Apache MXNet 1.8.0 is a minor release that includes backward compatible changes.
As MXNet follows semver, this release includes backwards-compatible fixes and new features only.
Discussion on the roadmap: https://github.com/apache/incubator-mxnet/issues/18800
Release Managers:
- Sam Skalicky (https://github.com/samskalicky)
- Joseph Evans (https://github.com/josephevans)
- Chaitanya Bapat (https://github.com/ChaiBapchya)
Scope
Reference: MXNet 1.8.0 Roadmap
Issue | PRs | Contributor(s) | Notes |
---|---|---|---|
Upgrade oneDNN version to v1.6 | https://github.com/apache/incubator-mxnet/pull/18867 | @bartekkuncer | Done |
BatchNorm performance improvement for axis!=1 | https://github.com/apache/incubator-mxnet/pull/18676 | @stu1130 | Done |
Users unable to create convolutional layers with large number of filters and large spatial kernel due to upper bound | https://github.com/apache/incubator-mxnet/pull/18910 | @wkcn | Done |
Backward inference support | https://github.com/apache/incubator-mxnet/pull/18895 | @Kh4L | Done |
TensorRT-Gluon Partition API & TRT 7 Suppoer | https://github.com/apache/incubator-mxnet/pull/18916 | @Kh4L | Done |
ONNX v1.7 support | https://github.com/apache/incubator-mxnet/pull/19017 | @josephevans | Done |
Stop packaging GPL libquadmath.so | https://github.com/apache/incubator-mxnet/pull/19055 | Done | |
Change Partition API's options_map to std::unordered_map | https://github.com/apache/incubator-mxnet/pull/18964 | @Kh4L | Done |
Support extra inputs for subgraph ops | https://github.com/apache/incubator-mxnet/pull/18894 | @samskalicky | Done |
Fix race condition in NaiveEngine::PushAsync | https://github.com/apache/incubator-mxnet/pull/19122 | @szha | Done |
Update CUB and include it only for CUDA < 11 | https://github.com/apache/incubator-mxnet/pull/18975 | @Kh4L | |
TensorRT: add INT8 with calibration | https://github.com/apache/incubator-mxnet/pull/19011 | @Kh4L | |
Fix the error of gradient of np.pad | https://github.com/apache/incubator-mxnet/pull/19044 | @CassiniXu | Done |
Intgemm support (8-bit matrix multiply library for x86 CPUs) | https://github.com/apache/incubator-mxnet/pull/19099 | kpuatamazon | |
Update to thrust 1.9.8 on Windows | https://github.com/apache/incubator-mxnet/pull/18977 | @stu1130 | |
Fix for duplicate subgraph inputs/outputs | https://github.com/apache/incubator-mxnet/pull/16131 | @samskalicky | |
Add cmake flag USE_FATBIN_COMPRESSION, ON by default for CUDA >= 11 | https://github.com/apache/incubator-mxnet/pull/19123 | Done | |
Release Timeline
Following timeline is based on everything goes well.(Added some buffer time)
Step | Tasks | Goal | Actual | Comments |
---|---|---|---|---|
Code Freeze and release start | Track ongoing PRs | September 18th | 9/18 | |
Cut the release branch | Check license headers | September 19th | 9/19 | |
Make code changes with necessary version updates on v1.x | September 20th | 9/21 | #19196 | |
Cut the release branch | September 21st | 9/19 | ||
Test the release and tag the release | Nightly test, Jenkins CI | September 25th | 9/24 | |
RAT check | September 25th | 9/24 | ||
Tag RC0 | September 26th | |||
Package artifacts and validate | Create release artifacts | September 26th | ||
Validate release package | September 26th | |||
Test release package | September 26th | |||
Begin apache voting | Start vote on dev@ | September 26th | ||
Start vote on general@ | October 15th | |||
Finalizing and posting the release | Create the final release tag on github | October 30th | ||
Rename, resign and upload the src tar to final dir | October 30th | |||
Update the website using tag | October 31st | |||
Release the official pip package | November 1st | |||
Release the official docker images | November 1st | |||
After 24 hrs, validate the packages are uploaded | November 3rd | |||
Draft the offical announce email and review | November 3rd | |||
Send out the email on announce@ | November 4th | |||
Update the apache blog | November 4th | |||
Update the aws blog | November 4th | |||
Send internal announcement | November 4th |