[Benchmark] Prefil-only benchmark scripts

[sundar24295s]

Motivation

• This PR extracts and refactors benchmark scripts from Dynamic Batch Tokenizer PR #9382 and PR #9748 to create standalone benchmark scripts with an option to enable garbage collection freeze add this this PR #9241 and profiling.

Modifications

Changes

• bench_score.py - Benchmarks SGLang's /v1/score API performance
• bench_embeddings.py - Benchmarks SGLang's /v1/embeddings API performance
• util.py - Shared utilities for HTTP requests, profiling, and result processing

Key Capabilities:

• Configurable load testing with RPS, duration, and batch size controls
• Distribution patterns - Poisson and constant request distributions
• GC freeze integration - Optional garbage collection freezing for consistent performance
• Profiler support - Built-in torch profiler integration with customizable output directories
• Comprehensive metrics - P50/P90/P99 latencies, success/failure rates, minute-by-minute breakdowns
• Warmup requests - Automatic server warmup before benchmark execution

Benchmarking

• How to run

# Score API benchmarking
python bench_score.py

# Embeddings API benchmarking  
python bench_embeddings.py

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

[gemini-code-assist]

Summary of Changes

Hello @sundar24295s, 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 a dedicated set of benchmark scripts for SGLang's prefill-only operations, specifically targeting the /v1/score and /v1/embeddings APIs. It refactors existing benchmarking logic into a shared utility module and enhances the testing capabilities with configurable load patterns, GC freezing, and profiling support, aiming to provide more robust and detailed performance insights.

Highlights

• Dedicated Prefill-Only Benchmarks: New standalone scripts (bench_score.py, bench_embeddings.py) are introduced under a new benchmark/prefill_only directory for focused performance testing of SGLang's /v1/score and /v1/embeddings APIs.
• Shared Utility Module: Common benchmarking functionalities, including HTTP request handling, profiling setup, and result processing, have been extracted into a new util.py module, promoting code reusability and maintainability.
• Enhanced Load Testing Configuration: The benchmarks now support configurable parameters such as requests per second (RPS), test duration, and batch sizes, allowing for flexible and precise load simulations.
• Request Distribution Patterns: Users can choose between Poisson and constant request distribution patterns to simulate various real-world traffic scenarios.
• Garbage Collection Freeze & Profiling: Optional garbage collection freezing is integrated for consistent performance measurements, alongside built-in torch profiler support with customizable output directories for detailed performance analysis.
• Comprehensive Metrics Reporting: The scripts provide detailed performance metrics including P50/P90/P99 latencies, success/failure rates, and minute-by-minute breakdowns, along with automatic server warmup before execution.

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. 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 refactors the benchmark scripts by extracting common logic into a util.py module, which is a great improvement for code reuse and maintainability. My review focuses on improving the efficiency of the new benchmark scripts. I've identified a couple of places where the tokenizer is loaded repeatedly, and I've provided suggestions to load it only once and pass it as an argument to the relevant functions. This should make the benchmark setup faster and more efficient.