Sheng Zha commented on INCUBATOR-253:

> Justin: 3rd party distribution also have different branding and trademark 
> policies they need to comply with. In general they would also need to be 
> based on released code.

> Sheng: I don't think the incubator has the right to require this given that 
> the apache license v2 allows redistribution with or without modification as 
> long as the conditions are met.

> Justin: The incubator has every right to ask podlings to comply with ASF 
> policy. ASF policy is on top of what the Apache License allows.

Agreed on the last comment. This is truism and (I believe) not relevant here, 
given that we were talking about third-party releases. I included the full 
context above for your convenience.

The responsibility of the PPMC in this case would be to make sure users of our 
project are sufficiently notified that these are not first-party releases and 
that they contain additional components that pose further restriction than what 
ALv2 license grants. And we the PPMC fully intend to do so. As mentioned, we 
will propose the option we want to pursue in full observation of the Apache's 
policies and we can continue the discussion on how to do this properly.


> The incubator also has the right to terminate a project if it doesn't comply 
>with policy.

Please realize that if certain constraints affect the viability of our 
ecosystem, a threat to terminate carries little weight to us. Also, one 
difficulty we've been having is caused by individuals imposing their opinions 
as policies on us. Examples include draft release policies that haven't 
sufficiently considered podling's release needs or passed proper Apache 

That said, we are still working hard to find a way to make MXNet viable in 
Apache incubator. In the unfortunate circumstances in which this is no longer 
possible, or that we have policy violations that the incubator can no longer 
tolerate to the extent that termination is warranted, let us know.


At the core, the problem is that GPU acceleration, one of the two pillars of 
modern deep learning frameworks (w/ the other pillar being 
auto-differentiation) only receives the status of category-X optional features 
at Apache. As a result, MXNet has no choice but to seek ways to accommodate 
both the goal of protecting the freedom for the use of our software and the 
goal of having a useful deep learning framework. Thus, we test many unclear 
definitions in the scope of category-X optional features. We also do not expect 
the situation to change as new hardware accelerator vendors are unlikely to 
allow reverse engineering which is incompatible with ALv2. There have been 
several generations of popular deep learning frameworks and we realize now that 
it's no coincidence that none was in Apache before MXNet.

As PPMC, we fully intend to uphold our responsibility of protecting our users 
from legal risks as part of Apache and we are learning our lessons for this. 
Keep in mind that we also have the responsibility to be a useful modern deep 
learning framework. We will not compromise one responsibility for the other. My 
hope is that the incubator is willing to work with us through rational 
discussion, and help guide us to our goal in the welcoming way of Apache.

> Issues with MXNet releases and their distribution
> -------------------------------------------------
>                 Key: INCUBATOR-253
>                 URL: https://issues.apache.org/jira/browse/INCUBATOR-253
>             Project: Incubator
>          Issue Type: Improvement
>            Reporter: Justin Mclean
>            Assignee: Justin Mclean
>            Priority: Major
> The main issues are:
> 1. Source and convenance binary releases containing Category X licensed code.
> 2. Website giving access to downloads of non released/unapproved code.
> 3. Website giving access to releases containing Category X licensed code.
> 4. Web site doesn't given enough warning to users of the issues with non 
> (P)PMC releases or making it clear that these are not ASF releases.
> 5. Maven releases containing Category X licensed code.
> 6. PiPy releases containing Category X licensed code.
> 7. Docker releases containing Category X licensed code.
> 8 Docker releases containing unreleased/unapproved code.
> 9. Trademark and branding issues with PiPy and Docker releases. 
> 10. Trademark and brand issues with naming of releases. 
> 11. Developer releases available to users and public searchable 
> https://repo.mxnet.io / https://dist.mxnet.io
> 12. Releases and other nightly builds on https://repo.mxnet.io / 
> https://dist.mxnet.io containing category X licensed code.
> 13. Lack of clarity on all platforms for what is an ASF release and what is 
> not.
> 14. Branding and release of 3rd parties containing unreleased code. (e.g. 
> https://docs.nvidia.com/deeplearning/frameworks/mxnet-release-notes/rel_20-03.html)
> For PiPy see:
> https://pypi.org/project/mxnet/
> For Docker see:
> https://hub.docker.com/u/mxnet
> For web site pages see:
> https://mxnet.apache.org/get_started?
> https://mxnet.apache.org/get_started/download
> I may of missed something, if so please add it.

This message was sent by Atlassian Jira

To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org
For additional commands, e-mail: general-h...@incubator.apache.org

Reply via email to