PrefixRepetitionRandomDataset #20638
Conversation
Signed-off-by: Seiji Eicher <seiji@anyscale.com>
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @eicherseiji, 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 enhances the benchmarking suite by adding a new synthetic dataset type, PrefixRepetitionRandomDataset. This dataset is specifically designed to simulate workloads with shared prefixes, enabling more accurate performance evaluations for systems optimized for such patterns. The changes include the implementation of the new dataset class and its full integration into the existing benchmark_serving script with configurable parameters.
Highlights
- New Benchmark Dataset: Introduced
PrefixRepetitionRandomDatasetinbenchmarks/benchmark_dataset.py. This new dataset class generates synthetic prompts where a fixed number of unique prefixes are randomly created, and then a specified number of random suffixes are appended to each prefix to form individual requests. This is useful for benchmarking scenarios that involve repeated prefixes, such as those leveraging KV cache optimizations. - Integration into Benchmarking Framework: Integrated the new
PrefixRepetitionRandomDatasetintobenchmarks/benchmark_serving.py. Users can now select this dataset using the--dataset-name prefix_repetitioncommand-line argument. - New Command-Line Arguments: Added a new argument group 'repeated prefix dataset options' to
benchmark_serving.py. This group includes arguments like--repeated-prefix-prompts-per-prefix,--repeated-prefix-prefix-len,--repeated-prefix-suffix-len,--repeated-prefix-num-prefixes, and--repeated-prefix-output-len, allowing for fine-grained control over the generation parameters of thePrefixRepetitionRandomDataset.
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 is currently in preview and 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 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
-
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. ↩
There was a problem hiding this comment.
Code Review
This pull request introduces a new PrefixRepetitionRandomDataset for benchmarking scenarios with repeated prefixes. The implementation is mostly correct, but there are a few areas for improvement.
The main issue is that the sample method in the new dataset has a signature that is incompatible with the base class, which I've flagged as a high-severity issue. I've also suggested a refactoring to reduce code duplication within the sample method for better maintainability.
The changes in benchmark_serving.py correctly integrate the new dataset, but the call to the sample method needs to be updated to match the corrected signature.
Overall, the changes are good, and with these fixes, the code will be more robust and maintainable.
Signed-off-by: Seiji Eicher <seiji@anyscale.com>
benchmarks/benchmark_dataset.py
Outdated
|
|
||
| class PrefixRepetitionRandomDataset(BenchmarkDataset): | ||
| # Default values copied from benchmark_serving.py for the repeated prefix dataset. | ||
| DEFAULT_PROMPTS_PER_PREFIX = 200 |
There was a problem hiding this comment.
The default value of 200 per prefix seems high, but yeah :)
| requests.append( | ||
| SampleRequest( | ||
| prompt=prompt, | ||
| prompt_len=prompt_len, |
There was a problem hiding this comment.
if you sum them in the str space it may not preserve the same sum of prefix and decode relation in the token space.
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.Purpose
Test Plan
Test Result
(Optional) Documentation Update