[Benchmarks] Add memory tracking to serving benchmark

[sfeng33]

Purpose

Partially fix #16353
Add memory usage tracking to benchmark_serving.py via two new metrics: peak_memory_gb and memory_per_request_mb.
These memory metrics will automatically be appended to the JSON output in the included in existing CI (for text model).

Test Plan

Run benchmark_serving.py with command:

  vllm serve microsoft/Phi-3.5-mini-instruct \
    --host 0.0.0.0 \
    --port 8000 \
    --trust-remote-code \
    --max-model-len 4096 \
    --gpu-memory-utilization 0.8

  python benchmarks/benchmark_serving.py \
    --backend vllm \
    --base-url http://localhost:8000 \
    --model microsoft/Phi-3.5-mini-instruct \
    --dataset-name random \
    --num-prompts 10 \
    --save-result \
    --result-filename test_memory_fix.json

Test Output

============ Serving Benchmark Result ============
Successful requests:                     10
Benchmark duration (s):                  1.95
Total input tokens:                      10240
Total generated tokens:                  1280
Request throughput (req/s):              5.12
Output token throughput (tok/s):         655.08
Total Token throughput (tok/s):          5895.72
Peak memory usage (GiB):                 1.59
Memory per request (MiB):                162.35
---------------Time to First Token----------------
Mean TTFT (ms):                          235.98
Median TTFT (ms):                        249.65
P99 TTFT (ms):                           365.82
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          13.34
Median TPOT (ms):                        13.24
P99 TPOT (ms):                           14.67
---------------Inter-token Latency----------------
Mean ITL (ms):                           13.34
Median ITL (ms):                         12.53
P99 ITL (ms):                            52.02
==================================================

Remaining work (Future PRs)

1. Extend to other benchmarks: Add memory tracking to benchmark_throughput.py and benchmark_latency.py
2. Multimodal CI tests: Add actual multimodal models to CI test configs
3. Memory regression detection: Once we obtain a great threshold

[github-actions]

👋 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 fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

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 ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Summary of Changes

Hello @sfeng33, 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 significantly enhances the benchmark_serving.py script by introducing comprehensive memory usage tracking. It adds two new metrics, peak memory usage and memory per request, which are now reported in both the console output and the JSON results. This provides critical insights into the memory footprint of serving models, addressing a partial fix for issue #16353 and laying the groundwork for future memory regression detection.

Highlights

• Memory Tracking Integration: Integrated peak memory usage tracking into benchmark_serving.py to capture peak_memory_gb and memory_per_request_mb.
• New Metrics: Introduced two new metrics, peak_memory_gb and memory_per_request_mb, to the benchmark results, providing deeper insights into memory consumption during serving.
• Reporting Enhancements: Updated the benchmark output (console and JSON) and documentation (README, buildkite descriptions) to include and display these newly tracked memory metrics.
• Cross-Platform Memory Utility: Added a get_memory_usage utility function in benchmark_utils.py that leverages resource.getrusage to accurately measure peak memory across different operating systems (Linux/macOS).

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

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 adds memory tracking to the serving benchmark. However, there's a critical cross-platform compatibility issue in benchmark_utils.py due to the use of the resource module, which will cause the code to fail on Windows. I've provided comments and suggestions to address this, as well as a minor suggestion to improve documentation.