Skip to content

Allow CuPy 9 - #205

Merged
ajschmidt8 merged 4 commits into
rapidsai:branch-0.20from
jakirkham:allow_cupy_9
Apr 28, 2021
Merged

ajschmidt8 merged 4 commits into
rapidsai:branch-0.20from
jakirkham:allow_cupy_9

Conversation

@jakirkham

Copy link
Copy Markdown
Member

Relax the CuPy version constraint to allow CuPy 9 to be used as well.

@jakirkham
jakirkham requested a review from a team as a code owner January 15, 2021 01:06

@ajschmidt8 ajschmidt8 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@jakirkham

Copy link
Copy Markdown
Member Author

That makes sense and agree. Let's hold off on this for a bit then.

@ajschmidt8

Copy link
Copy Markdown
Member

That makes sense and agree. Let's hold off on this for a bit then.

@jakirkham, sounds good to me. do you want to close this PR or convert it to a draft in the meantime? This might help avoid some confusion because I see Dante is also changing the spec in NVIDIA/cuml#3378 (which will also need to be tested with all the repos).

@jakirkham
jakirkham marked this pull request as draft January 15, 2021 15:39
@jakirkham

Copy link
Copy Markdown
Member Author

Good point. Marked as draft. We can revisit once we are ready 🙂

@mike-wendt

Copy link
Copy Markdown
Contributor

Should this be merged with #208 or the other way around?

@jakirkham

Copy link
Copy Markdown
Member Author

Would do Dante's PR first. I think we can wait on this one for a bit. Happy to resolve conflicts here after that goes in.

@raydouglass
raydouglass changed the base branch from branch-0.18 to branch-0.19 March 4, 2021 13:56
@raydouglass

Copy link
Copy Markdown
Contributor

@jakirkham What's the status of this?

@jakirkham

Copy link
Copy Markdown
Member Author

CuPy has been releasing some betas (which we have packaged in conda-forge). These have seen some use already. The CuPy 9 RC is due out 25 March. CuPy 9 final is due out 22 April. More details here ( cupy/cupy#3891 )

This is close to RAPIDS 0.19 release schedule though the fact that RAPIDS 0.19 is due out 15 April (a week before CuPy 9) could make things tricky if we want to include CuPy 9 in RAPIDS 0.19.

Here are some options:

  1. We discuss with CuPy devs about bumping up their release to align with ours
  2. We push out the RAPIDS release a week or so to line up with CuPy's
  3. We hold off on this until RAPIDS 0.20

Thoughts? 🙂

@pentschev

Copy link
Copy Markdown
Member

I would like to point out that CuPy is today maintained by only 3-4 people and they're very often working real hard (including weekends) when release dates are close, so if we want to align 0.19 and CuPy 9 I would suggest we try to align RAPIDS release to CuPy's to alleviate pressure on them.

@kkraus14

kkraus14 commented Mar 4, 2021

Copy link
Copy Markdown
Contributor

I would like to point out that CuPy is today maintained by only 3-4 people and they're very often working real hard (including weekends) when release dates are close, so if we want to align 0.19 and CuPy 9 I would suggest we try to align RAPIDS release to CuPy's to alleviate pressure on them.

RAPIDS 0.19 release is aligned to GTC so we cannot push it at this point.

@pentschev

Copy link
Copy Markdown
Member

RAPIDS 0.19 release is aligned to GTC so we cannot push it at this point.

Ok, then our alternatives are:

  • We don't tie RAPIDS 0.19 to CuPy 9;
  • We ask CuPy maintainers to release earlier because of GTC/RAPIDS 0.19 -- this will possibly require us to allocate a couple engineers in the 2-3 weeks before CuPy's release to help ensure everything goes smoothly.

We have a biweekly call with PFN next Tuesday, so it would be good if we have a decision on what to do by Monday early afternoon ET.

@jakirkham

Copy link
Copy Markdown
Member Author

Do we know if PfN is presenting at GTC? Maybe they would be interested in releasing around the same time for the same reason

@pentschev

Copy link
Copy Markdown
Member

Do we know if PfN is presenting at GTC? Maybe they would be interested in releasing around the same time for the same reason

They've been offered to submit proposals, I don't know what happened to those, I'll check.

@pentschev

Copy link
Copy Markdown
Member

I checked and there will only be an Optuna talk from PFN, no CuPy talks at GTC.

@pentschev

Copy link
Copy Markdown
Member

Reminder that we have the CuPy biweekly meeting tomorrow with PFN. If we need to check whether CuPy 9 release can be done earlier to meet our needs, please let me know by 3pm ET today !

@ajschmidt8
ajschmidt8 changed the base branch from branch-0.19 to branch-0.20 April 23, 2021 19:36
@jakirkham
jakirkham marked this pull request as ready for review April 26, 2021 18:58
@ajschmidt8

Copy link
Copy Markdown
Member

The below libraries have cupy pins that are <9. I think we'll want to get them updated before merging this PR so that we don't run into version conflicts in the containers.

cusignal has <9.1 which should be fine, but may not hurt to update to <10.0.0a0 also

@jakirkham

Copy link
Copy Markdown
Member Author

@jakirkham

Copy link
Copy Markdown
Member Author

All of those PRs have passed 😄

Asking relevant teams for reviews

@jakirkham

Copy link
Copy Markdown
Member Author

Looks like all of those PRs are now merged. Please let me know if there's anything else we need to do here 🙂

@ajschmidt8

Copy link
Copy Markdown
Member

Looks like all of those PRs are now merged. Please let me know if there's anything else we need to do here

@jakirkham, just updated the branch here. assuming CI succeeds again, we can go ahead and merge 👍.

@jakirkham

Copy link
Copy Markdown
Member Author

Looks like it passes! 🟢 😄 🎉

@ajschmidt8
ajschmidt8 merged commit b4c9e06 into rapidsai:branch-0.20 Apr 28, 2021
@jakirkham
jakirkham deleted the allow_cupy_9 branch April 28, 2021 22:43
@jakirkham

Copy link
Copy Markdown
Member Author

Thanks AJ! 😄

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants