model-pruning

Reduce LLM size and accelerate inference using pruning techniques like Wanda and SparseGPT. Use when compressing models without retraining, achieving 50% sparsity with minimal accuracy loss, or enabling faster inference on hardware accelerators. Covers unstructured pruning, structured pruning, N:M sparsity, magnitude pruning, and one-shot methods.

Quality

75%

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

Fix and improve this skill with Tessl

tessl review fix ./skills/emerging-techniques/model-pruning/SKILL.md

Quality

Content

57%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.

The body is highly actionable with concrete code across the main pruning methods, but it is verbose, re-explains basic concepts, inlines material that should live in reference files, and lacks validation checkpoints for its destructive pruning operations.

Suggestions

Move the duplicated Wanda implementation and the strategy/deployment deep-dive into references/ (and link them from the body) to remove the ~150 lines of redundancy with references/wanda.md and cut basic concept explanations Claude already knows.

Add an inline validation checkpoint in the production pipeline (e.g. verify achieved sparsity and run a quick perplexity sanity check before saving) with a fix-and-retry loop, so the destructive prune step is guarded.

Replace non-executable snippets with runnable equivalents: drop the fake `from sparsegpt import SparseGPT` import, define or stub `load_calibration_data()`, and fix the evaluation snippet so it does not subtract result dicts.

Dimension	Reasoning	Score
Conciseness	The ~485-line body re-explains basic concepts Claude already knows (magnitude pruning, structured vs unstructured) and duplicates the full wanda_prune implementation already present in references/wanda.md, so while much of the code is useful, several sections are padded and could be tightened.	3 / 5
Actionability	Provides multiple concrete, mostly copy-paste-ready functions (wanda_prune, nm_prune, production pipeline, evaluation), but a few examples are non-executable: `from sparsegpt import SparseGPT` is not a real importable package, `load_calibration_data()` is undefined, and `weight^2 / diag(Hessian)` is pseudocode.	4 / 5
Workflow Clarity	The production pipeline is sequenced (load → calibrate → prune → fine-tune → save) and ends with an evaluation section, but pruning is a destructive weight-modifying operation with no inline validate-before-save checkpoint or fix-and-retry feedback loop, so workflow clarity is capped at 3 per the destructive-operation rule.	3 / 5
Progressive Disclosure	Section headers give some structure, but a large volume of content that belongs in separate reference files (full strategy implementations, comparison tables, core-concept explanations) is inlined, and the existing references/wanda.md is never linked from the body while duplicating its code.	3 / 5
	Total	13 / 20 Passed

Description

92%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.

The description is specific, complete, and distinct: it states concrete capabilities, an explicit Use-when trigger clause, and names specialized techniques that separate it from generic compression skills. Its only weakness is slightly incomplete coverage of natural user synonyms.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions ("Reduce LLM size", "accelerate inference", "compressing models without retraining", "achieving 50% sparsity", "enabling faster inference on hardware accelerators") plus five named pruning techniques, giving comprehensive coverage of the domain.	5 / 5
Completeness	Explicitly answers "what" (reduce size / accelerate inference via Wanda and SparseGPT pruning) and "when" with a concrete "Use when compressing models without retraining, achieving 50% sparsity..., or enabling faster inference on hardware accelerators" trigger clause.	5 / 5
Trigger Term Quality	Good natural keyword coverage ("compressing models", "without retraining", "50% sparsity", "faster inference", "hardware accelerators", "pruning"), but a few common user phrasings (e.g. "make the model smaller", "speed up inference") and synonyms are missing.	4 / 5
Distinctiveness Conflict Risk	Names specialized methods (Wanda, SparseGPT, N:M sparsity) that establish a clear niche with distinct triggers and minimal overlap risk with other skills.	5 / 5
	Total	19 / 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/emerging-techniques/model-pruning/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.