Rebuild for pytorch 2.10 + python 3.14 + CUDA 13.0

[regro-cf-autotick-bot]

This PR has been triggered in an effort to update pytorch210.

Notes and instructions for merging this PR:

1. Please merge the PR only after the tests have passed.
2. Feel free to push to the bot's branch to update this PR if needed.

Please note that if you close this PR we presume that the feedstock has been rebuilt, so if you are going to perform the rebuild yourself don't close this PR until the your rebuild has been merged.

---

If this PR was opened in error or needs to be updated please add the bot-rerun label to this PR. The bot will close this PR and schedule another one. If you do not have permissions to add this label, you can use the phrase @conda-forge-admin, please rerun bot in a PR comment to have the conda-forge-admin add it for you.

_{This PR was created by the regro-cf-autotick-bot. The regro-cf-autotick-bot is a service to automatically track the dependency graph, migrate packages, and propose package version updates for conda-forge. Feel free to drop us a line if there are any issues! This PR was generated by https://github.com/regro/cf-scripts/actions/runs/21440468042 - please use this URL for debugging.}

[conda-forge-admin]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found it was in an excellent condition.

[h-vetinari]

A few jobs did not start; the server is very congested at the moment. Please wait with restarting those until the build out for pytorch is mostly through (this feedstock will likely see new migrators for python 3.14 and CUDA 13.0 anyway due to that merge, one could even decide to wait for those; in any case, let's get the current pytorch builds out the door first)

[weiji14]

Ah sure, just saw that got merged. We can wait until the weekend to rerun with py314.

[weiji14]

@h-vetinari, I'll go with your plan at conda-forge/tensorflow-feedstock#470 (comment) to cherry-pick in #45 and #46 into this PR to conserve CI resources. Assuming that it's ok to push a commit since the tensorflow builds are almost done now.

[h-vetinari]

We have one more big build ongoing for a necessary bugfix to pytorch 2.9. The queue should clear up in ~24h or so. Please wait until then with running CI here.

And yes, combining #45 & #46 with this PR would be great! 🙏

[h-vetinari]

And yes, combining #45 & #46 with this PR would be great!

The way I suggest to do this: cherry-pick all the migrator commits, but leave out the rerender commits from the bot. Then do a single rerender commit for all three migrations (which will conveniently also have your authorship so it can run on cirun)

[weiji14]

Oh nice. I realized after rerendering that there are separate CUDA 12.9 and 13.0 builds now, and across 2 platform (linux-64 & linux-aarch64) multiplied by 5 python versions, that totals up to 20 builds... Thinking if we should drop the CUDA 12.9 builds here, or some of the older python versions?

[h-vetinari]

It's OK... mostly. We don't really have a choice but to support CUDA 12 & 13 for a while, not least since CUDA 13 dropped a bunch of GPU architectures. CUDA 13 is still just rolling out. The answer is mostly: skip stuff (e.g. everything but a single python version) during development/debugging, and then build out the full matrix only once before the merge.

[weiji14]

It's OK... mostly. We don't really have a choice but to support CUDA 12 & 13 for a while, not least since CUDA 13 dropped a bunch of GPU architectures. CUDA 13 is still just rolling out.

Well, flash-attn is for compute capability 8.0+ (Ampere or above), and CUDA 13.0 is compatible to 7.5+ (Turing) if I'm not mistaken. So we could realistically drop CUDA 12.9, assuming people update their drivers. Cutting 10 builds of 8 hours each is a lot of savings.

[h-vetinari]

assuming people update their drivers

well... 😅

If people only updated their OSes, and their drivers, and their libraries, and their python version, life as a distributor would certainly be a heck of a lot easier 😆

In practice, removing support for anything not already dead as a doornail always leads to lots of acrimonious discussion, because no-one likes having something taken away. And in any case, CUDA 12 is faaaaar from that still. For context, we dropped CUDA 11.8 only ~8 months ago.

[h-vetinari]

2/3 GPU jobs for pytorch just finished (and the remaining one should be done soon), so please feel free to go ahead here. There are many other feedstocks in the queue though, so if you happen to not see this (it's the weekend...), I might trigger other builds in the meantime.

[weiji14]

Cool, I'll trigger 4 builds (Python 3.14 only) as a test for now. Make sure that the combination of Pytorch 2.10 on CUDA12.9/13.0 works for both linux-64/linux-aarch64.

[conda-forge-admin]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found it was in an excellent condition.

I do have some suggestions for making it better though...

For recipe/meta.yaml:

• ℹ️ The recipe is not parsable by parser conda-souschef (grayskull). This parser is not currently used by conda-forge, but may be in the future. We are collecting information to see which recipes are compatible with grayskull.
• ℹ️ The recipe is not parsable by parser conda-recipe-manager. The recipe can only be automatically migrated to the new v1 format if it is parseable by conda-recipe-manager.

_{This message was generated by GitHub Actions workflow run https://github.com/conda-forge/conda-forge-webservices/actions/runs/21554512251. Examine the logs at this URL for more detail.}

[h-vetinari]

One of the agents lost its connection. You can either cancel & restart, push another commit, or just live with testing 3 builds (which I think is fine too). If you want to push a change, you could add sm_110 (renamed from sm_101) for CUDA 13, for example

-    - TORCH_CUDA_ARCH_LIST=8.0;8.6;8.9;9.0;10.0;12.0+PTX
+    - TORCH_CUDA_ARCH_LIST=8.0;8.6;8.9;9.0;10.0;12.0+PTX       # [(cuda_compiler_version or "").startswith("12")]
+    - TORCH_CUDA_ARCH_LIST=8.0;8.6;8.9;9.0;10.0;11.0;12.0+PTX  # [(cuda_compiler_version or "").startswith("13")]

[h-vetinari]

or just live with testing 3 builds (which I think is fine too)

I was reminded of issues with GCC 14 + CUDA 13 on aarch for pytorch, so I restarted the failed job, which was just that combination

[h-vetinari]

OK this all looks fine now. :)

[h-vetinari]

Please wait with merging until the linux portion of https://github.com/conda-forge/onnxruntime-feedstock/actions/runs/21577043973 has run through - after that you're good to go!

[weiji14]

Cool, I'll merge this in an hour (without the ci skip) unless there's any last minute concerns. Edit: I'll wait for onnxruntime 🙂

If you want to push a change, you could add sm_110 (renamed from sm_101) for CUDA 13

sm_110 seems to be just for the Jetson Thor chips according to https://developer.nvidia.com/cuda/gpus? Will probably skip that unless someone actually needs it for robotics or something.

[h-vetinari]

Alright, let's get this one built!