gptq

Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

72%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The content is highly actionable and well-structured with genuine progressive disclosure, but it runs long and omits validation checkpoints in its quantization workflow. Tightening benchmark/model-list material and adding an explicit verify-quantization step would lift the two weaker dimensions.

Suggestions

Add an explicit validation/verification checkpoint in the 'Quantize your own model' workflow (e.g., reload the saved model and run a quick perplexity or generation sanity check before pushing).

Move the large benchmark tables and 'Supported models' list into a reference file (e.g. references/benchmarks.md), keeping only a one-line summary in SKILL.md to improve conciseness.

Consolidate the three near-duplicate BaseQuantizeConfig blocks into a single configurable example with a small parameter trade-off table, reducing repetition.

Dimension	Reasoning	Score
Conciseness	The ~445-line body is mostly efficient and free of basic-concept fluff, but large benchmark tables, a supported-models list, and near-duplicate config blocks could be tightened or pushed to references.	2 / 3
Actionability	Executable, copy-paste-ready code appears throughout — install commands, load/quantize/kernels/QLoRA/multi-GPU examples — with concrete model names and parameters.	3 / 3
Workflow Clarity	The quantize-your-own-model flow is sequenced clearly, but there are no explicit validation/verification checkpoints for an operation that produces model artifacts, which caps the score at 2.	2 / 3
Progressive Disclosure	A clear overview links via markdown to three real one-level-deep references (calibration.md, integration.md, troubleshooting.md), each verified to exist, with content appropriately split.	3 / 3
	Total	10 / 12 Passed

Description

100%

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is strong across all dimensions: it states concrete capabilities, natural trigger terms, and clear when-to-use guidance while occupying a distinct niche. It is one of the better-formed skill descriptions, with only minor technical jargon (model size tokens, 'perplexity degradation') that slightly limits lay-user trigger resonance.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — '4-bit quantization', '4× memory reduction', 'faster inference (3-4× speedup)', 'QLoRA fine-tuning' — matching the comprehensive-actions anchor.	3 / 3
Completeness	Explicitly answers both what ('Post-training 4-bit quantization for LLMs with minimal accuracy loss') and when ('Use for deploying large models... when you need 4× memory reduction... or for faster inference').	3 / 3
Trigger Term Quality	Covers natural terms a user would say ('quantization', 'consumer GPUs', '4-bit', 'memory reduction', 'faster inference', 'QLoRA', 'fine-tuning') with good breadth.	3 / 3
Distinctiveness Conflict Risk	Occupies a clear niche (GPTQ 4-bit post-training quantization) with distinct triggers and an explicit 'Known Conflicts' note, making misfiring unlikely.	3 / 3
	Total	12 / 12 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/gptq/SKILL.md
Commit: 52845c3

Reviewed: about 10 hours ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.