Switch to release channel for PyTorch + CUDA 13

[alexbarghi-nv]

Switches from the nightly channel to the release channel for PyTorch with CUDA 13.

[greptile-apps]

Greptile Overview

Greptile Summary

Switches PyTorch dependency for CUDA 13 from nightly channel to stable release channel now that PyTorch 2.9.0 with CUDA 13 support has been released.

• Removed nightly channel URL (/whl/nightly/cu130) in favor of stable release URL (/whl/cu130) for CUDA 13 in dependencies.yaml:380 and ci/test_wheel_cugraph-pyg.sh:23
• Updated PyTorch version constraint from >=2.9.0.dev0 to >=2.9.0 in dependencies.yaml:392 and python/cugraph-pyg/pyproject.toml:58
• Removed TODO comments referencing the PyTorch CUDA 13 tracking issue
• CUDA 12 continues to use cu126 stable channel (confirmed by commit 1d1238e)

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The changes are straightforward and well-contained: switching from nightly to stable PyTorch releases for CUDA 13. All changes are consistent across files (script, dependencies, and generated pyproject.toml). The version constraint change from dev0 to stable is appropriate. CUDA 12 path remains unchanged at cu126.
• No files require special attention

Important Files Changed

File Analysis

Filename	Score	Overview
ci/test_wheel_cugraph-pyg.sh	5/5	Updated PyTorch index URL for CUDA 13 from nightly to stable release channel
dependencies.yaml	5/5	Removed nightly channel references and updated PyTorch version constraint from >=2.9.0.dev0 to >=2.9.0 for CUDA 13
python/cugraph-pyg/pyproject.toml	5/5	Updated copyright year and PyTorch version constraint from >=2.9.0.dev0 to >=2.9.0 (auto-generated from dependencies.yaml)

Sequence Diagram

sequenceDiagram
    participant CI as CI Workflow
    participant Script as test_wheel_cugraph-pyg.sh
    participant Deps as dependencies.yaml
    participant PyPI as PyTorch Index
    participant Tests as Test Suite
    
    CI->>Script: Execute with RAPIDS_CUDA_VERSION
    Script->>Script: Extract CUDA_MAJOR from version
    
    alt CUDA_MAJOR == "12"
        Script->>PyPI: Request PyTorch from cu126 release channel
    else CUDA_MAJOR == "13"
        Script->>PyPI: Request PyTorch from cu130 release channel (was nightly)
    end
    
    PyPI->>Script: Return PyTorch >=2.9.0 (was >=2.9.0.dev0)
    Script->>Tests: Install wheels with PyTorch
    Tests->>CI: Execute pytest suite

Loading

[greptile-apps]

_{2 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[gforsyth]

Looks like at least one test job has the pytorch index set to nightlies:

if [[ "${CUDA_MAJOR}" == "12" ]]; then
  PYTORCH_INDEX="https://download.pytorch.org/whl/cu126"
else
  PYTORCH_INDEX="https://download.pytorch.org/whl/nightly/cu130"
fi

that's in test_wheel_cugraph-pyg.sh

[greptile-apps]

_{3 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{3 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[alexbarghi-nv]

Looks like at least one test job has the pytorch index set to nightlies:

if [[ "${CUDA_MAJOR}" == "12" ]]; then
  PYTORCH_INDEX="https://download.pytorch.org/whl/cu126"
else
  PYTORCH_INDEX="https://download.pytorch.org/whl/nightly/cu130"
fi

that's in test_wheel_cugraph-pyg.sh

Thanks for the fix. Is there any way we can remove that from the wheel scripts? Would be good to reduce technical debt here.

[gforsyth]

Thanks for the fix. Is there any way we can remove that from the wheel scripts? Would be good to reduce technical debt here.

I'd like to get this unblocked for the release, but I've made a follow-up issue to remove the branching from the test scripts for afterwards.

[gforsyth]

Ok, we're green. Admin merging this to get around the check-nightly-ci failure

[jameslamb]

Exciting, thanks @alexbarghi-nv !!!

And we should hopefully have CUDA 13 conda packages soon too 😁

conda-forge/pytorch-cpu-feedstock#457