skills.rest
hey-kong
hey-kongCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

sglang-sota-performance

Coordinate benchmarking, profiling, patching, and revalidation across SGLang, vLLM, and TensorRT-LLM.

Updated Jun 29, 2026
Search Tags:#workflow#performance#benchmarking#reproducibility#profiler#sglang#kernel-analysis
One-click install
npx skills add https://github.com/hey-kong/sglang --skill sglang-sota-performance-hey-kong
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: sglang-sota-performance
Source: https://github.com/hey-kong/sglang/tree/main/.claude/skills/sglang-sota-performance
Command: npx skills add https://github.com/hey-kong/sglang --skill sglang-sota-performance-hey-kong

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Coordinate an end-to-end performance optimization loop for a single model across frameworks to determine best deployment and patch opportunities.

Core Features & Use Cases

  • End-to-end optimization loop across frameworks (SGLang, vLLM, and TensorRT-LLM) to find the fastest, most reliable deployment.
  • Automated benchmarking, profiling, and patch validation with reproducible artifact layouts for auditability.
  • Structured, reusable workflow that guides diagnosis from root-cause analysis to patch validation and revalidation.

Quick Start

Run the SOTA loop for a single model to compare frameworks, profile SGLang, patch as needed, and revalidate.

Frequently Asked Questions about sglang-sota-performance

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

FAQPage Schema
How do I optimize LLM deployment performance across vLLM and TensorRT-LLM?▼

To optimize LLM deployment performance, you can run an end-to-end loop that benchmarks, profiles, and patches models across frameworks like SGLang, vLLM, and TensorRT-LLM to find the fastest deployment configuration.

What is the best way to analyze kernel bottlenecks during model serving?▼

Analyzing kernel bottlenecks involves using a structured profiling workflow that diagnoses root causes in target hardware workloads and guides reproducible patching to resolve performance limitations.

Can I compare SGLang and vLLM performance for a specific hardware workload?▼

Yes, you can compare SGLang and vLLM performance by providing a model id, hardware details, and workload inputs to coordinate automated benchmarking and determine the best deployment command.

How do I ensure reproducibility when tuning model performance?▼

To ensure reproducibility when tuning model performance, use an artifact layout to organize runs, logs, and patches, enabling auditability and structured revalidation of performance improvements.

What inputs are required to start end-to-end performance profiling for a single model?▼

Starting end-to-end performance profiling requires inputs like model id, hardware details, workload specifications, frameworks to compare, and an artifact layout to organize runs, logs, and patches.

Related Skills

Explore More Software Engineering Skills→
annealing-inversionannealing-inversion

sglang-sota-performance

Benchmark SGLang against vLLM and TensorRT-LLM, then patch bottlenecks.

Community
Advanced
handsomelong922handsomelong922

serving-llms-vllm

Deploy scalable LLM serving with vLLM for OpenAI-compatible endpoints.

Community
Advanced
BBufBBuf
721

sglang-sota-performance

Benchmark models across SGLang, vLLM, and TensorRT-LLM to identify performance gaps.

Community
Advanced
n0ann0an
103

xcode-build-orchestrator

Benchmark Xcode builds, analyze with specialist skills, and apply approved fixes.

Community
Advanced