don't build images w cuda 130 since we don't have flash attention wheels

[winglian]

Description

Motivation and Context

How has this been tested?

Screenshots (if appropriate)

Types of changes

Social Handles (Optional)

Summary by CodeRabbit

• Chores
  • Removed CUDA 13.0.0 configuration from build workflows.
  • Enhanced Docker build to support all PyTorch 2.9.x versions instead of requiring an exact version match.

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

[coderabbitai]

📝 Walkthrough

Walkthrough

The pull request modifies CI/build configuration and Docker setup. It comments out a CUDA 13.0 matrix entry in GitHub Actions workflows and changes the flash_attn installation condition in the Dockerfile from an exact PyTorch version match to a regex pattern matching any 2.9.x version.

Changes

Cohort / File(s)	Change Summary
CI/Build Configuration .github/workflows/main.yml	Comments out CUDA 13.0 matrix entry (cuda: 130, cuda_version 13.0.0, Python 3.11, PyTorch 2.9.1, axolotl_extras) in GitHub Actions workflows
Docker Configuration docker/Dockerfile-base	Replaces exact PyTorch version match "2.9.1" with regex pattern ^2\.9\.[0-9]+$ for flash_attn installation condition

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• build cuda 13.0.0 base image with 2.9.0 #3229: Directly adds the CUDA 13.0.0 matrix entry that this PR comments out, representing inverse changes to the same build configuration.
• install flash attention in 2.9.0 base images #3224: Modifies the same flash_attn installation conditional in docker/Dockerfile-base for PyTorch 2.9.x variants.
• build docker images for 2.9.x #3273: Updates PyTorch 2.9.x / CUDA 12.8 matrix entries and adjusts Dockerfile versioning logic for overlapping PyTorch versions.

Suggested labels

ready to merge

Suggested reviewers

• SalmanMohammadi
• NanoCode012

Pre-merge checks

✅ 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 directly and clearly summarizes the main change: disabling CUDA 13.0 image builds due to unavailable flash attention wheels.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

---

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

📜 Review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between afe18ac and 412bee3.

📒 Files selected for processing (2)

• .github/workflows/main.yml
• docker/Dockerfile-base

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (14)

• GitHub Check: PyTest from Source Dist (3.11, 2.9.0)
• GitHub Check: PyTest (3.11, 2.8.0)
• GitHub Check: PyTest (3.11, 2.9.0)
• GitHub Check: PyTest (3.11, 2.9.1)
• GitHub Check: PyTest from Source Dist (3.11, 2.8.0)
• GitHub Check: PyTest from Source Dist (3.11, 2.9.1)
• GitHub Check: build-base (130, 13.0.0, 3.11, 2.9.1, 9.0+PTX, Dockerfile-base)
• GitHub Check: build-base (128, 12.8.1, 3.11, 2.8.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
• GitHub Check: build-base (128, 12.8.1, 3.11, 2.9.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
• GitHub Check: build-base (128, 12.8.1, 3.11, 2.9.1, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
• GitHub Check: build-base-uv (128, 12.8.1, 3.11, 2.8.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-uv-base)
• GitHub Check: build-base-uv (128, 12.8.1, 3.11, 2.9.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-uv-base)
• GitHub Check: build-base-uv (128, 12.8.1, 3.11, 2.9.1, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-uv-base)
• GitHub Check: build-base-uv (130, 13.0.0, 3.11, 2.9.1, 9.0+PTX, Dockerfile-uv-base)

🔇 Additional comments (2)

.github/workflows/main.yml (2)

34-38: LGTM!

The CUDA 13.0 matrix entry is correctly disabled by commenting it out, which aligns with the PR objective to stop building images without flash attention wheel support.

---

101-105: LGTM!

Consistent removal of CUDA 13.0 support across all build jobs.

[coderabbitai]

⚠️ Potential issue | 🔴 Critical

Critical syntax error: =~ requires double brackets in Bash.

The regex operator =~ only works with [[ (double brackets) in Bash, not with [ (single bracket/test command). This will cause a build failure or incorrect conditional evaluation.

🔎 Proposed fix

-RUN if [ "$PYTORCH_VERSION" =~ ^2\.9\.[0-9]+$ ] && [ "$CUDA" = "128" ] ; then \
+RUN if [[ "$PYTORCH_VERSION" =~ ^2\.9\.[0-9]+$ ]] && [[ "$CUDA" = "128" ]] ; then \

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	RUN if [ "$PYTORCH_VERSION" =~ ^2\.9\.[0-9]+$ ] && [ "$CUDA" = "128" ] ; then \
	RUN if [[ "$PYTORCH_VERSION" =~ ^2\.9\.[0-9]+$ ]] && [[ "$CUDA" = "128" ]] ; then \

🤖 Prompt for AI Agents

In docker/Dockerfile-base around line 54, the conditional uses the Bash regex
operator with single brackets (`[ "$PYTORCH_VERSION" =~ ... ]`) which is
invalid; replace the single-bracket test with a Bash double-bracket conditional
(`[[ "$PYTORCH_VERSION" =~ ^2\.9\.[0-9]+$ ]] && [ "$CUDA" = "128" ]`) or convert
the entire check to a POSIX-safe alternative (e.g., use grep -E on the variable)
and ensure the Dockerfile uses bash (#!/usr/bin/env bash or `SHELL
["/bin/bash","-c"]`) if relying on `[[ ... =~ ... ]]`.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!