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

llm-torch-profiler-analysis

Analyze LLM torch-profiler traces to produce kernel, overlap, and fuse tables.

721|65|Updated Apr 1, 2026
Search Tags:#analysis#llm#triage#overlap#fusion#torch-profiler#kernel-table
One-click install
npx skills add https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS --skill llm-torch-profiler-analysis
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: llm-torch-profiler-analysis
Source: https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS/tree/main/skills/llm-torch-profiler-analysis
Command: npx skills add https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS --skill llm-torch-profiler-analysis

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill provides a standardized triage view for LLM torch-profiler traces across multiple backends, producing three tables that help identify kernel bottlenecks, potential overlaps, and fusion opportunities.

Core Features & Use Cases

  • Unified three-table triage for kernel timing, overlap opportunities, and fuse-pattern discovery from a single trace or live profiling session.
  • Supports single-trace analysis from trace.json(trace.gz) files or trace directories, and two-trace analysis using mapping and formal traces captured from running servers.
  • Generates actionable guidance for performance optimization and architecture decisions.

Quick Start

Run the triage against a local trace or start a live profiling session against a running server and read the three output tables.

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 identify performance bottlenecks in LLM torch-profiler traces?▼

To identify performance bottlenecks in LLM torch-profiler traces, parse single trace.json files or directories to generate a three-table triage report highlighting kernel timing, overlap opportunities, and fusion patterns.

Can I analyze dual-trace workflows from running servers using torch-profiler?▼

Yes, you can analyze dual-trace workflows using mapping and formal traces captured from running servers to produce unified triage results for kernel bottlenecks and fusion opportunities across multiple backends.

What is the best way to discover kernel fusion opportunities from a trace directory?▼

The best way to discover kernel fusion opportunities from a trace directory is to run a unified triage analysis that parses the trace data and renders a dedicated fuse table for optimization guidance.

Does torch-profiler triage support both local trace files and live profiling sessions?▼

Yes, torch-profiler triage supports both local single-trace analysis from trace.json or trace.gz files and live profiling sessions against running servers to render kernel, overlap, and fuse tables.

How do I find overlap opportunities in my LLM trace.json file?▼

To find overlap opportunities in an LLM trace.json file, run the triage analysis to parse the trace and render an overlap table that identifies potential performance overlaps across backends.

Related Skills

Explore More Software Engineering Skills→
annealing-inversionannealing-inversion

llm-torch-profiler-analysis

Analyze LLM torch-profiler traces into kernel, overlap, and fuse pattern reports.

Community
Advanced
dyyounggdyyoungg

llm-torch-profiler-analysis

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

Community
Advanced
FutureMLS-LabFutureMLS-Lab
556

sglang-torch-profiler-analysis

Analyze SGLang torch-profiler traces to identify dominant GPU kernels and fusion patterns.

Official
Advanced
sqjiansqjian

sglang-torch-profiler-analysis

Analyze SGLang torch-profiler traces for kernel overlaps and fusion opportunities.

Community
Intermediate