Commits

Breadcrumbs

History for

onmaster

User selector

All users

Datepicker

All time

Commit History

Commits on Mar 23, 2019

• Fix typo in setup.py error message on torch version check (NVIDIA#219)

  

  Cubbee

  authored and

  mcarilli

  committed
  dc55a99

  Copy full SHA for dc55a99

Commits on Mar 22, 2019

• Check cuda version (NVIDIA#216)

  Show description for 5b8faa2
  mcarilli

  authored
  5b8faa2

  Copy full SHA for 5b8faa2

Commits on Mar 19, 2019

• 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 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

• Removing deprecated scale_check_overflow kernel

  definitelynotmcarilli

  committed
  8f53411

  Copy full SHA for 8f53411

Commits on Mar 5, 2019

• Documentation updates

  definitelynotmcarilli

  committed
  7f39db9

  Copy full SHA for 7f39db9

Commits on Mar 4, 2019

• Cleaning up READMEs

  definitelynotmcarilli

  committed
  df83b67

  Copy full SHA for df83b67

Commits on Feb 19, 2019

• Reworked multi tensor apply, added tests

  definitelynotmcarilli

  committed
  6763a8b

  Copy full SHA for 6763a8b

Commits on Feb 11, 2019

• Stashing work

  definitelynotmcarilli

  committed
  fad78c1

  Copy full SHA for fad78c1

Commits on Feb 4, 2019

• Restoring fused inf/nan check + downscale kernel

  definitelynotmcarilli

  committed
  fd03f26

  Copy full SHA for fd03f26

Commits on Dec 12, 2018

• Warning instead of error if nvcc is not found

  definitelynotmcarilli

  committed
  197bcc4

  Copy full SHA for 197bcc4

Commits on Oct 31, 2018

• [WIP] Fused layer norm cuda (NVIDIA#69)

  Show description for 1b9b65c
  

  thorjohnsen

  authored and

  mcarilli

  committed
  1b9b65c

  Copy full SHA for 1b9b65c

Commits on Oct 30, 2018

• Remove arch from adam compile options

  

  ngimel

  authored
  45f030d

  Copy full SHA for 45f030d

Commits on Oct 29, 2018

• Merging in fused adam optimizer, additional DDP features tested in 18.10 (NVIDIA#60)

  Show description for e0bc5d6
  mcarilli

  authored
  e0bc5d6

  Copy full SHA for e0bc5d6

Commits on Oct 23, 2018

• [syncBN] (NVIDIA#48)

  Show description for 81eef1e
  

  jjsjann123

  authored and

  mcarilli

  committed
  81eef1e

  Copy full SHA for 81eef1e

Commits on Jul 23, 2018

• Switch to simple Python-only install, in preparation for upstreaming C++ backend.

  definitelynotmcarilli

  committed
  d695b68

  Copy full SHA for d695b68

Commits on Jul 5, 2018

• Update setup.py for NVIDIA#23

  mcarilli

  authored
  247349f

  Copy full SHA for 247349f

Commits on Jul 4, 2018

• docs: minor spelling tweaks (NVIDIA#25)

  

  brettkoonce

  authored and

  mcarilli

  committed
  d74fda2

  Copy full SHA for d74fda2

Commits on Jun 24, 2018

• Response to issue NVIDIA#16: more comprehensive instructions in examples/docker

  definitelynotmcarilli

  committed
  2e7d799

  Copy full SHA for 2e7d799

Commits on Jun 21, 2018

• import ctypes for line 51 of setup.py

  

  cclauss

  authored
  5f2c649

  Copy full SHA for 5f2c649

Commits on Jun 14, 2018

• Pulling in old logic to manually look for CUDA_HOME in Pytorch <= 0.4 to allow cross-compilation

  definitelynotmcarilli

  committed
  98fa5a3

  Copy full SHA for 98fa5a3

Commits on Jun 7, 2018

• Reverting setup.py to rely on cpp_extension's native CUDA_HOME, because upstream fixed cross-compilation

  definitelynotmcarilli

  committed
  bf85538

  Copy full SHA for bf85538

• Updating setup.py to enable cross-compilation

  definitelynotmcarilli

  committed
  70b3377

  Copy full SHA for 70b3377

Commits on Jun 6, 2018

• Macros based on torch.__version__ to compile with 0.4 and 0.5

  definitelynotmcarilli

  committed
  d506eff

  Copy full SHA for d506eff

Commits on May 26, 2018

• Fleshed out Cuda version checking and compiling for multiple arches

  definitelynotmcarilli

  committed
  fb7d4e1

  Copy full SHA for fb7d4e1

Commits on May 25, 2018

• Transferred backend and build system to use Pytorch C++ extension + ATen dispatch.

  definitelynotmcarilli

  committed
  d17a015

  Copy full SHA for d17a015

Commits on May 18, 2018

• Initial support for automatic mixed precision

  

  carlc-nv

  authored and

  definitelynotmcarilli

  committed
  e733e78

  Copy full SHA for e733e78

Commits on May 15, 2018

• Fix build for pytorch post 0.4

  csarofeen

  committed
  83acda9

  Copy full SHA for 83acda9

Commits on May 4, 2018

• Don't need libcuda for building.

  csarofeen

  committed
  9b76c2a

  Copy full SHA for 9b76c2a

Commits on May 2, 2018

• Editing setup.py to locate nvcc and detect cuda major version more strictly

  definitelynotmcarilli

  committed
  5dfa4c3

  Copy full SHA for 5dfa4c3

Commits on May 1, 2018

• Try to improve robustness of finding cuda in build. Try to support building with CUDA 8.

  csarofeen

  committed
  5709cfb

  Copy full SHA for 5709cfb

• Don't compile cuda code if running clean.

  csarofeen

  committed
  1cea100

  Copy full SHA for 1cea100

Commits on Apr 25, 2018

• Initial release

  csarofeen

  committed
  2fa4dba

  Copy full SHA for 2fa4dba

Loading