skills.rest
QingHuan-6
QingHuan-6COMMUNITY
Explore all skills by this author →
GitHub

llm-torch-profiler-analysis

Triage torch-profiler traces into kernel, overlap, and fuse tables across LLM frameworks.

Updated May 30, 2026
Search Tags:#LLM#triage#overlap#fusion#kernel-analysis#torch-profiler#TensorRT-LLM
One-click install
npx skills add https://github.com/QingHuan-6/SglangGen --skill llm-torch-profiler-analysis-qinghuan-6
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: llm-torch-profiler-analysis
Source: https://github.com/QingHuan-6/SglangGen/tree/main/.claude/skills/llm-torch-profiler-analysis
Command: npx skills add https://github.com/QingHuan-6/SglangGen --skill llm-torch-profiler-analysis-qinghuan-6

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires triage_kernel_helpers, triage_overlap_helpers, profile_common, and includes scripts (resource) components.

What problem does it solve?

This Skill provides a unified triage report for LLM torch-profiler traces across SGLang, vLLM, and TensorRT-LLM, helping engineers quickly identify kernel, overlap, and fuse opportunities.

Core Features & Use Cases

  • Single-trace triage: generate a three-table diagnostic from one trace for quick insight into kernel performance, overlap potential, and fuse opportunities.
  • Two-trace multi-backend analysis: compare mapping and formal traces to attribute performance patterns across frameworks.
  • Live profiling support: drive live profiling against a running server and summarize results into actionable guidance.
  • Output format: three tables (kernel, overlap-opportunity, fuse-pattern) plus a concise summary of dominant patterns and recommended next steps.

Quick Start

Run the unified analysis on an existing trace or live server to generate the three-trace tables and a compact summary.

Frequently Asked Questions about llm-torch-profiler-analysis

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

FAQPage Schema
How do I triage torch-profiler traces to find kernel and overlap opportunities for LLM inference?▼

Triage torch-profiler traces by analyzing them to produce structured kernel, overlap, and fuse insights. This generates three diagnostic tables and a summary of dominant optimization opportunities for frameworks like SGLang, vLLM, and TensorRT-LLM.

Can I compare torch-profiler performance patterns across vLLM and TensorRT-LLM?▼

Yes, you can compare performance patterns across vLLM and TensorRT-LLM using two-trace multi-backend analysis. It attributes performance differences by comparing mapping and formal traces from a single profiling session.

What is the best way to analyze live-server profiling results for LLM workloads?▼

Analyze live-server profiling results by driving live profiling against a running server. This workflow summarizes live torch-profiler data into actionable guidance and structured diagnostic tables.

Does this torch-profiler triage support both static traces and live profiling workflows?▼

Yes, the torch-profiler triage applies to both static trace files and live-server profiling workflows. It processes single or paired traces to deliver cross-framework comparisons and live diagnostic insights.

What specific optimization patterns are identified when analyzing LLM torch-profiler traces?▼

Analyzing LLM torch-profiler traces identifies kernel performance issues, overlap potential, and fuse opportunities. It outputs three structured tables plus a concise summary of dominant patterns and recommended next steps.

Related Skills

Explore More Software Engineering Skills→
dyyounggdyyoungg

llm-torch-profiler-analysis

Analyze LLM torch-profiler traces to identify kernel, overlap, and fusion patterns.

Community
Advanced
BBufBBuf
721

sglang-sota-performance

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

Community
Advanced
alibabaalibaba
71

gpu-kernel-bottleneck-analysis

Analyzes GPU kernel performance using Roofline models and profiling data.

Official
Advanced
QingHuan-6QingHuan-6

llm-serving-auto-benchmark

Benchmark SGLang, vLLM, and TensorRT-LLM serving to produce Markdown/CSV summaries.

Community
Advanced