sglang-diffusion-performance

Benchmark and tune SGLang diffusion performance configurations across GPUs and VRAM budgets.

Updated May 14, 2026
One-click install
npx skills add https://github.com/sqjian/sglang --skill sglang-diffusion-performance-sqjian
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: sglang-diffusion-performance
Source: https://github.com/sqjian/sglang/tree/main/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-performance
Command: npx skills add https://github.com/sqjian/sglang --skill sglang-diffusion-performance-sqjian

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Benchmark and tune diffusion performance configurations for SGLang.

Core Features & Use Cases

  • Benchmark native vs diffusers backends to identify best-performing pipelines.
  • Guidance for GPU/VRAM budgets and multi-GPU setups to optimize latency and memory.
  • Use Case: quickly select flags and environment settings that maximize throughput for a given model.

Quick Start

Run a diffusion performance benchmark for your model and GPU to identify the best flags and batch settings.

Frequently Asked Questions about sglang-diffusion-performance

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I benchmark diffusion performance in SGLang to find the best GPU settings?

Benchmark diffusion performance in SGLang by running native and diffusers backend comparisons to identify the fastest pipelines. This process applies environment variables and documented flag recommendations to maximize throughput for your specific model and GPU setup.

What is the best way to optimize diffusion model latency and VRAM usage for multi-GPU setups?

Optimize diffusion model latency and VRAM by tuning performance configurations for multi-GPU setups within SGLang. The benchmarking process evaluates different batch settings and flags to identify configurations that maximize memory efficiency and throughput.

Does SGLang support benchmarking native backends against diffusers baselines for diffusion models?

Yes, SGLang supports benchmarking native backends against diffusers baselines for diffusion models. This comparison identifies the best-performing pipelines by applying reproducible environment setups and documented flag recommendations.

How do I select the right flags to maximize throughput for my diffusion model on a specific GPU?

Select flags to maximize throughput by running diffusion performance benchmarks tailored to your model and GPU. The benchmark evaluates various configurations, including batch settings and environment variables, to ensure optimal latency and memory usage.

Do I need specific environment setup scripts to ensure reproducible diffusion benchmarking in SGLang?

Yes, you need environment setup scripts and environment variables to ensure reproducible diffusion benchmarking in SGLang. These scripts establish the baseline configurations required to accurately compare native vs diffusers backends across various GPUs and VRAM budgets.

Why does my diffusion performance benchmark fail to reflect expected VRAM limits during multi-GPU testing?

Diffusion performance benchmarks may fail to reflect expected VRAM limits if environment variables and flag recommendations are not correctly applied. Ensuring reproducible benchmarking requires proper environment setup scripts tailored to your specific GPU and multi-GPU configurations.