deprecate torch 2.6.0 support

[winglian]

Description

With pytorch currently @ 2.8.0; 2.6.0 is now 2 releases old and we should recommend that users should be on 2.8.0 or 2.7.1 at worst.

Summary by CodeRabbit

• Chores

  • Streamlined CI build matrices by removing older CUDA/PyTorch combos (e.g., PyTorch 2.6.0, CUDA 12.6.x) and standardizing on newer sets (PyTorch 2.7.1/2.8.0, CUDA 12.8.1).
  • Aligned nightly workflows to use updated CUDA/PyTorch versions.

• Tests

  • Reduced test matrix variants to PyTorch 2.7.1 and 2.8.0.
  • Updated nightly and e2e test configurations to CUDA 12.8.1.
  • Removed legacy multi-GPU entry; remaining scenarios unchanged.

• Documentation

  • Updated Quick Start to require PyTorch ≥ 2.7.1.

[coderabbitai]

📝 Walkthrough

Walkthrough

The PR updates GitHub Actions workflow matrices by removing PyTorch 2.6.0/CUDA 12.6.3 entries, elevating supported combinations to PyTorch 2.7.1/2.8.0 and CUDA 12.8.1 in nightly workflows, and aligns tests accordingly. README is updated to reflect the new minimum PyTorch version (≥2.7.1).

Changes

Cohort / File(s)	Summary
CI matrix pruning (remove 2.6.0 / CUDA 12.6.3) .github/workflows/base.yml, .github/workflows/main.yml, .github/workflows/multi-gpu-e2e.yml, .github/workflows/tests.yml	Drops matrix entries using PyTorch 2.6.0 with CUDA 12.6.3 across base images, main builds, multi-GPU e2e, and standard tests; retains newer 2.7.1/2.8.0 combinations.
Nightly matrix bumps (2.7.1/2.8.0, CUDA 12.8.1) .github/workflows/nightlies.yml, .github/workflows/tests-nightly.yml	Updates nightly build/test matrices to PyTorch 2.7.1/2.8.0 and CUDA 12.8.1; adjusts include entries accordingly; no structural workflow changes.
Docs README.md	Updates minimum required PyTorch version from ≥2.6.0 to ≥2.7.1.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• Add 2.8.0 base images and uv images #3034 — Updates base CI matrices to add CUDA 12.8.1 and PyTorch 2.8.0 entries, overlapping with this PR’s matrix adjustments.
• Add support for PyTorch 2.8.0 #3106 — Adjusts workflow matrices to promote PyTorch 2.8.0/CUDA 12.8.1, closely aligned with these version updates.
• run nightly-vs-upstream-main on 2.7.1 and multi-gpu also #2929 — Modifies tests-nightly and docker e2e matrices around PyTorch/CUDA versions, similar files and intent.

Suggested labels

ready to merge

Suggested reviewers

• djsaunde
• SalmanMohammadi

Pre-merge checks and finishing touches

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title Check	✅ Passed	The title “deprecate torch 2.6.0 support” concisely and accurately describes the primary change in the pull request, which is the removal of all PyTorch 2.6.0 entries from the CI matrices and documentation, making it clear to reviewers what the main intent of the PR is.
Docstring Coverage	✅ Passed	No functions found in the changes. Docstring coverage check skipped.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch deprecate-2.6.0

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

📖 Documentation Preview: https://68dfd35eba80467ebb93d768--resonant-treacle-0fd729.netlify.app

Deployed on Netlify from commit 867c99c

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!