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

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.

A dense, highly actionable reference with executable code and good progressive disclosure, weakened mainly by missing validation/feedback checkpoints in the merge and training workflows and some verbosity in benchmark sections.

Suggestions

Add an explicit validation gate to the merge workflow (e.g., evaluate the adapter on held-out data and confirm metrics before merge_and_unload) so destructive merge/deploy steps have a checkpoint.

Add a validate→fix→retry loop to the CUDA OOM troubleshooting sequence (try fix, re-run training, confirm memory headroom before proceeding) rather than presenting fixes as a flat list.

Tighten or move the three benchmark tables (memory, speed, MMLU) into references/ to reduce token cost on the main page while keeping the headline trade-offs.

Dimension	Reasoning	Score
Conciseness	Mostly efficient with code and tables rather than concept tutorials, but the long benchmark sections and repeated config patterns across methods (IA3, Prefix Tuning) add tokens that could be tightened.	2 / 3
Actionability	Fully executable code with concrete LoraConfig, BitsAndBytesConfig, and TrainingArguments snippets plus exact parameter values and install commands, copy-paste ready rather than pseudocode.	3 / 3
Workflow Clarity	Decision trees and sequenced quick-start are clear, but the merge/deploy and training workflows lack explicit validation checkpoints (e.g., evaluate-before-merge gate), and the rubric caps batch/transform operations at 2 when feedback loops are missing.	2 / 3
Progressive Disclosure	Overview-shaped body with two clearly signaled, one-level-deep references (advanced-usage.md, troubleshooting.md) that are real files, with detail appropriately pushed off the main page.	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.

A strong, third-person description that states concrete capabilities and explicit use-when triggers tied to GPU memory and adapter serving, with good natural keyword coverage and low conflict risk.

Dimension	Reasoning	Score
Specificity	Names concrete methods (LoRA, QLoRA) and specific scenarios ('7B-70B', 'train <1% of parameters', 'multi-adapter serving') rather than vague language, matching the multiple-specific-actions anchor.	3 / 3
Completeness	Explicitly answers both what ('Parameter-efficient fine-tuning... using LoRA, QLoRA, and 25+ methods') and when via a clear 'Use when...' clause with multiple triggers.	3 / 3
Trigger Term Quality	Covers natural user terms ('fine-tuning', 'LoRA', 'QLoRA', 'limited GPU memory', 'multi-adapter serving') a user would actually say, with common variations, beyond a single keyword.	3 / 3
Distinctiveness Conflict Risk	Occupies a clear niche (PEFT/adapter fine-tuning on memory-constrained GPUs) with distinct triggers unlikely to fire for unrelated training skills.	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/peft/SKILL.md
Commit: 52845c3

Reviewed: about 14 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.