skills.rest
yo-steven
yo-stevenCOMMUNITY
Explore all skills by this author →
GitHub

ad-graph-dump

Dump TensorRT-LLM AutoDeploy FX graphs after each registered transform.

Updated May 23, 2026
Search Tags:#environment variables#tensorrt-llm#autodeploy#fx graph#graph debugging#fusion verification#distributed rank-0
One-click install
npx skills add https://github.com/yo-steven/skills-exploration-20260522 --skill ad-graph-dump
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ad-graph-dump
Source: https://github.com/yo-steven/skills-exploration-20260522/tree/main/skills/TensorRT-LLM/ad-graph-dump
Command: npx skills add https://github.com/yo-steven/skills-exploration-20260522 --skill ad-graph-dump

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Ad-graph-dump helps you debug TensorRT-LLM AutoDeploy by making the intermediate FX graph structure visible after each transform, so you can confirm what rewrites happened and where a specific subgraph came from.

Core Features & Use Cases

  • Per-transform FX graph dumps: Writes a text dump after every registered AutoDeploy transform runs, capturing the module’s graph state immediately after that transform.
  • Deterministic file naming for pipeline order: Produces lexicographically sortable filenames that track run order and transform stage/keys for easy before/after comparisons.
  • Rank-aware safe dumping: Only rank 0 writes files, avoiding duplicated outputs in distributed runs.
  • Actionable for fusion and subgraph verification: Use dumps to verify fusion matches, check whether wrapper/metadata nodes appear or disappear, and correlate transform logs to on-disk graph output.

Quick Start

Set AD_DUMP_GRAPHS_DIR to a writable directory, then run your AutoDeploy entrypoint so that the pipeline emits numbered graph dump files after each transform.

Frequently Asked Questions about ad-graph-dump

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

FAQPage Schema
How do I debug TensorRT-LLM AutoDeploy FX graph rewrites?▼

Debug TensorRT-LLM AutoDeploy FX graph rewrites by setting the AD_DUMP_GRAPHS_DIR environment variable to a writable directory, which triggers text dumps of the graph after each registered transform for verification.

How can I verify fusion and subgraph changes in TensorRT-LLM AutoDeploy?▼

Verify fusion and subgraph changes in TensorRT-LLM AutoDeploy by inspecting the dumped FX graphs, allowing you to check whether wrapper or metadata nodes appear or disappear after specific transforms.

Does AutoDeploy graph dumping work in distributed multi-rank environments?▼

AutoDeploy graph dumping works safely in distributed environments by restricting file output to rank 0 only, preventing duplicated graph dumps across multiple ranks during a distributed run.

How do I track the execution order of AutoDeploy transforms on disk?▼

Track the execution order of AutoDeploy transforms through deterministic, lexicographically sortable filenames that encode run order and transform stage keys for straightforward before and after comparisons.

Related Skills

Explore More Software Engineering Skills→
yo-stevenyo-steven

ad-add-fusion-transformation

Add TensorRT-LLM AutoDeploy fusion transformation passes for graph pattern rewriting.

Community
Advanced
DJRHailsDJRHails
2

import-papers

Extract metadata from PDFs and generate BibTeX entries.

Community
Advanced
yo-stevenyo-steven

ad-conf-check

Verify applied AutoDeploy YAML config keys against TensorRT-LLM server logs.

Community
Advanced
Devsoul2026Devsoul2026
150

tensorrt-llm

Optimize and deploy LLM inference on NVIDIA GPUs with TensorRT-LLM.

Community
Advanced