use cuda 12.9.1 and add python 3.12 to base images

[winglian]

Description

Motivation and Context

How has this been tested?

AI Usage Disclaimer

Screenshots (if appropriate)

Types of changes

Social Handles (Optional)

Summary by CodeRabbit

• Chores
  • Added support for CUDA 12.9.1 and CUDA 13.0.0 configurations
  • Extended testing matrix to include Python 3.12
  • Updated build configurations for PyTorch 2.9.1 compatibility

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

📝 Walkthrough

Walkthrough

Expands CI/CD workflow matrices by adding CUDA 12.9.1 and 13.0.0 configurations with Python 3.12 and PyTorch 2.9.1 to build-base and build-base-uv sections in base.yml, and adds Python 3.12 to the test matrix in tests.yml. Configuration-only changes.

Changes

Cohort / File(s)	Summary
Workflow matrix expansions .github/workflows/base.yml	Adds CUDA 12.9.1 (cuda_version: "129") and CUDA 13.0.0 (cuda_version: "130") with Python 3.12 and PyTorch 2.9.1 to both build-base and build-base-uv matrix sections. Includes corresponding cudnn_version, torch_cuda_arch_list, Dockerfile paths, and platform configurations.
Test matrix expansion .github/workflows/tests.yml	Extends python_version matrix from ["3.11"] to ["3.11", "3.12"] for PyTest coverage.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

Possibly related PRs

• build cuda 13.0.0 base image with 2.9.0 #3229: Modifies the same workflow matrices to add CUDA 13.0.0 build configuration for PyTorch 2.9.0, directly overlapping with this PR's CUDA 13.0.0 additions.
• use updated version of prebuilt wheels for flash attention for cu130 #3342: Adds CUDA 13.0.0 (cu130) matrix entries for PyTorch 2.9.1 across workflow configurations with identical version pairings.
• Torch 2.9.1 base images #3268: Updates base Dockerfile images to PyTorch 2.9.1, complementing the PyTorch 2.9.1 matrix additions in this PR.

Suggested labels

ready to merge

Suggested reviewers

• djsaunde
• SalmanMohammadi
• NanoCode012

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately and concisely describes the main changes: adding CUDA 12.9.1 and Python 3.12 support to base images, which aligns with the workflow modifications.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

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.}

[coderabbitai]

Actionable comments posted: 1

🤖 Fix all issues with AI agents

In @.github/workflows/tests.yml:
- Around line 57-58: Update the GitHub Actions workflow to include Python 3.12
in the pytest-sdist job matrix: locate the pytest-sdist job in
.github/workflows/tests.yml and add "3.12" to the python_version matrix (the
matrix key and the job name pytest-sdist are the targets); ensure the matrix
mirrors the pytest job's python_version list so pytest-sdist runs for both 3.11
and 3.12 unless there is an intentional reason to keep it out, in which case add
a comment documenting the limitation.

🧹 Nitpick comments (1)

.github/workflows/base.yml (1)

54-77: Consider adding e2e test coverage for new configurations.

The new CUDA 12.9.1 and Python 3.12 base images are being built, but the e2e tests in tests.yml (docker-e2e-tests) don't yet include corresponding matrix entries. You may want to add e2e coverage for these configurations in a follow-up PR to validate the images work correctly in GPU environments.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

🧩 Analysis chain

🏁 Script executed:

# First, let's examine the .github/workflows/tests.yml file
# Check the pytest job around line 57 and pytest-sdist around line 147
head -170 .github/workflows/tests.yml | tail -120 | cat -n

Repository: axolotl-ai-cloud/axolotl

Length of output: 5081

---

🏁 Script executed:

# Get more context around the specific lines
sed -n '45,75p' .github/workflows/tests.yml | cat -n

Repository: axolotl-ai-cloud/axolotl

Length of output: 1241

---

🏁 Script executed:

# Check the pytest-sdist job around line 147
sed -n '135,160p' .github/workflows/tests.yml | cat -n

Repository: axolotl-ai-cloud/axolotl

Length of output: 1011

---

🌐 Web query:

PyTorch 2.8 2.9 Python 3.12 compatibility support

💡 Result:

Short answer: Yes — PyTorch 2.8 and 2.9 include support for Python 3.12 (prebuilt wheels/binaries are provided for those releases). [1][2][3]

Sources:

• PyTorch/FBGEMM compatibility table showing Python 3.12 for PyTorch 2.8.x and 2.9.x. [1]
• PyTorch 2.8 release announcement. [2]
• PyTorch 2.9 release announcement (notes include support across the 2.9 release family). [3]

If you need links to installation commands (pip/conda) for a specific OS/CUDA combo, tell me your platform and CUDA (or “CPU-only”) and I’ll fetch the exact command.

---

Ensure pytest-sdist job is also updated for Python 3.12 testing.

The pytest job correctly tests both Python 3.11 and 3.12 with PyTorch 2.8.0, 2.9.0, and 2.9.1 (all versions support Python 3.12). However, the pytest-sdist job currently tests only Python 3.11. Update the pytest-sdist job's matrix to include Python 3.12 for consistency and complete coverage, or document if this limitation is intentional.

🤖 Prompt for AI Agents

In @.github/workflows/tests.yml around lines 57 - 58, Update the GitHub Actions
workflow to include Python 3.12 in the pytest-sdist job matrix: locate the
pytest-sdist job in .github/workflows/tests.yml and add "3.12" to the
python_version matrix (the matrix key and the job name pytest-sdist are the
targets); ensure the matrix mirrors the pytest job's python_version list so
pytest-sdist runs for both 3.11 and 3.12 unless there is an intentional reason
to keep it out, in which case add a comment documenting the limitation.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!