add migration for pytorch 2.8

[das-intensity]

Adding migration for pytorch 2.8 as it's been out for almost 3 weeks now.

We still have the migration for 2.7 active https://conda-forge.org/status/migration/?name=pytorch27 so I've pinned this to wait on that one.

However since those package PRs have been WIP for ~3months, I suspect we should just close out that migration.

[das-intensity]

@h-vetinari @jeongseok-meta what do you think? Ok to cancel pytorch27 migration in favor of this? Do I just delete the migration yaml? I've never closed out a migration before.

[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.

[traversaro]

Probably it make sense to first package PyTorch 2.8 before starting a migration for it (see conda-forge/pytorch-cpu-feedstock#409)?

[das-intensity]

@traversaro ahh yes you're right, I saw https://anaconda.org/conda-forge/pytorch show 2.8.0 and figured it'd been migrated, but I see conda-forge/pytorch-cpu-feedstock#410 - will convert to draft until conda-forge/pytorch-cpu-feedstock#409 lands.

[das-intensity]

Re-opened as conda-forge::pytorch has updated to 2.8

[h-vetinari]

I sent #7821 so we can start this on a clean slate. Thanks!

[conda-forge-admin]

Hi! This is the friendly conda-forge automerge bot!

I considered the following status checks when analyzing this PR:

• linter: passed
• github-actions: passed

Thus the PR was passing and merged! Have a great day!

[Luthaf]

I have one question about this migration: it looks like pytorch is properly pinned and pulled at version 2.8, but this is not the case for libtorch:

• in Rebuild for pytorch 2.8 libmetatensor-torch-feedstock#29, none of the file in .ci_support had a libtorch == 2.8 pin added
• The previous build of this feedstock (libmetatensor-torch) already pulled libtorch 2.8, and I had to manually run a migration like this one on python-metatensor-torch, which itself has a dependency on pytorch (python-metatensor-torch v0.8.1 python-metatensor-torch-feedstock#13)

Could it be that libtorch is not properly handled while pytorch is?

[h-vetinari]

That's because the recipe

  host:
    - libmetatensor >=0.1.15,<0.2
    # We build against the CPU version of torch to not have to deal with CUDA
    # compilers (torch's CMake targets tries to find CUDA compilers even when
    # there is no CUDA code to build). This does not impact the `run`
    # dependency, since libtorch `run_exports` are variant agnostic
    - libtorch * cpu*

does not have a pin that the bot recognizes. You have to add a bare - libtorch line for the pinning to take hold.

[Luthaf]

Ok, that makes sense, I'll have a look. I guess I can add both - libtorch and - libtorch * cpu* in there. Thanks!

[h-vetinari]

Ah, I had forgotten that there had been an open issue that I still wanted to fix before the migration: conda-forge/cudnn-feedstock#124

I think we need to lift the cudnn pin for the migration to actually work