[Benchmark][Bugfix] Fix Dataset Length Calculation

[robertgshaw2-redhat]

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

Purpose

• fix dataset

Test Plan

• launch server

vllm serve RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic

• run an example sweep

CONCURRENCIES=(1 25 50 100 150 200)

INPUT_LEN=650
OUTPUT_LEN=150
PREFIX_LEN=500
CC_MULT=30
MODEL=${MODEL:-RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic}
TOKENIZER=${TOKENIZER:-RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic}


for CONCURRENCY in "${CONCURRENCIES[@]}";
do
    NUM_PROMPTS=$(($CC_MULT * $CONCURRENCY))
    
    echo ""
    echo "===== RUNNING $MODEL FOR $NUM_PROMPTS PROMPTS WITH CONCURRENCY $CONCURRENCY ====="
    echo ""

    python3 ../benchmarks/benchmark_serving.py \
        --served-model-name $MODEL \
        --model $TOKENIZER \
        --dataset-name random \
        --random-prefix-len $PREFIX_LEN \
        --random-input-len $INPUT_LEN \
        --random-output-len $OUTPUT_LEN \
        --max-concurrency $CONCURRENCY \
        --num-prompts $NUM_PROMPTS \
        --seed $(date +%s) \
        --percentile-metrics ttft,tpot,itl,e2el \
        --metric-percentiles 90,95,99 \
        --ignore-eos

done

Test Result

I get the proper output length reported.

(Optional) Documentation Update

[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 @robertgshaw2-redhat, 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 addresses a critical bug in the benchmark dataset generation process. The primary goal is to ensure that the total input length for sampled requests is accurately determined based on the actual tokenized sequence, rather than relying on potentially incorrect or approximated values. This correction is vital for maintaining the integrity and reliability of benchmark results, as incorrect length calculations could skew performance metrics.

Highlights

• Bugfix: Corrected the calculation of total_input_len within the sample function in benchmark_dataset.py. Previously, it used prefix_len + int(input_lens[i]), which could lead to inaccuracies. The fix now uses len(re_encoded_sequence) to ensure the input length accurately reflects the actual token count of the re-encoded prompt.

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 fixes the calculation of total_input_len in the RandomDataset's sampling logic, ensuring the prompt_len in SampleRequest accurately reflects the token length of the generated prompt after re-encoding and truncation. This is an important bugfix for the accuracy of the benchmark's reported metrics.

[gemini-code-assist]

The total_input_len is now correctly calculated as the length of re_encoded_sequence. This ensures that the prompt_len in SampleRequest accurately reflects the token length of the generated prompt after re-encoding and truncation.