Thanks Dave, Henry, Ming for the discussions so far.
Please allow me to try to dissect and summarize the
discussion points and answers so far:
D0: Release process
- The PPMC has a well document release process[1] and has proj
D1: Use of github and dev@
- We follow the principle “everything
I’m changing my vote to +1 (binding)
Best of luck TVM.
Regards,
Dave
> On Aug 27, 2020, at 10:29 AM, Dave Fisher wrote:
>
> I’m -1 (binding), because this is being rushed. I’ll be willing to change
> after there is actual discussion on general@.
>
>> On Aug 27, 2020, at 10:02 AM, Tianqi
+1 (binding)
On Thu, Aug 27, 2020 at 10:02 AM Tianqi Chen wrote:
> Dear Incubator:
>
>
>
> Following a heated discussion with great support from our mentors,
>
> committers and community members.
>
> The Apache(incubating) TVM community has formally voted for graduation and
>
> bring the case
+1 (binding)
On Fri, Aug 28, 2020 at 10:21 PM Felix Cheung
wrote:
> +1 (binding)
>
>
> On Thu, Aug 27, 2020 at 10:02 AM Tianqi Chen wrote:
>
> > Dear Incubator:
> >
> >
> >
> > Following a heated discussion with great support from our mentors,
> >
> > committers and community members.
> >
> >
Hi,
I’m also concerned that the person who seems responsible for these docker
images seems to be the proposed chair of the project.
Thanks,
Justin
-
To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org
For
Hi,
> Then why do they have 40K+ downloads? It also doesn’t explain the version
> numbers.
Sorry 4K I misread that, but even that seem high for something that is used for
testing.
Thanks,
Justin
-
To unsubscribe, e-mail:
Hi,
> - As you can see that the domain redirects to https://tvm.apache.org/.
> So hopefully there is no confusion here.
In some cases it does not. e.g. discuss.tvm.ai IMO this is a graduation blocker.
> P2: releases
> - The PPMC has a well documented release process[1]
> that anyone can go
Hi,
On that installation pages it does says "We provide docker utility scripts to
help developers to setup development environment. They are also helpful run
through TVM demo and tutorials.” and "You can find some un-official prebuilt
images in https://hub.docker.com/r/tvmai/.” IMO this
Hi,
> I am sure there are many other binary packaging out there that use Apache
> project names, for example this one for Apache Spark:
> https://hub.docker.com/r/bitnami/spark/
What other Apache projects do (Spark in particular for various reasons) are not
always good examples to follow. But
Hi,
The "install from source” page should probably point people to the source
releases rather than the latest code in GitHub.
Thanks,
Justin
-
To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org
For additional
Thanks Justin!
Docker Images and Downloads Stats
- These images are indeed only used for testing purposes.
Note that there are more than 150 PRs every month, results in more than
tens of thousands of tests being triggered.
Every now and then the CI machine need to pull the fresh new image
To follow up the comment on docker image, as I mentioned in my previous
email.
We are changing the use of docker hub name to a new name that does not
contain tvm https://github.com/apache/incubator-tvm/pull/6360
The tvmai docker hub will be removed after the PR is merged
As I stated in my
Thanks TQ, I understand the situation for the current Docker images.
The reason I mentioned about having official Docker images for TVM in the
ASF Docker hub to help differentiate between community/ external one, and
the one officially come from Apache TVM (P)PMC.
- Henry
On Fri, Aug 28, 2020,
Sure, that is easy to fix and good suggestion. But, hope it is not a
blocker issue.
On Fri, Aug 28, 2020, 5:09 PM Justin Mclean
wrote:
> Hi,
>
> The "install from source” page should probably point people to the source
> releases rather than the latest code in GitHub.
>
> Thanks,
> Justin
>
Look like TQ has given you answer and proposed solution about the Docker
images.
Hope that helps.
On Fri, Aug 28, 2020, 5:01 PM Justin Mclean
wrote:
> Hi,
>
> > I am sure there are many other binary packaging out there that use Apache
> > project names, for example this one for Apache Spark:
>
The Apache does have official Docker hub [1].
So, the TVM PPMC could, in the near future, create Docker images for Apache
TVM for official distribution to help people quickly consume and run the
source code releases.
[1] https://hub.docker.com/u/apache
On Fri, Aug 28, 2020, 4:29 PM Tianqi
Thanks Justin for bringing up these points!
P0: tvm.ai domain
- As you can see that the domain redirects to https://tvm.apache.org/.
So hopefully there is no confusion here.
P2: releases
- The PPMC has a well documented release process[1]
that anyone can go and produce a release.
- While we
Hi Justin,
I am sure there are many other binary packaging out there that use Apache
project names, for example this one for Apache Spark:
https://hub.docker.com/r/bitnami/spark/
Are you concern about the name? The Docker image hub clearly stated this is
not official Docker image from Apache TVM
Hi Justin:
Again to further clarify the Docker image.
First of all, the PPMC is 100% with you that it is NOT fine to
release docker images with the Apache brand, or not following the
apache official release tag.
The test cache binary used by TVM CI is a different thing.
As the docker image
HI,
I’m -1 (binding), please see the discussion thread for my concerns.
Thanks,
Justin
-
To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org
For additional commands, e-mail: general-h...@incubator.apache.org
Hi,
I also have several concerns.
1. Use of the tvm.ai domain - has permission been obtained from trademark to
use this domain? Who has control of it?
2. With only 2 releases with 1 release manager I don’t think the project is
mature enough to graduate,
3. Branding/trademarks on these
Thanks Henry.
In this case, we choose the thirdparty for binary testing cache in
particular to simplify the trademark concerns
(since we do not want to vet the license of the test env). Which I believe
is allowed.
In terms of the docker image source, the community strives to release
source
Thanks Justin and Henry for the discussion thread.
Please allow me to dissect and summarize again some of the discussion
points:)
C0: Docker Image
- There was a mis-understanding that the docker image like ci-gpu
contains tvm. They do not, and instead contain an environment to
build and run
Makes sense. Do you think the recent changes to Docker Hub will affect
this? It might be a good idea to move the images into the Apache org, even
if they’re not for end users.
On Fri, Aug 28, 2020 at 20:52 Tianqi Chen wrote:
> Thanks Matt!
>
>
>
> That was indeed the approach we used before.
my -1 non-binding.
see discuss thread about my concerns.
Justin Mclean 于 2020年8月29日周六 上午5:59写道：
> HI,
>
> I’m -1 (binding), please see the discussion thread for my concerns.
>
> Thanks,
> Justin
> -
> To unsubscribe, e-mail:
tvmlang.org was an old url by the project that eventually got deprecated at
some time point, it now redirects to the https://tvm.apache.org/
The PMC makes sure to redirect to the static Apache project when possible
for official project related information.
Due to the mechanism of the discourse
I think only two releases and only one release manager are blocking issues.
Considering that TVM merges 150 PRs a month on average, but only two
releases within a year and a half, that is, users use codes that are much
behind the master, which is not friendly to the community.
Second, there is
Thanks Matt!
The main reason we used thirdparty(as stated above) is that the TVM PPMC
believes it is important to
protect the Apache brand and release policy(artifact being compatible
apache).
Given some of the optional test packages (e.g. pytorch/tensorflow) are not
what we could control,
and it
We have also talked to Apache INFRA about the use of the https://ci.tvm.ai/ due
to the need of special setups in terms of GPU based resources etc.
So far we have received warm help from the INFRA and they are super helpful
in helping setting up CI hooks so that the community can enjoy quick
Thanks Matt!
That was indeed the approach we used before. The main problem of this
approach is
- There could be certain upstream changes (say tensorflow get updated) that
may not retrigger the rebuild
- The CI instance itself can quickly get crowded by the historical cached
images and causes disk
Thanks Ming.
Certainly, different PMC can take a different approach towards release.
We are not suggesting that every PMC should take our approach.
In the particular case of TVM, the community has been working on several
major features including Ansor, uTVM and it does not make sense to cut
the
Thanks Justin. To further address your listed points:
P0: tvm.io
The original project only uses the domain tvm.ai, which is redirected to
the https://tvm.apache.org/.
Other tvm related domain names have nothing to do with ApacheTVM. Note that
because TVM as a three letter abbreviation is quite
I would like to revisit a bit about the release topic to shed light about
the release process.
The TVM PPMC has always been focusing on producing high quality releases.
- The project has produced five major (non-apache) releases prior joining
Apache
- While most incubator projects start with a
Is there even a need to upload the Docker images for CI? Docker recognizes
layers by checksums, so CI build agents will cache the image layers
regardless of whether or not you upload them to Docker Hub. Layers get
rebuilt when the Docker file changes or you force an update.
On Fri, Aug 28, 2020
Hi,
Thanks for the response. I think the docker issue is resolved.
I still have concerns about use of tvm.io. I have discovered several other
subdomains (six in total) that do not redirect to the Apache site, there might
be others. It's still unclear who has control of this domain name. The
Hi,
Sorry my mistake, I meant to say tvm.ai not tvm.io, but as I said subdomains on
tvm.ai are not redirecting. For instance discuss.tvm.ai stays as it is. Other
sub-domains include ci.tvm.ai , tracking.tvm.ai and docs.tvm.ai that I’ve
found. There might be others, of those four, only docs
36 matches
Mail list logo
