You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mxnet.apache.org by Pedro Larroy <pe...@gmail.com> on 2018/05/02 10:56:27 UTC
Re: [VOTE] Release Apache MXNet(incubating) version 1.2.0.RC1
For me the situation with DMLC is problematic.
I often find myself having to fix things in the DMLC subrepos.
* These changes are impossible to test with the MXNet CI system without
doing shenanigans like changing the submodules to my own forks.
* Slows down development as fixes need to be merged on the DMLC subrepo
first, then MXNet updated with the new subrepo commit, effectively doubling
the turnaround time for any fix / commit.
* The DMLC subrepos are in a different organization and subject to
different set of committers and rules.
I think we should have a plan forward to have MXNet be more self contained,
so it's faster and easier to contribute.
Pedro.
On Tue, May 1, 2018 at 1:25 AM, Hen <ba...@apache.org> wrote:
> Naive question - How often does this happen? It should be rare that a
> project needs to send a PR to a dependency, and much rarer that it blocks a
> release.
>
> It sounds like it’s too coupled a dependency. I suspect the DMLC code
> situation is going to be a major blocker to any chance of graduating.
>
> Hen
>
> On Mon, Apr 30, 2018 at 1:23 PM Gautam <ga...@gmail.com> wrote:
>
> > +1 on merging this "https://github.com/dmlc/mshadow/pull/319 "
> >
> > I am curious how are we tracking the sub-modules's PRs which are really
> > important for MXNet?
> > This PR has been waiting to merge for almost 4 months.
> >
> >
> >
> > On Mon, Apr 30, 2018 at 1:51 AM, Pedro Larroy <
> > pedro.larroy.lists@gmail.com>
> > wrote:
> >
> > > -1
> > >
> > > We should merge this and update mshadow before the next release:
> > > https://github.com/dmlc/mshadow/pull/319 so we compile cuda for
> Volta.
> > >
> > > On Sat, Apr 28, 2018 at 12:53 AM, Steffen Rochel <
> > steffenrochel@gmail.com>
> > > wrote:
> > >
> > > > Hi Chris - acknowledge that building the docs is not as good as it
> > should
> > > > be and needs to be improved. Is it worse compared to 1.1.0 release to
> > > > consider as release blocker?
> > > >
> > > >
> > > > On Fri, Apr 27, 2018 at 9:53 AM Chris Olivier <cjolivier01@gmail.com
> >
> > > > wrote:
> > > >
> > > > > -1
> > > > >
> > > > > Building the docs locally is an absolute nightmare. I haven;t been
> > > able
> > > > to
> > > > > get it to work yet.
> > > > >
> > > > > On Thu, Apr 26, 2018 at 3:36 PM, Marco de Abreu <
> > > > > marco.g.abreu@googlemail.com> wrote:
> > > > >
> > > > > > Hello,
> > > > > >
> > > > > > I'd like to request to pause this vote since I have identified an
> > > issue
> > > > > > with our CMakeLists, breaking all UNIX builds with the latest
> > version
> > > > > > (3.11) of cmake. This issue is tracked at [1]. The PR to fix this
> > is
> > > > > > available at [2].
> > > > > >
> > > > > > Best regards,
> > > > > > Marco
> > > > > >
> > > > > >
> > > > > > [1]: https://github.com/apache/incubator-mxnet/issues/10708
> > > > > > [2]: https://github.com/apache/incubator-mxnet/pull/10712
> > > > > >
> > > > > >
> > > > > > On Thu, Apr 26, 2018 at 3:05 PM, Sheng Zha <sz...@gmail.com>
> > > wrote:
> > > > > >
> > > > > > > -1 for the following reasons:
> > > > > > >
> > > > > > > 1. due to addition of support for fp16, the build breaks for
> > > windows
> > > > > and
> > > > > > > older version of osx (clang 8 for example). fix is in
> > > > > > > https://github.com/dmlc/mshadow/pull/333
> > > > > > >
> > > > > > > 2. due to addition of quantized fully connected op, cuda 7.5
> > build
> > > is
> > > > > > > broken. Jun Wu is tracking the issue.
> > > > > > >
> > > > > > > -sz
> > > > > > >
> > > > > > > On Thu, Apr 26, 2018 at 3:01 PM, Anirudh <
> anirudh2290@gmail.com>
> > > > > wrote:
> > > > > > >
> > > > > > > > Hi all,
> > > > > > > >
> > > > > > > > As part of RC1 release, We have addressed the issue with
> > respect
> > > to
> > > > > > > > asymmetric padding in ONNX Import module (
> > > > > > > > https://github.com/apache/incubator-mxnet/pull/10676).
> > > > > > > > We have also added existing silent failures for MXNet Conv
> and
> > > the
> > > > > > > > incompatibility in behavior for certain use cases between
> MXNet
> > > > > without
> > > > > > > > MKLDNN and MXNet with MKLDNN support to the release notes. We
> > > have
> > > > > > marked
> > > > > > > > both ONNX and MKLDNN support as "Experimental" in the release
> > > > notes.
> > > > > > The
> > > > > > > > tutorial
> > > > > > > > <https://github.com/apache/incubator-mxnet/blob/master/
> > > > > > > > docs/tutorials/onnx/fine_tuning_gluon.md>
> > > > > > > > which was called out still seems to be failing with cpu
> context
> > > and
> > > > > we
> > > > > > > have
> > > > > > > > mentioned this as a known issue in the release notes. Since,
> > both
> > > > > > MKLDNN
> > > > > > > > support and ONNX import module have been marked experimental
> > this
> > > > > > should
> > > > > > > > not be a blocking issue.
> > > > > > > >
> > > > > > > > I would like to propose a vote to release Apache MXNet
> > > (incubating)
> > > > > > > version
> > > > > > > > 1.2.0.RC1. Voting will start now (Thursday, April 26th) and
> end
> > > at
> > > > > 3:00
> > > > > > > PM
> > > > > > > > PDT, Monday, April 30th.
> > > > > > > >
> > > > > > > > Link to release notes:
> > > > > > > > https://cwiki.apache.org/confluence/display/MXNET/
> > > > > > > > Apache+MXNet+%28incubating%29+1.2.0+Release+Notes
> > > > > > > >
> > > > > > > > Link to release candidate 1.2.0.rc1:
> > > > > > > >
> > https://github.com/apache/incubator-mxnet/releases/tag/1.2.0.rc1
> > > > > > > >
> > > > > > > > Voting results for 1.2.0.rc0:
> > > > > > > > https://lists.apache.org/thread.html/
> > > > c79327b822c47ec4f97dad463fc6cd
> > > > > > > > fe2aa517772d22eca24c5c300f@%3Cdev.mxnet.apache.org%3E
> > > > > > > >
> > > > > > > > View this page, click on “Build from Source”, and use the
> > source
> > > > code
> > > > > > > > obtained from the 1.2.0.rc1 tag:
> > > > > > > > https://mxnet.incubator.apache.org/install/index.html
> > > > > > > >
> > > > > > > > (Note: The README.md points to the 1.2.0 tag and does not
> work
> > at
> > > > the
> > > > > > > > moment.)
> > > > > > > >
> > > > > > > > Please see a preview of the 1.2.0 version of the website at:
> > > > > > > > http://mxnet.incubator.apache.org/versions/1.2.0/index.html
> > > > > > > >
> > > > > > > >
> > > > > > > > Anirudh
> > > > > > > >
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
> >
> >
> > --
> > Best Regards,
> > Gautam Kumar
> >
>
Re: [VOTE] Release Apache MXNet(incubating) version 1.2.0.RC1
Posted by "Zheng, Da" <dz...@amazon.com>.
I have to agree that the DMLC subrepos do make the development much more difficult sometimes.
On 5/2/18, 3:57 AM, "Pedro Larroy" <pe...@gmail.com> wrote:
For me the situation with DMLC is problematic.
I often find myself having to fix things in the DMLC subrepos.
* These changes are impossible to test with the MXNet CI system without
doing shenanigans like changing the submodules to my own forks.
* Slows down development as fixes need to be merged on the DMLC subrepo
first, then MXNet updated with the new subrepo commit, effectively doubling
the turnaround time for any fix / commit.
* The DMLC subrepos are in a different organization and subject to
different set of committers and rules.
I think we should have a plan forward to have MXNet be more self contained,
so it's faster and easier to contribute.
Pedro.
On Tue, May 1, 2018 at 1:25 AM, Hen <ba...@apache.org> wrote:
> Naive question - How often does this happen? It should be rare that a
> project needs to send a PR to a dependency, and much rarer that it blocks a
> release.
>
> It sounds like it’s too coupled a dependency. I suspect the DMLC code
> situation is going to be a major blocker to any chance of graduating.
>
> Hen
>
> On Mon, Apr 30, 2018 at 1:23 PM Gautam <ga...@gmail.com> wrote:
>
> > +1 on merging this "https://github.com/dmlc/mshadow/pull/319 "
> >
> > I am curious how are we tracking the sub-modules's PRs which are really
> > important for MXNet?
> > This PR has been waiting to merge for almost 4 months.
> >
> >
> >
> > On Mon, Apr 30, 2018 at 1:51 AM, Pedro Larroy <
> > pedro.larroy.lists@gmail.com>
> > wrote:
> >
> > > -1
> > >
> > > We should merge this and update mshadow before the next release:
> > > https://github.com/dmlc/mshadow/pull/319 so we compile cuda for
> Volta.
> > >
> > > On Sat, Apr 28, 2018 at 12:53 AM, Steffen Rochel <
> > steffenrochel@gmail.com>
> > > wrote:
> > >
> > > > Hi Chris - acknowledge that building the docs is not as good as it
> > should
> > > > be and needs to be improved. Is it worse compared to 1.1.0 release to
> > > > consider as release blocker?
> > > >
> > > >
> > > > On Fri, Apr 27, 2018 at 9:53 AM Chris Olivier <cjolivier01@gmail.com
> >
> > > > wrote:
> > > >
> > > > > -1
> > > > >
> > > > > Building the docs locally is an absolute nightmare. I haven;t been
> > > able
> > > > to
> > > > > get it to work yet.
> > > > >
> > > > > On Thu, Apr 26, 2018 at 3:36 PM, Marco de Abreu <
> > > > > marco.g.abreu@googlemail.com> wrote:
> > > > >
> > > > > > Hello,
> > > > > >
> > > > > > I'd like to request to pause this vote since I have identified an
> > > issue
> > > > > > with our CMakeLists, breaking all UNIX builds with the latest
> > version
> > > > > > (3.11) of cmake. This issue is tracked at [1]. The PR to fix this
> > is
> > > > > > available at [2].
> > > > > >
> > > > > > Best regards,
> > > > > > Marco
> > > > > >
> > > > > >
> > > > > > [1]: https://github.com/apache/incubator-mxnet/issues/10708
> > > > > > [2]: https://github.com/apache/incubator-mxnet/pull/10712
> > > > > >
> > > > > >
> > > > > > On Thu, Apr 26, 2018 at 3:05 PM, Sheng Zha <sz...@gmail.com>
> > > wrote:
> > > > > >
> > > > > > > -1 for the following reasons:
> > > > > > >
> > > > > > > 1. due to addition of support for fp16, the build breaks for
> > > windows
> > > > > and
> > > > > > > older version of osx (clang 8 for example). fix is in
> > > > > > > https://github.com/dmlc/mshadow/pull/333
> > > > > > >
> > > > > > > 2. due to addition of quantized fully connected op, cuda 7.5
> > build
> > > is
> > > > > > > broken. Jun Wu is tracking the issue.
> > > > > > >
> > > > > > > -sz
> > > > > > >
> > > > > > > On Thu, Apr 26, 2018 at 3:01 PM, Anirudh <
> anirudh2290@gmail.com>
> > > > > wrote:
> > > > > > >
> > > > > > > > Hi all,
> > > > > > > >
> > > > > > > > As part of RC1 release, We have addressed the issue with
> > respect
> > > to
> > > > > > > > asymmetric padding in ONNX Import module (
> > > > > > > > https://github.com/apache/incubator-mxnet/pull/10676).
> > > > > > > > We have also added existing silent failures for MXNet Conv
> and
> > > the
> > > > > > > > incompatibility in behavior for certain use cases between
> MXNet
> > > > > without
> > > > > > > > MKLDNN and MXNet with MKLDNN support to the release notes. We
> > > have
> > > > > > marked
> > > > > > > > both ONNX and MKLDNN support as "Experimental" in the release
> > > > notes.
> > > > > > The
> > > > > > > > tutorial
> > > > > > > > <https://github.com/apache/incubator-mxnet/blob/master/
> > > > > > > > docs/tutorials/onnx/fine_tuning_gluon.md>
> > > > > > > > which was called out still seems to be failing with cpu
> context
> > > and
> > > > > we
> > > > > > > have
> > > > > > > > mentioned this as a known issue in the release notes. Since,
> > both
> > > > > > MKLDNN
> > > > > > > > support and ONNX import module have been marked experimental
> > this
> > > > > > should
> > > > > > > > not be a blocking issue.
> > > > > > > >
> > > > > > > > I would like to propose a vote to release Apache MXNet
> > > (incubating)
> > > > > > > version
> > > > > > > > 1.2.0.RC1. Voting will start now (Thursday, April 26th) and
> end
> > > at
> > > > > 3:00
> > > > > > > PM
> > > > > > > > PDT, Monday, April 30th.
> > > > > > > >
> > > > > > > > Link to release notes:
> > > > > > > > https://cwiki.apache.org/confluence/display/MXNET/
> > > > > > > > Apache+MXNet+%28incubating%29+1.2.0+Release+Notes
> > > > > > > >
> > > > > > > > Link to release candidate 1.2.0.rc1:
> > > > > > > >
> > https://github.com/apache/incubator-mxnet/releases/tag/1.2.0.rc1
> > > > > > > >
> > > > > > > > Voting results for 1.2.0.rc0:
> > > > > > > > https://lists.apache.org/thread.html/
> > > > c79327b822c47ec4f97dad463fc6cd
> > > > > > > > fe2aa517772d22eca24c5c300f@%3Cdev.mxnet.apache.org%3E
> > > > > > > >
> > > > > > > > View this page, click on “Build from Source”, and use the
> > source
> > > > code
> > > > > > > > obtained from the 1.2.0.rc1 tag:
> > > > > > > > https://mxnet.incubator.apache.org/install/index.html
> > > > > > > >
> > > > > > > > (Note: The README.md points to the 1.2.0 tag and does not
> work
> > at
> > > > the
> > > > > > > > moment.)
> > > > > > > >
> > > > > > > > Please see a preview of the 1.2.0 version of the website at:
> > > > > > > > http://mxnet.incubator.apache.org/versions/1.2.0/index.html
> > > > > > > >
> > > > > > > >
> > > > > > > > Anirudh
> > > > > > > >
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
> >
> >
> > --
> > Best Regards,
> > Gautam Kumar
> >
>
Re: [VOTE] Release Apache MXNet(incubating) version 1.2.0.RC1
Posted by Marco de Abreu <ma...@googlemail.com>.
Totally agree with Henry and Pedro.
On Wed, May 2, 2018 at 12:56 PM, Pedro Larroy <pe...@gmail.com>
wrote:
> For me the situation with DMLC is problematic.
>
> I often find myself having to fix things in the DMLC subrepos.
>
> * These changes are impossible to test with the MXNet CI system without
> doing shenanigans like changing the submodules to my own forks.
> * Slows down development as fixes need to be merged on the DMLC subrepo
> first, then MXNet updated with the new subrepo commit, effectively doubling
> the turnaround time for any fix / commit.
> * The DMLC subrepos are in a different organization and subject to
> different set of committers and rules.
>
> I think we should have a plan forward to have MXNet be more self contained,
> so it's faster and easier to contribute.
>
> Pedro.
>
>
> On Tue, May 1, 2018 at 1:25 AM, Hen <ba...@apache.org> wrote:
>
> > Naive question - How often does this happen? It should be rare that a
> > project needs to send a PR to a dependency, and much rarer that it
> blocks a
> > release.
> >
> > It sounds like it’s too coupled a dependency. I suspect the DMLC code
> > situation is going to be a major blocker to any chance of graduating.
> >
> > Hen
> >
> > On Mon, Apr 30, 2018 at 1:23 PM Gautam <ga...@gmail.com> wrote:
> >
> > > +1 on merging this "https://github.com/dmlc/mshadow/pull/319 "
> > >
> > > I am curious how are we tracking the sub-modules's PRs which are really
> > > important for MXNet?
> > > This PR has been waiting to merge for almost 4 months.
> > >
> > >
> > >
> > > On Mon, Apr 30, 2018 at 1:51 AM, Pedro Larroy <
> > > pedro.larroy.lists@gmail.com>
> > > wrote:
> > >
> > > > -1
> > > >
> > > > We should merge this and update mshadow before the next release:
> > > > https://github.com/dmlc/mshadow/pull/319 so we compile cuda for
> > Volta.
> > > >
> > > > On Sat, Apr 28, 2018 at 12:53 AM, Steffen Rochel <
> > > steffenrochel@gmail.com>
> > > > wrote:
> > > >
> > > > > Hi Chris - acknowledge that building the docs is not as good as it
> > > should
> > > > > be and needs to be improved. Is it worse compared to 1.1.0 release
> to
> > > > > consider as release blocker?
> > > > >
> > > > >
> > > > > On Fri, Apr 27, 2018 at 9:53 AM Chris Olivier <
> cjolivier01@gmail.com
> > >
> > > > > wrote:
> > > > >
> > > > > > -1
> > > > > >
> > > > > > Building the docs locally is an absolute nightmare. I haven;t
> been
> > > > able
> > > > > to
> > > > > > get it to work yet.
> > > > > >
> > > > > > On Thu, Apr 26, 2018 at 3:36 PM, Marco de Abreu <
> > > > > > marco.g.abreu@googlemail.com> wrote:
> > > > > >
> > > > > > > Hello,
> > > > > > >
> > > > > > > I'd like to request to pause this vote since I have identified
> an
> > > > issue
> > > > > > > with our CMakeLists, breaking all UNIX builds with the latest
> > > version
> > > > > > > (3.11) of cmake. This issue is tracked at [1]. The PR to fix
> this
> > > is
> > > > > > > available at [2].
> > > > > > >
> > > > > > > Best regards,
> > > > > > > Marco
> > > > > > >
> > > > > > >
> > > > > > > [1]: https://github.com/apache/incubator-mxnet/issues/10708
> > > > > > > [2]: https://github.com/apache/incubator-mxnet/pull/10712
> > > > > > >
> > > > > > >
> > > > > > > On Thu, Apr 26, 2018 at 3:05 PM, Sheng Zha <szha.pvg@gmail.com
> >
> > > > wrote:
> > > > > > >
> > > > > > > > -1 for the following reasons:
> > > > > > > >
> > > > > > > > 1. due to addition of support for fp16, the build breaks for
> > > > windows
> > > > > > and
> > > > > > > > older version of osx (clang 8 for example). fix is in
> > > > > > > > https://github.com/dmlc/mshadow/pull/333
> > > > > > > >
> > > > > > > > 2. due to addition of quantized fully connected op, cuda 7.5
> > > build
> > > > is
> > > > > > > > broken. Jun Wu is tracking the issue.
> > > > > > > >
> > > > > > > > -sz
> > > > > > > >
> > > > > > > > On Thu, Apr 26, 2018 at 3:01 PM, Anirudh <
> > anirudh2290@gmail.com>
> > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi all,
> > > > > > > > >
> > > > > > > > > As part of RC1 release, We have addressed the issue with
> > > respect
> > > > to
> > > > > > > > > asymmetric padding in ONNX Import module (
> > > > > > > > > https://github.com/apache/incubator-mxnet/pull/10676).
> > > > > > > > > We have also added existing silent failures for MXNet Conv
> > and
> > > > the
> > > > > > > > > incompatibility in behavior for certain use cases between
> > MXNet
> > > > > > without
> > > > > > > > > MKLDNN and MXNet with MKLDNN support to the release notes.
> We
> > > > have
> > > > > > > marked
> > > > > > > > > both ONNX and MKLDNN support as "Experimental" in the
> release
> > > > > notes.
> > > > > > > The
> > > > > > > > > tutorial
> > > > > > > > > <https://github.com/apache/incubator-mxnet/blob/master/
> > > > > > > > > docs/tutorials/onnx/fine_tuning_gluon.md>
> > > > > > > > > which was called out still seems to be failing with cpu
> > context
> > > > and
> > > > > > we
> > > > > > > > have
> > > > > > > > > mentioned this as a known issue in the release notes.
> Since,
> > > both
> > > > > > > MKLDNN
> > > > > > > > > support and ONNX import module have been marked
> experimental
> > > this
> > > > > > > should
> > > > > > > > > not be a blocking issue.
> > > > > > > > >
> > > > > > > > > I would like to propose a vote to release Apache MXNet
> > > > (incubating)
> > > > > > > > version
> > > > > > > > > 1.2.0.RC1. Voting will start now (Thursday, April 26th) and
> > end
> > > > at
> > > > > > 3:00
> > > > > > > > PM
> > > > > > > > > PDT, Monday, April 30th.
> > > > > > > > >
> > > > > > > > > Link to release notes:
> > > > > > > > > https://cwiki.apache.org/confluence/display/MXNET/
> > > > > > > > > Apache+MXNet+%28incubating%29+1.2.0+Release+Notes
> > > > > > > > >
> > > > > > > > > Link to release candidate 1.2.0.rc1:
> > > > > > > > >
> > > https://github.com/apache/incubator-mxnet/releases/tag/1.2.0.rc1
> > > > > > > > >
> > > > > > > > > Voting results for 1.2.0.rc0:
> > > > > > > > > https://lists.apache.org/thread.html/
> > > > > c79327b822c47ec4f97dad463fc6cd
> > > > > > > > > fe2aa517772d22eca24c5c300f@%3Cdev.mxnet.apache.org%3E
> > > > > > > > >
> > > > > > > > > View this page, click on “Build from Source”, and use the
> > > source
> > > > > code
> > > > > > > > > obtained from the 1.2.0.rc1 tag:
> > > > > > > > > https://mxnet.incubator.apache.org/install/index.html
> > > > > > > > >
> > > > > > > > > (Note: The README.md points to the 1.2.0 tag and does not
> > work
> > > at
> > > > > the
> > > > > > > > > moment.)
> > > > > > > > >
> > > > > > > > > Please see a preview of the 1.2.0 version of the website
> at:
> > > > > > > > > http://mxnet.incubator.apache.
> org/versions/1.2.0/index.html
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Anirudh
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> > >
> > >
> > > --
> > > Best Regards,
> > > Gautam Kumar
> > >
> >
>