skills.rest
BBuf
BBufCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

sglang-glm5-glm51-optimization

Identify and vet GLM-5/5.1 optimization changes in SGLang PR diffs.

721|65|Updated Apr 1, 2026
Search Tags:#validation#optimization#fp8#cookbook#glm-5#glm-5.1#nsa
One-click install
npx skills add https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS --skill sglang-glm5-glm51-optimization
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: sglang-glm5-glm51-optimization
Source: https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS/tree/main/skills/model-optimization/sglang/sglang-glm5-glm51-optimization
Command: npx skills add https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS --skill sglang-glm5-glm51-optimization

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

PR-backed and current-main optimization manual for GLM-5 and GLM-5.1 in SGLang. Use when Codex needs to recover, extend, or audit GLM-5 DSA/NSA/NSA indexer paths, GLM-5.1 FP8/MXFP4/NVFP4, NextN/MTP, dense-attention threshold, NSA TileLang/AITER, tool templates, EAGLE, PCG, AMD/Blackwell/NPU validation, or GLM-5 cookbook recipes.

Core Features & Use Cases

  • PR-backed optimization cards and diffs for GLM-5/5.1.
  • Guidance for runtime tuning, documentation, and validation (EAGLE/PCG, AMD/Blackwell validation, GLM-5 cookbook recipes).
  • Use Case: A developer audits a GLM-5.1 FP8 deployment and aligns it with the main branch to ensure stability across QA and production.

Quick Start

Review the latest PR dossier in references/pr-history.md and apply the recommended changes to the model config and runtime setup.

Frequently Asked Questions about sglang-glm5-glm51-optimization

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

FAQPage Schema
How do I optimize GLM-5.1 FP8 deployment in SGLang?▼

To optimize GLM-5.1 FP8 deployment in SGLang, audit the PR dossier in references/pr-history.md and apply recommended changes to model config and runtime setup. This ensures stability across QA and production by aligning with the main branch.

What is the best way to validate NSA indexing changes for GLM-5?▼

Validating NSA indexing changes for GLM-5 involves using a PR-diff-based change-card workflow with cross-referenced evidence from PR history and runtime validation data. This approach identifies and vets optimization changes effectively.

Does SGLang support GLM-5.1 MXFP4 and NVFP4 quantization?▼

SGLang supports GLM-5.1 optimization with MXFP4 and NVFP4 quantization. The Skill provides PR-backed optimization cards and diffs to guide runtime tuning and configuration for these specific formats.

How do I configure dense-attention thresholds for GLM-5 models?▼

Configuring dense-attention thresholds for GLM-5 models requires reviewing the PR-backed optimization manual. It provides specific guidance on setting thresholds and validating changes across different runtime scenarios.

Can I use this Skill for AMD and Blackwell hardware validation of GLM-5?▼

Yes, this Skill supports AMD, Blackwell, and GB300 validation scenarios for GLM-5. It offers guidance for runtime tuning, documentation, and validation workflows across these hardware platforms.

Why should I use a PR-diff-based workflow for GLM-5 optimization?▼

A PR-diff-based workflow for GLM-5 optimization guarantees cross-referenced evidence from PR history and runtime validation data. This method ensures changes are vetted and aligned with the main branch for stable deployments.

Related Skills

Explore More Software Engineering Skills→
BBufBBuf
721

vllm-glm5-glm51-optimization

Document PR-backed optimization paths for GLM-5 integration in vLLM.

Community
Advanced
ever-oliever-oli
3

serving-llms-vllm

Deploy and manage vLLM servers for high-throughput LLM inference.

Community
Advanced
BBufBBuf
721

sglang-glm45-optimization

Optimize GLM-4.5 MoE loading and runtime tuning in SGLang.

Community
Advanced
BBufBBuf
721

sglang-ernie45-optimization

Identify Ernie4.5-VL optimization practices from PR-diff evidence in SGLang runtime.

Community
Advanced