[CI] Enable Qwen3-Omni Performance Benchmark

[sammysun0711]

Motivation

This PR aim to enable Qwen3-Omni-Instruct performance benchmark in CI.

Modifications

Accuracy Tests

Benchmarking and Profiling

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.

[Copilot]

Pull request overview

This PR enables performance benchmarking for the Qwen3-Omni model in the CI pipeline. It adds a new performance benchmark step to the GitHub Actions workflow and configures the benchmark parameters to test with multiple images at higher resolution.

Key Changes:

• Added performance benchmark workflow step for Qwen3-Omni in GitHub Actions
• Updated server launch configuration with new environment variables and adjusted memory/prefill parameters
• Modified benchmark parameters to test with 20 images at 960x1280 resolution instead of 1 image at 800x800

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
scripts/ci/sglang_benchmark_workflow.sh	Added ROCm environment variables, adjusted chunked-prefill and max-prefill-tokens sizes, reduced cuda-graph-max-bs, and updated performance benchmark parameters for Qwen3-Omni
.github/workflows/sglang_benchmark_workflow.yaml	Added new "Run performance benchmark" step that executes the performance benchmark for Qwen3-Omni model

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.