Skip to content
Back to skills

Profiling Debugging

ASecurity

"Profile and debug ExecuTorch execution with ETRecord, ETDump,

  • 247 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added September 8, 2026
toolspythonbashdebuggingapibackendperformance

Works with

  • api

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned September 8, 2026

npx -y skills add VectorSpaceLab/AREX-Skill --skill profiling-debugging --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Profiling Debugging?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Profiling Debugging
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vectorspacelab-profiling-debugging/badge)](https://www.skillsdirectory.com/skills/vectorspacelab-profiling-debugging)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: profiling-debugging
description: "Profile and debug ExecuTorch execution with ETRecord, ETDump,
  Inspector, visualization, delegate debug data, intermediate outputs, and
  runtime/export triage."
disable-model-invocation: true
metadata:
  disco-role: operating
license: NOASSERTION
---

# profiling-debugging

Use this sub-skill when the user asks about ExecuTorch profiling, ETDump, ETRecord, Inspector, delegate debugging, model visualization, memory-planning inspection, intermediate outputs, or runtime/export accuracy/performance triage.

## Route Here For

- Enabling and collecting ETDump at runtime.
- Generating ETRecord during export and using it to map runtime events to graph/source/module hierarchy.
- Loading artifacts with `executorch.devtools.Inspector`.
- Comparing delegate and CPU outputs, inspecting numerical divergence, and deciding when to route QNN-specific per-layer debugging to `../qualcomm/SKILL.md`.
- Understanding memory planning and visualization artifacts.

## Fast Path

1. Ask which artifacts the user already has: `.pte`, ETRecord, ETDump, debug buffer, eager reference outputs, backend logs.
2. If no ETRecord exists and source model/export is available, regenerate export with ETRecord enabled before collecting runtime data.
3. If only ETDump exists, perform raw timing inspection but explain that source/module attribution may be limited.
4. Use [profiling workflows](references/profiling-workflows.md) and [API reference](references/api-reference.md).
5. For artifact sanity checks, run:

```bash
python scripts/inspect_etdump_summary.py --etdump path/to/run.etdp --etrecord path/to/model.etrecord
```

The helper checks file presence and Inspector import availability; it does not parse private model data unless the installed package supports it.

## Cross-Links

- Export and ETRecord generation details: `../export-runtime/SKILL.md`.
- QNN-specific intermediate output debugger: `../qualcomm/SKILL.md`.
- Backend choice and fallback behavior: `../backend-selection/SKILL.md`.

Files in this skill

  • SKILL.md2 KB
  • references/api-reference.md984 B
  • references/profiling-workflows.md1.5 KB
  • references/troubleshooting.md988 B
  • scripts/inspect_etdump_summary.py1.4 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.