"Evaluates exported MMDeploy models, profiles latency, runs
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill validation --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Validation?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-validation)More formats (shields.io, HTML) on the badges page.
---
name: validation
description: "Evaluates exported MMDeploy models, profiles latency, runs
regression matrices, and generates backend coverage tables."
disable-model-invocation: true
metadata:
disco-role: operating
license: Apache 2.0
---
# Validation
Use this sub-skill when a task asks to evaluate an exported MMDeploy backend model, measure inference speed, run a regression matrix, build a supported-backend table, or interpret validation/profiling reports.
## Route by task
- **Metric-aware backend evaluation:** read [evaluation](references/evaluation.md), then use [scripts/test.py](scripts/test.py) when the user needs dataset metrics, visualized outputs, speed-test logging during evaluation, backend model-file selection, or `--cfg-options` overrides.
- **Latency-only profiling:** read [profiling](references/profiling.md), then use [scripts/profiler.py](scripts/profiler.py) when the user asks for warmup/iteration-controlled latency or custom batch-size/shape timing without evaluator metrics.
- **Maintainer regression matrix:** read [regression](references/regression.md), then use [scripts/regression_test.py](scripts/regression_test.py) for codebase/backend/model filters, conversion-only versus precision modes, checkpoint caching, and merged report workbooks.
- **Supported model/backend table:** read [supported model tables](references/supported-model-tables.md), then use [scripts/generate_md_table.py](scripts/generate_md_table.py) to turn regression matrix YAML into Markdown support tables.
- **Unexpected failure, skip, or slow run:** read [troubleshooting](references/troubleshooting.md) before classifying a problem as a model defect.
## Guardrails
- Use `test.py` for validation/evaluation and optional speed logs; use `profiler.py` for latency-only measurements. Do not treat profiler output as metric evidence.
- Regression runs can download checkpoints, call conversion workflows, and exercise heavyweight backends. Always narrow `--codebase`, `--backends`, and `--models` unless the user explicitly requests a full matrix.
- Backend installation/build problems belong to the backend workflow. In this validation sub-skill, record whether a backend is unavailable, skipped, failed, or blocked, and route installation remediation elsewhere.
- Graph rewrite or optimizer implementation changes belong to the extensibility workflow. Use validation here only to observe downstream metric/latency/report impact.

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!