Now, the GPU tests CI action automatically appends the result to prevent it from re-running.

[finbarrtimbers]

After a successful run, the action will append "GPU_TESTS=[experiment ID](beaker link)" to the PR. This will prevent it from running again.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[gemini-code-assist]

Summary of Changes

Hello @finbarrtimbers, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces robust verification mechanisms for critical training artifacts, such as model checkpoints and DeepSpeed states, ensuring their validity and completeness. Concurrently, it streamlines the continuous integration workflow for GPU tests by automating the recording of experiment results directly into the pull request, thereby preventing unnecessary re-executions and improving efficiency.

Highlights

• Artifact Verification: Implemented new verification functions (verify_model_checkpoint, verify_deepspeed_state_checkpoint, verify_traces) to ensure the integrity and completeness of saved model checkpoints, DeepSpeed states, and training traces.
• Integrated Verification Checks: Integrated these verification checks into the model saving processes (maybe_save_checkpoint, save_final_model) and DeepSpeed state saving logic to proactively detect corrupted or incomplete artifacts during training.
• CI Automation for GPU Tests: Enhanced the GPU test CI script (run_gpu_tests_on_beaker.sh) to output experiment metadata (ID and URL) to GITHUB_OUTPUT, enabling automated updates to pull request descriptions and preventing redundant test runs.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Ignored Files

• Ignored by pattern: .github/workflows/** (1)
  • .github/workflows/tests.yml

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces verification functions for model checkpoints, DeepSpeed states, and training traces to enhance the robustness of the training pipeline. These checks are integrated into the saving process. Additionally, the GPU test CI script is updated to output Beaker experiment details, which is a useful improvement for CI/CD. The changes are well-structured, but I have a couple of suggestions to improve the conciseness and efficiency of the new verification functions.

I am having trouble creating individual review comments. Click here to see my feedback.

open_instruct/grpo_fast.py (150-164)

The logic for checking for a valid checkpoint can be significantly simplified. The initial checks for marker_path and config_path can be combined, and the if block to set the weights_valid boolean flag can be replaced by directly returning the result of the boolean expression. This will make the code more concise and easier to read.

    if not marker_path.exists() or not config_path.exists():
        return False

    return (safetensors_path.exists() and safetensors_path.stat().st_size > 1024) or (
        pytorch_path.exists() and pytorch_path.stat().st_size > 1024
    )

open_instruct/grpo_fast.py (195-196)

For checking if any rollout files exist, using any() with a generator expression is more efficient than creating a full list with list() and then checking its length. any() will stop iterating as soon as it finds the first matching file, which can be faster if there are many files.

    return any(rollouts_path.glob(f"{run_name}_rollouts_*.jsonl"))