2a5

Commits on Jun 13, 2019

• disable_allreduce -> _disable_allreduce

  definitelynotmcarilli

  committed
  ae7f0de

  Copy full SHA for ae7f0de

• Add option to turn on/off allreduce in DDP (useful for gradient accumulation) (#356)

  

  thorjohnsen

  authored and

  mcarilli

  committed
  1c2ba89

  Copy full SHA for 1c2ba89

Commits on Jun 11, 2019

• Allow multi_tensor_lamb to update fp16 params

  definitelynotmcarilli

  committed
  47e3367

  Copy full SHA for 47e3367

Commits on Jun 7, 2019

• Fix for https://github.com/NVIDIA/apex/issues/344

  definitelynotmcarilli

  committed
  0466713

  Copy full SHA for 0466713

Commits on Jun 6, 2019

• Making O1 the default opt level

  definitelynotmcarilli

  committed
  1dca16c

  Copy full SHA for 1dca16c

Commits on Jun 4, 2019

• Adding min_loss_scale and max_loss_scale arguments to amp.initialize

  definitelynotmcarilli

  committed
  b82c6bd

  Copy full SHA for b82c6bd

Commits on May 31, 2019

• Multi tensor lamb optimizer (#334)

  Show description for 8be5b6b
  

  thorjohnsen

  authored and

  mcarilli

  committed
  8be5b6b

  Copy full SHA for 8be5b6b