Released Chainer v6.3.0

3 views
Skip to first unread message

Hiroyuki Yamazaki

unread,
Aug 22, 2019, 4:08:31 AM8/22/19
to Chainer User Group
We have released Chainer v6.3.0. The release note is:

This is the release note of v6.3.0. See here for the complete list of solved issues and merged PRs.


Highlights

  • NumPy 1.17 is now officially supported.

New Features

  • Add automatic management of snapshots (deletion and load) (#7862)

Enhancements

  • Fix Adam FP16 overflow on gpu kernels (#7780)
  • Make F.average as accurate as backend (#7782)
  • Fix type_check error message on evaluating bool expression (#7801)
  • Fix module in msg of type_check (#7810)
  • Fix F.clip for NumPy 1.17 (#7855)

Bug Fixes

  • Fix Parameter.dtype for uninitialized parameter (#7749)
  • Fix UpdateRule.use_fp32_update for uninitialized parameter (#7751)
  • Avoid unload module call in PureNcclCommunicator (#7787)
  • Fix TypeError in max_pooling_2d (#7789, thanks @ishanrai05!)
  • Fix create_mnbn_model() bug (#7846)
  • Fix backward of split_axis for intel64 when grad_ouputs contains None (#7931)
  • Fix precision in F.max_pooling_2d (#7933)
  • Fix backends.copyto from/to chainerx (#7934)
  • Fix in-place update of arrays in Link.serialize and optimizers.Adam (#7941)
  • Fix ChainerX non native deserialization (#7954)
  • Fix multi-device loss scaling (#7968)

Documentation

  • Fix static_graph docs code examples (#7884)
  • Add 1.17 to supported NumPy versions (#7961)

Tests

  • Fix test of chx.reshape (#7792)
  • Revert #6754 (Fix Travis with macOS) (#7800)
  • Fix a typo in test_communicator (#7822)
  • Fix F.clipped_relu test for NumPy 1.17 (#7854)
  • Switch current directory in Jenkins tests (#7856)
  • Fix flaky TestHuberLoss (#7869)
  • Configure tolerances of F.average_pooling_2d test (#7870)
  • Refactor convolution functions tests (#7873)
  • Relax tolerances in convolution link tests when using old cuDNN (#7878)
  • Fix chainerx.logsumexp test tolerance (#7889)
  • Relax tolerances in convolution function tests when using old cuDNN (#7895)
  • Sample stable inputs in tests of group normalization (#7899)
  • Relax float16 tolerances in ChainerX binary math tests (#7908)
  • Avoid ndarray.data access and fix wrong test (#7913)
  • Avoid unstable inputs in tests of decorrelated batch normalization (#7915)
  • Avoid testing F.cast from negative floating-point to unsigned (#7944)
  • Relax fp16 tolerances in TestContrastive (#7959)
  • Relax fp16 tolerance in TrueDiv test (#7972)
  • Fix tolerance in L.CRF1d test (#7977)
Reply all
Reply to author
Forward
0 new messages