hqq-quantization

Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for fast quantization workflows, or when deploying with vLLM or HuggingFace Transformers.

Quality

75%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./skills/optimization/hqq/SKILL.md

Quality

Content

67%Weight 40%Scale 1-5

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

A comprehensive, highly actionable skill body with strong executable examples and clear workflow sequencing, supported by real reference files. Its main weakness is conciseness — overlapping examples inflate the body — and minor actionability gaps from undefined variables in a couple of snippets.

Suggestions

Consolidate redundant quantization examples: keep one canonical BaseQuantizeConfig/HQQLinear snippet and let the others reference it, removing overlap between 'Basic quantization', 'Core concepts', and 'Workflow 1'.

Define or scaffold placeholder variables in Workflow 2 and the PEFT section (e.g. `tokenizer`, `train_dataset`, `input_tensor`) so examples are fully self-contained and copy-paste runnable.

Move the backends selection table and per-layer mixed-precision config detail into references/advanced-usage.md, keeping the body to a concise overview pointer.

Dimension	Reasoning	Score
Conciseness	Mostly efficient and free of basic-concept padding, but the ~434-line body has notable redundancy — the basic quantization, core-concepts config, and Workflow 1 examples overlap heavily, and several integration sections repeat similar loading boilerplate.	3 / 5
Actionability	Abundant copy-paste-ready executable code covering installation, quantization, HF/vLLM/PEFT integration, and troubleshooting; docked one point because a few examples reference undefined variables (e.g. `input_tensor`, `tokenizer` in Workflow 2, `train_dataset`).	4 / 5
Workflow Clarity	Workflows 1 and 2 are clearly numbered sequences with a verify-quality checkpoint and a benchmarking step, plus a structured 'Common issues' section; minor gaps are the missing explicit validate-then-retry feedback loop and undefined variables in Workflow 2.	4 / 5
Progressive Disclosure	Good section structure with two real, well-signaled one-level-deep references (references/advanced-usage.md, references/troubleshooting.md); docked one point because substantial API-style content (backends table, mixed-precision and per-layer configs) is inlined and could live in the advanced reference.	4 / 5
	Total	15 / 20 Passed

Description

83%Weight 40%Scale 1-5

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

A strong, third-person description that pairs a clear capability statement with an explicit multi-trigger 'Use when' clause. It is specific and well-differentiated, with only minor gaps in action coverage and natural-term synonyms.

Dimension	Reasoning	Score
Specificity	Names the domain (Half-Quadratic Quantization for LLMs) and several concrete actions ('quantizing models to 4/3/2-bit precision', 'fast quantization workflows', 'deploying with vLLM or HuggingFace Transformers'), with only minor coverage gaps such as PEFT/saving.	4 / 5
Completeness	Explicitly answers both 'what' ('Half-Quadratic Quantization for LLMs without calibration data') and 'when' via a clear 'Use when...' clause with multiple concrete trigger phrases.	5 / 5
Trigger Term Quality	Includes natural user-facing terms ('quantizing models', '4/3/2-bit precision', 'calibration datasets', 'vLLM', 'HuggingFace Transformers') but omits common synonyms like 'compression' or 'memory reduction' that users might say.	4 / 5
Distinctiveness Conflict Risk	The 'without calibration data' framing carves a clear niche distinct from GPTQ/AWQ, though the broad 'quantizing models' phrasing carries minor overlap risk with bitsandbytes/GPTQ skills.	4 / 5
	Total	17 / 20 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: OpenLAIR/dr-claw
Path: skills/optimization/hqq/SKILL.md
Commit: 4b35756

Reviewed: 1 day 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.