peft-fine-tuning

Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you need to train <1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.

Quality

93%

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

87%

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

A high-quality, executable reference skill with lean code-first content and clean one-level progressive disclosure. Its only meaningful gap is that the core training workflow lacks embedded validation checkpoints, relying on scattered troubleshooting guidance instead.

Suggestions

Add an explicit validation checkpoint inside the quick-start training flow (e.g., 'After trainer.train(), load the saved adapter with PeftModel.from_pretrained and run a sanity generate() before merging') so the validate-then-proceed pattern is sequenced rather than implied.

Convert the 'Common issues' troubleshooting entries into a short feedback-loop format (symptom -> check -> fix -> re-run) tied to the main workflow steps, turning scattered guidance into explicit error-recovery checkpoints.

Insert a brief 'Verify before merging' gate in the merge section that pairs with the existing 'Evaluate on held-out data' best practice, since merging is the irreversible step in this workflow.

Dimension	Reasoning	Score
Conciseness	Dense, code-first reference content with terse, operationally useful comments and no explaining of concepts Claude already knows; comparison tables and benchmarks earn their tokens as actionable reference data.	3 / 3
Actionability	Provides complete, copy-paste-ready Python and YAML for LoRA, QLoRA, IA3, prefix tuning, merging, multi-adapter serving, and TRL/Axolotl/vLLM integration — fully executable, not pseudocode.	3 / 3
Workflow Clarity	The quick-start recipe is a clear linear sequence with one inline check (print_trainable_parameters showing expected 0.17%), but verifications are scattered across 'Common issues'/'Best practices' rather than embedded as explicit validate-then-proceed checkpoints in the flow.	2 / 3
Progressive Disclosure	SKILL.md is a concise overview with well-signaled, one-level-deep markdown links to references/advanced-usage.md and references/troubleshooting.md; both files exist and contain no nested cross-references, so content is appropriately split with easy navigation.	3 / 3
	Total	11 / 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.

A strong, specific description with an explicit 'Use when...' trigger clause, natural keywords, and a distinct niche. It answers both what the skill does and when to invoke it without fluff or over-claims.

Dimension	Reasoning	Score
Specificity	Names concrete capabilities and methods — 'Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods' plus 'multi-adapter serving' — matching the multi-action anchor.	3 / 3
Completeness	Clearly states what it does (PEFT via LoRA/QLoRA/25+ methods) and includes an explicit 'Use when fine-tuning large models... when you need to train <1% of parameters... or for multi-adapter serving' trigger clause, answering both what and when.	3 / 3
Trigger Term Quality	Uses natural terms a user would say when requesting this — 'fine-tuning large models (7B-70B)', 'limited GPU memory', 'LoRA', 'QLoRA', 'multi-adapter serving' — with good coverage of variations.	3 / 3
Distinctiveness Conflict Risk	A clear niche (parameter-efficient fine-tuning) with distinct, specific triggers unlikely to fire for unrelated skills; third-person voice with no first/second-person phrasing.	3 / 3
	Total	12 / 12 Passed

Validation

81%

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 — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning
metadata_field	'metadata' should map string keys to string values	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	13 / 16 Passed

Repository: NousResearch/hermes-agent
Path: optional-skills/mlops/peft/SKILL.md
Commit: 4c9628e

Reviewed: about 13 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.