Commits

Breadcrumbs

History for

onmaster

User selector

All users

Datepicker

All time

Commit History

Commits on May 31, 2019

• Multi tensor lamb optimizer (NVIDIA#334)

  Show description for 8be5b6b
  

  thorjohnsen

  authored and

  mcarilli

  committed
  8be5b6b

  Copy full SHA for 8be5b6b

• Give multi-tensor L2 norm the ability to compute norms per-tensor as well as globally (NVIDIA#333)

  Show description for 93338e6
  mcarilli

  authored
  93338e6

  Copy full SHA for 93338e6

Commits on Apr 26, 2019

• Replace type().ScalarType() with scalar_type() (NVIDIA#272)

  Show description for 855808f
  

  ptrblck

  authored and

  mcarilli

  committed
  855808f

  Copy full SHA for 855808f

Commits on Apr 10, 2019

• Quick kernel to clean up l2norm

  definitelynotmcarilli

  committed
  683b6e0

  Copy full SHA for 683b6e0

• Kernel + sizes stress test

  definitelynotmcarilli

  committed
  1a48b26

  Copy full SHA for 1a48b26

Commits on Apr 9, 2019

• Simple cut of the kernel in place

  definitelynotmcarilli

  committed
  e57f5d0

  Copy full SHA for e57f5d0

Commits on Apr 8, 2019

• Fix for NVIDIA#246

  definitelynotmcarilli

  committed
  03100f4

  Copy full SHA for 03100f4

Commits on Apr 4, 2019

• WIP: Handle arbitrary combinations of optimizers/models/losses (NVIDIA#232)

  Show description for 3f87614
  mcarilli

  authored
  3f87614

  Copy full SHA for 3f87614

Commits on Mar 21, 2019

• Use build macro for backward compat

  syed-ahmed

  committed
  0f5e3fe

  Copy full SHA for 0f5e3fe

• Rename IntList to IntArrayRef

  syed-ahmed

  committed
  2a46709

  Copy full SHA for 2a46709

Commits on Mar 19, 2019

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

  definitelynotmcarilli

  committed
  ac7dbf6

  Copy full SHA for ac7dbf6

• Fixing interaction of DDP with dynamic loss scaling

  definitelynotmcarilli

  committed
  8437d29

  Copy full SHA for 8437d29

• Multi-tensor axpby kernel for more flexible unscaling (groundwork for NVIDIA#163 and NVIDIA#179 fix)

  definitelynotmcarilli

  committed
  5e55200

  Copy full SHA for 5e55200

Commits on Mar 15, 2019

• Anticipating upstream #17996

  definitelynotmcarilli

  committed
  2c8e1c8

  Copy full SHA for 2c8e1c8

Commits on Mar 12, 2019

• Forward/backward compatibility around pytorch 3aeb78, to fix NVIDIA#191

  definitelynotmcarilli

  committed
  42180bd

  Copy full SHA for 42180bd

Commits on Mar 10, 2019

• fix includes

  

  Natalia Gimelshein

  committed
  f34686f

  Copy full SHA for f34686f

• Removing deprecated scale_check_overflow kernel

  definitelynotmcarilli

  committed
  8f53411

  Copy full SHA for 8f53411

Commits on Mar 7, 2019

• Merge pull request NVIDIA#173 from NVIDIA/api_refactor

  Show description for 4714497
  mcarilli

  authored
  4714497

  Copy full SHA for 4714497

Commits on Mar 3, 2019

• Bug fix in next power of 2

  mkolod

  committed
  ca6c276

  Copy full SHA for ca6c276

Commits on Feb 28, 2019

• Comprehensive tests for cross product of options

  definitelynotmcarilli

  committed
  d24c25b

  Copy full SHA for d24c25b

Commits on Feb 24, 2019

• Stashing work

  definitelynotmcarilli

  committed
  d137b80

  Copy full SHA for d137b80

Commits on Feb 22, 2019

• Allow multi-tensor unscale to handle FP16 output, so it can also be used for copy-scatter. Rename some options.

  definitelynotmcarilli

  committed
  80a3f3c

  Copy full SHA for 80a3f3c

Commits on Feb 19, 2019

• Reworked multi tensor apply, added tests

  definitelynotmcarilli

  committed
  6763a8b

  Copy full SHA for 6763a8b

Commits on Feb 13, 2019

• New API tentatively works on resnet50, ready for stress testing.

  definitelynotmcarilli

  committed
  889d171

  Copy full SHA for 889d171

Commits on Feb 11, 2019

• Stashing work

  definitelynotmcarilli

  committed
  fad78c1

  Copy full SHA for fad78c1

Commits on Feb 8, 2019

• stashing work

  definitelynotmcarilli

  committed
  1f693b9

  Copy full SHA for 1f693b9

Commits on Feb 6, 2019

• Tests and resnet50 example work

  definitelynotmcarilli

  committed
  a5bc76d

  Copy full SHA for a5bc76d

• ready for testing

  definitelynotmcarilli

  committed
  6e9159d

  Copy full SHA for 6e9159d

Commits on Feb 5, 2019

• New downscale kernel is working but not perf tested

  definitelynotmcarilli

  committed
  337056c

  Copy full SHA for 337056c

• Merge pull request NVIDIA#146 from NVIDIA/restore_fused_kernel

  Show description for 45537d3
  mcarilli

  authored
  45537d3

  Copy full SHA for 45537d3

Commits on Feb 4, 2019

• Restoring fused inf/nan check + downscale kernel

  definitelynotmcarilli

  committed
  fd03f26

  Copy full SHA for fd03f26

Commits on Feb 1, 2019

• allowing syncBN to run with affine = False

  jjsjann123

  committed
  223a47e

  Copy full SHA for 223a47e

Commits on Jan 18, 2019

• patching grid reduction to be volta-safe

  jjsjann123

  committed
  38bada2

  Copy full SHA for 38bada2

Commits on Jan 15, 2019

• [sync BN nhwc]

  Show description for 443fa76
  jjsjann123

  committed
  443fa76

  Copy full SHA for 443fa76

Commits on Nov 6, 2018

• [syncBN]

  Show description for ee67e56
  jjsjann123

  committed
  ee67e56

  Copy full SHA for ee67e56

Pagination

Previous

Next