Add benchmark

[chenyushuo]

Description

As the title says.

Checklist

Please check the following items before code is ready to be reviewed.

• Code has passed all tests
• Docstrings have been added/updated in Google Style
• Documentation has been updated
• Code is ready for review

[gemini-code-assist]

Summary of Changes

Hello @chenyushuo, 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!

I've introduced new benchmarking capabilities to the project. This pull request primarily adds a dedicated Python script for running benchmarks, along with configuration templates for two specific datasets: gsm8k and countdown. The goal is to provide a structured way to evaluate model performance.

Highlights

• New Benchmarking Script: I've added benchmark/bench.py, a Python script that orchestrates benchmark runs. It dynamically prepares configuration files based on command-line arguments and launches the trinity system.
• Dataset-Specific Benchmark Configurations: I've included two new YAML configuration templates, countdown-template.yaml and gsm8k-template.yaml, which define the specific settings for running benchmarks on these respective datasets.
• Synchronizer Actor Retrieval Improvement: I've made a minor adjustment in trinity/manager/synchronizer.py to optimize the retrieval of the synchronizer actor, ensuring it's returned directly upon creation.

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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

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 issue 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 or fill out our survey to provide feedback.

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 a benchmarking script and configuration templates, which is a valuable addition for performance evaluation. However, I've identified a few areas for improvement. The countdown-template.yaml contains hardcoded, user-specific paths that will prevent the benchmark from running in different environments. I've marked these as critical issues. Additionally, the main benchmark script, benchmark/bench.py, could be improved for better maintainability and robustness. My review includes specific suggestions to address these points.

[chenyushuo]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces a benchmarking script and associated configuration files for gsm8k and countdown datasets. The changes also include a minor style fix in a test file and a logical correction in the Synchronizer actor retrieval.

My review focuses on the new benchmark script (benchmark/bench.py). I've identified a critical race condition in the distributed setup that could lead to errors on worker nodes, and a high-severity security issue related to using an unsafe YAML loader. I've provided detailed suggestions for how to fix these issues. The other changes look good.