Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Feb 25, 2019

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

  definitelynotmcarilli

  committed
  500939d

  Copy full SHA for 500939d

• Forward+backward compatibility fix around pytorch/pytorch#15744

  definitelynotmcarilli

  committed
  7c36c41

  Copy full SHA for 7c36c41

Commits on Feb 24, 2019

• Merge pull request NVIDIA#167 from NVIDIA/syncbn_api_update

  Show description for 7784ee4
  mcarilli

  authored
  7784ee4

  Copy full SHA for 7784ee4

Commits on Feb 21, 2019

• update SyncBatchNorm API for primitive implementation to support apex.parallel.convert_syncbn_model

  jjsjann123

  committed
  bc76b01

  Copy full SHA for bc76b01

Commits on Feb 20, 2019

• Merge pull request NVIDIA#148 from NVIDIA/ekrimer_add_groupbn_test

  Show description for 8e8dd35
  mcarilli

  authored
  8e8dd35

  Copy full SHA for 8e8dd35

• Update README.md

  mcarilli

  authored
  187ed33

  Copy full SHA for 187ed33

• Update README.md

  mcarilli

  authored
  5c78a50

  Copy full SHA for 5c78a50

• Update README.md

  mcarilli

  authored
  8542db2

  Copy full SHA for 8542db2

• Update README.md

  mcarilli

  authored
  0ff493d

  Copy full SHA for 0ff493d

• Update README.md

  mcarilli

  authored
  6212302

  Copy full SHA for 6212302

Commits on Feb 16, 2019

• moved process group creation into apex so it can be called by users

  ekrimer

  committed
  37cd5df

  Copy full SHA for 37cd5df

• fixing it to work properly in multi-node environment

  ekrimer

  committed
  e49dca6

  Copy full SHA for e49dca6

• adding test_groups.py to unit_test.sh

  ekrimer

  committed
  598fbc8

  Copy full SHA for 598fbc8

Commits on Feb 10, 2019

• Lazily trigger fused_norm for FusedAdam error on Windows

  definitelynotmcarilli

  committed
  69907b8

  Copy full SHA for 69907b8

Commits on Feb 8, 2019

• printout message update

  ekrimer

  committed
  f572555

  Copy full SHA for f572555

• a test and example for sync (group) bn with group_size<world_size

  ekrimer

  committed
  18d412a

  Copy full SHA for 18d412a

Commits on Feb 6, 2019

• Merge pull request NVIDIA#144 from jma127/master

  Show description for 1b90385
  

  ngimel

  authored
  1b90385

  Copy full SHA for 1b90385

• Some documentation cleanup

  definitelynotmcarilli

  committed
  b2f63c4

  Copy full SHA for b2f63c4

• Merge branch 'master' into api_refactor

  definitelynotmcarilli

  committed
  2cbca1a

  Copy full SHA for 2cbca1a

• Tests for the fused downscale kernel

  definitelynotmcarilli

  committed
  340e71a

  Copy full SHA for 340e71a

• Merge branch 'new_downscale_kernel'

  definitelynotmcarilli

  committed
  8818ba9

  Copy full SHA for 8818ba9

• 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

• Better FP16 support in pytorch fp16 utils.

  Show description for 713e0fb
  jma127

  committed
  713e0fb

  Copy full SHA for 713e0fb

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

  definitelynotmcarilli

  committed
  9288ba5

  Copy full SHA for 9288ba5

• Removing patching of loss.backward, which appears to cause memory leaks (reference cycles?) in some models

  definitelynotmcarilli

  committed
  a11c45a

  Copy full SHA for a11c45a

• New downscale kernel is working but not perf tested

  definitelynotmcarilli

  committed
  337056c

  Copy full SHA for 337056c

• Merge pull request NVIDIA#123 from donglixp/patch-1

  Show description for 57ad184
  mcarilli

  authored
  57ad184

  Copy full SHA for 57ad184

• Merge pull request NVIDIA#146 from NVIDIA/restore_fused_kernel

  Show description for 45537d3
  mcarilli

  authored
  45537d3

  Copy full SHA for 45537d3

• Only warn once in LossScaler constructor

  definitelynotmcarilli

  committed
  03b0eeb

  Copy full SHA for 03b0eeb

• FP16 grad downscale (which shouldn't happen in user code) fallback + warning

  definitelynotmcarilli

  committed
  a153c41

  Copy full SHA for a153c41

Commits on Feb 4, 2019

• Merge pull request NVIDIA#143 from NVIDIA/sbn_no_affine

  Show description for d81ed26
  mcarilli

  authored
  d81ed26

  Copy full SHA for d81ed26

• Restoring fused inf/nan check + downscale kernel

  definitelynotmcarilli

  committed
  fd03f26

  Copy full SHA for fd03f26

Commits on Feb 3, 2019

• Lazy imports to reduce error spam

  definitelynotmcarilli

  committed
  48299b0

  Copy full SHA for 48299b0

Commits on Feb 1, 2019

• async->non_blocking, module-specific logging

  definitelynotmcarilli

  committed
  cc85a2e

  Copy full SHA for cc85a2e

Pagination

Previous

Next