[Benchmark] DiT models Performance benchmark(T2I/I2I/T2V/TI2V)

[david6666666]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

support RFC #344

Test Plan

vllm serve Tongyi-MAI/Z-Image-Turbo --omni --port 8091 --cache-backend cache_dit
python3 benchmarks/benchmark_serving.py \
        --dataset vbench --task t2i --num-prompts 10 \
        --height 1024 --width 1024

vllm serve Qwen/Qwen-Image-Edit  --omni --port 8091 --cache-backend cache_dit
python3 benchmarks/benchmark_serving.py \
       --dataset vbench --task i2i --num-prompts 10 

vllm serve Wan-AI/Wan2.2-T2V-A14B-Diffusers --omni --port 8091 --cache-backend cache_dit
python3 benchmarks/benchmark_serving.py \
    --dataset vbench --task t2v --num-prompts 10 \
    --height 480 --width 640 --fps 16 --num-frames 80

Test Result

---

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.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[david6666666]

TOTO:

• refer [Bug]: Filename Mismatch, Incorrect Device Assignment, and Poor UX in Qwen2.5-Omni Example (2×H100 Setup) #370, Integrate tqdm or a native vLLM-style progress bar in the benchmark scripts
• Add t2i/t2v/i2i/i2v dataset benchmark

[david6666666]

close for test

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Guard empty dataset before resizing

When no I2V data is found and PIL is missing, _create_dummy_data() returns an empty list, which then flows into _resize_data() and triggers a ZeroDivisionError at self.args.num_prompts // len(data). This means running the benchmark without a dataset path on a machine lacking PIL (or with an empty dataset directory) will crash before any requests are sent. Consider handling len(data) == 0 explicitly (e.g., return early or raise a clearer error) to avoid this failure.

Useful? React with 👍 / 👎.

[david6666666]

@hsliuustc0106 @ZJY0516 PTAL

[ZJY0516]

We should rename this file name to diffusion_benchmark_serving.py or something else

[david6666666]

pr ready ptal, thx