Re: Issue with releases / feedback from ASF board
Thanks for the reply. Besides the clarification below, we will track and address here [1] the rest of the comments in this thread. > > - We will introduce source releases on PyPI and Maven using apache-mxnet > > name to provide users with the option to have Apache distribution and allow > > users to compile CUDA/cuDNN code as an option. > > I assume that mean you will two distribution on PyPI? Or is the intent to remove the other one? - We PPMC will create apache-mxnet package with Apache releases in the form of source distributions. This is intended as first-party release to comply with the draft release policy. - As individual, and with appropriate names and description as determined by trademark review, I intend to continue to provide binary releases on PyPI for the convenience of the community. This is out of necessity as compiling all the CUDA code in MXNet can take hours even on powerful CPUs. -sz [1] https://github.com/apache/incubator-mxnet/issues/18397 On Sat, May 23, 2020 at 5:15 PM Justin Mclean wrote: > Hi, > > > - MXNet publishes only nightly pre-release builds to dist.mxnet.io for > the > > verification purpose for projects in the ecosystem and for developers who > > work on the bleeding edge. > > IMO Something more needs to be done here, at the very least I would expect > to se a very large disclaimer that these are not releases for use of the > general public and for internal development use only and are not Apache > Releases. A google search brings up that page and it is incorrectly titled > "Apache MXNet Python Binary Distribution”. I see other download pages that > just contain links and no information to what the files are. > > It also needs to be clear what a user is installed from this install page > [1] which I assume is using the above? The GitHub download page [2] is also > confusing as it contains a mix of Apache and non-Apache releases. > > > - PyPI releases are not the act of this PPMC, but are from individuals > > acting as third-party for the convenience of the community. > > Currently this page isn’t in line with Apache trademark policy. > > > With my PPMC hat on, we will take the following actions: > > > > - We will try to take down problematic Maven releases immediately and > > discuss in the community how to proceed with future Maven releases. > > Thanks for that. > > > - We will introduce source releases on PyPI and Maven using apache-mxnet > > name to provide users with the option to have Apache distribution and > allow > > users to compile CUDA/cuDNN code as an option. > > I assume that mean you will two distribution on PyPI? Or is the intent to > remove the other one? > > > Finally, we appreciate any lesson other projects can share on licensing > and > > distribution that involves CUDA code. Thanks. > > I’m not aware of any other project that uses CUDU code. > > Thanks, > Justin > > 1. https://mxnet.apache.org/get_started/? > 2. https://github.com/apache/incubator-mxnet/releases > - > To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org > For additional commands, e-mail: general-h...@incubator.apache.org > >
Re: Issue with releases / feedback from ASF board
Hi, > Here are some examples using search: > - Arrow https://arrow.apache.org/docs/python/cuda.html "This functionality is optional and must have been enabled at build time. “ So you as a user needs to recompile the code, checking their downloads page this also seems to be the case and there’s no distribution of CUDA code. The others links you provided look similar, but I’m not involved in those projects and it would be best if you asked them. If you think they are distributing CUDA code then bring it up with that project, legal or the board. Thanks, Justin - To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org For additional commands, e-mail: general-h...@incubator.apache.org
Re: Issue with releases / feedback from ASF board
> > I’m not aware of any other project that uses CUDU code. > I want to point out that most machine learning and analytics related projects will likely need to involve (and likely distribute) CUDA code, this would include some of the current TLPs. Here are some examples using search: - Arrow https://arrow.apache.org/docs/python/cuda.html - SystemML https://systemml.apache.org/docs/1.2.0/gpu - Singa https://svn.apache.org/repos/infra/websites/production/singa/content/docs/gpu.html TQ > > 1. https://mxnet.apache.org/get_started/? > 2. https://github.com/apache/incubator-mxnet/releases > - > To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org > For additional commands, e-mail: general-h...@incubator.apache.org > >
Re: Issue with releases / feedback from ASF board
Hi, > - MXNet publishes only nightly pre-release builds to dist.mxnet.io for the > verification purpose for projects in the ecosystem and for developers who > work on the bleeding edge. IMO Something more needs to be done here, at the very least I would expect to se a very large disclaimer that these are not releases for use of the general public and for internal development use only and are not Apache Releases. A google search brings up that page and it is incorrectly titled "Apache MXNet Python Binary Distribution”. I see other download pages that just contain links and no information to what the files are. It also needs to be clear what a user is installed from this install page [1] which I assume is using the above? The GitHub download page [2] is also confusing as it contains a mix of Apache and non-Apache releases. > - PyPI releases are not the act of this PPMC, but are from individuals > acting as third-party for the convenience of the community. Currently this page isn’t in line with Apache trademark policy. > With my PPMC hat on, we will take the following actions: > > - We will try to take down problematic Maven releases immediately and > discuss in the community how to proceed with future Maven releases. Thanks for that. > - We will introduce source releases on PyPI and Maven using apache-mxnet > name to provide users with the option to have Apache distribution and allow > users to compile CUDA/cuDNN code as an option. I assume that mean you will two distribution on PyPI? Or is the intent to remove the other one? > Finally, we appreciate any lesson other projects can share on licensing and > distribution that involves CUDA code. Thanks. I’m not aware of any other project that uses CUDU code. Thanks, Justin 1. https://mxnet.apache.org/get_started/? 2. https://github.com/apache/incubator-mxnet/releases - To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org For additional commands, e-mail: general-h...@incubator.apache.org
Re: Issue with releases / feedback from ASF board
Hi Justin and incubator, Thank you for the feedback. 1. We fully intend to address the licensing issues in releases and software distributions as part of our efforts towards graduation. 2. To my knowledge, there was no intention from this PPMC to bypass the software release policy. To clarify on 2: - MXNet publishes only nightly pre-release builds to dist.mxnet.io for the verification purpose for projects in the ecosystem and for developers who work on the bleeding edge. - PyPI releases are not the act of this PPMC, but are from individuals acting as third-party for the convenience of the community. The binary releases there are not from dist.mxnet.io, but are produced from the same or similar build scripts without modification to the source code. - We made a mistake on Maven releases that we took the same solution as PyPI without carefully reviewing that it complies with the license requirement. With my PPMC hat off, as the individual who made releases to PyPI: - I do NOT intend to confuse the users of our PyPI releases to be the act of MXNet PPMC as I didn't use the apache-mxnet as PyPI package name, which is specified by the draft policy on releases [1]. I will initiate review with Apache Trademark to make sure it is clear. - I do NOT intend to confuse the users of our PyPI releases to be under Apache License 2.0 and it's an artifact from using the same build script. I will seek input from Apache Legal on LEGAL-515 to clarify what the proper licensing should be. With my PPMC hat on, we will take the following actions: - We will try to take down problematic Maven releases immediately and discuss in the community how to proceed with future Maven releases. - We will introduce source releases on PyPI and Maven using apache-mxnet name to provide users with the option to have Apache distribution and allow users to compile CUDA/cuDNN code as an option. Finally, we appreciate any lesson other projects can share on licensing and distribution that involves CUDA code. Thanks. Regards, Sheng On behalf of Apache MXNet (Incubating) PPMC On Fri, May 22, 2020 at 8:49 PM Justin Mclean wrote: > Hi, > > The incubator report had the following feedback: > Incubator needs to address the software distribution issues > regarding MXNet (not reporting this month). The PPMC is > effectively bypassing our software release policies by creating > distribution packages that are combined with non-open-source > platform libraries and publishing them on dist.mxnet.io, where > they are picked up and distributed using the PyPi channel to > all Python users. The resulting pages on PyPi mislead users > into thinking they are installing an Apache License 2.0 package > even though it contains software that we are not allowed to > distribute under our license. > https://issues.apache.org/jira/browse/LEGAL-515 > https://dist.mxnet.io/python/cu102 > https://pypi.org/project/mxnet-cu102/ > > I can see you have been discussing this here [1] so some progress has been > made. It would be a good idea to keep the Incubator PMC in the loop on what > actions the PMC is going to take to resolve this. > > Thanks, > Justin > > 1. https://s.apache.org/5102c > >
