Update torch and torchvision package versions for CUDA 13.x

[JanuszL]

• torch: 2.11.0+cu130 (CUDA 13.x)
• torchvision: 0.26.0+cu130 (CUDA 13.x)
• Broadened extra_index URL to https://download.pytorch.org/whl/
  to support both CUDA variants
• Updated python_max_ver to 3.14 for CUDA 13.x packages

Category:

Other (Configuration)

Description:

Updates the torch and torchvision package versions used in QA setup to the latest available releases, and broadens the PyTorch extra index URL so it can serve packages for CUDA 13.x builds.

Additional information:

Affected modules and functionalities:

• qa/setup_packages.py: Updated torch and torchvision versions and their extra index URL.
  • torch: 2.11.0+cu130 (CUDA 13.x)
  • torchvision: 0.26.0+cu130 (CUDA 13.x)
  • extra_index: https://download.pytorch.org/whl/cu128/ -> https://download.pytorch.org/whl/ (unified index for all CUDA variants)
  • Updated python_max_ver to 3.14 for CUDA 13.x torch/torchvision packages

Key points relevant for the review:

• The extra index URL was changed from the CUDA-128-specific path to the base PyTorch wheel index, which serves version-suffixed wheels (e.g. +cu128, +cu130) correctly for both CUDA variants.
• python_max_ver for CUDA 13.x packages bumped to 3.14 to align with the new torch 2.10.0 support matrix.

Tests:

• Existing tests apply
• New tests added
• N/A

Checklist

Documentation

• Existing documentation applies
• Documentation updated
• N/A

DALI team only

Requirements

• Implements new requirements
• Affects existing requirements
• N/A

REQ IDs: N/A

JIRA TASK: N/A

[greptile-apps]

Greptile Summary

This PR updates the PyTorch and torchvision versions in the QA package setup for CUDA 13.x builds (torch 2.11.0+cu130, torchvision 0.26.0+cu130), adds explicit +cu128 CUDA suffixes to the CUDA 12.x version strings, and broadens extra_index from the CUDA-128-specific path to the base PyTorch wheel index to serve both variants. The python_min_ver for CUDA 13.x packages is also bumped from 3.9 to 3.10, reflecting the torch 2.11.0 support matrix, and python_max_ver is extended to 3.14.

One minor note: the PR description references "torch 2.10.0" in its justification for the python_max_ver bump, but the actual version installed is 2.11.0+cu130.

Confidence Score: 5/5

Safe to merge — this is a straightforward version-bump and config change with no logic alterations.

All changes are additive configuration updates (version strings, index URL, Python version bounds). The explicit CUDA suffixes in version strings combined with the broader index URL is the correct approach for serving multi-CUDA-variant wheels. The only finding is a minor discrepancy in the PR description text (2.10.0 vs 2.11.0), which does not affect any code path.

No files require special attention.

Important Files Changed

Filename	Overview
qa/setup_packages.py	Updates torch (2.11.0+cu130) and torchvision (0.26.0+cu130) for CUDA 13.x, adds explicit +cu128 suffix to CUDA 12.x versions, broadens extra_index URL, and bumps python_min_ver to 3.10 for CUDA 13.x packages.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[QA Setup: Select torch/torchvision] --> B{CUDA Version?}
    B -- CUDA 12.x --> C["torch==2.7.1+cu128\n(Python 3.9-3.13)"]
    B -- CUDA 13.x --> D["torch==2.11.0+cu130\n(Python 3.10-3.14)"]
    C --> E["torchvision==0.22.1+cu128\n(Python 3.9-3.13)"]
    D --> F["torchvision==0.26.0+cu130\n(Python 3.10-3.14)"]
    E --> G["pip install --extra-index-url\nhttps://download.pytorch.org/whl/"]
    F --> G

Loading

_{Reviews (2): Last reviewed commit: "Fix" | Re-trigger Greptile}

[dali-automaton]

CI MESSAGE: [48509214]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [48509223]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [48664587]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [48664598]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [48664598]: BUILD FAILED

[dali-automaton]

CI MESSAGE: [48664587]: BUILD FAILED

[dali-automaton]

CI MESSAGE: [48664587]: BUILD PASSED

[dali-automaton]

CI MESSAGE: [48664598]: BUILD PASSED