peft

Fine-tune large LLMs with LoRA on limited GPU memory.

Quality

66%

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 ./optional-skills/mlops/peft/SKILL.md

Quality

Content

76%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 content is highly actionable with comprehensive executable examples and good progressive disclosure through reference files. Its main weaknesses are missing validation/feedback checkpoints in the training and merge workflows and some over-explanation of basic concepts.

Suggestions

Add explicit validation steps to destructive/batch workflows, e.g. 'Verify adapter loaded with model.print_trainable_parameters() before training' and 'Evaluate on held-out data; only merge when metrics pass.'

Trim tutorial-style explanations of concepts Claude already knows (rank/alpha rules, what nf4 means) to tighten conciseness.

Consider moving the methods comparison and per-architecture target_modules tables into a reference file to keep SKILL.md an overview.

Dimension	Reasoning	Score
Conciseness	The body is mostly lean with code blocks and dense tables carrying the content, though a few tutorial-style explanations of concepts Claude already knows (rank scaling, alpha rule of thumb) could be trimmed.	4 / 5
Actionability	It provides fully executable, copy-paste-ready code across LoRA, QLoRA, IA3, Prefix Tuning, merging, multi-adapter serving, and TRL/vLLM/Axolotl integration, covering the common cases with concrete commands and expected outputs.	5 / 5
Workflow Clarity	Sections are sequenced and a best-practices list exists, but the training and merge/push workflows lack explicit validation checkpoints and feedback loops for destructive or batch operations, which caps the score at 3.	3 / 5
Progressive Disclosure	Advanced and troubleshooting material is split into well-signaled, one-level-deep real reference files with descriptions, though the main body remains lengthy and inlines content that could stay overview-level.	4 / 5
	Total	16 / 20 Passed

Description

57%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 concise and names a concrete capability with natural trigger terms, but it lacks an explicit when-to-use clause and broader method coverage. Adding a "Use when..." sentence and synonyms like QLoRA/PEFT would lift completeness and trigger quality.

Suggestions

Add an explicit 'Use when...' clause, e.g. 'Use when fine-tuning LLMs on consumer GPUs or when the user mentions LoRA, QLoRA, or parameter-efficient fine-tuning.'

Include synonyms and method breadth such as QLoRA, PEFT, and adapters to improve trigger-term coverage and distinctiveness.

Mention concrete outcomes (e.g. train <1% of parameters, swap adapters at inference) to increase specificity beyond a single action.

Dimension	Reasoning	Score
Specificity	Names the domain and one concrete action ("Fine-tune large LLMs with LoRA") but offers no breadth of methods beyond LoRA, matching the anchor for naming the domain plus 1-2 concrete actions without comprehensive coverage.	3 / 5
Completeness	It clearly states what the skill does but provides no "Use when..." clause or equivalent trigger guidance, so per the cap it cannot exceed 3.	3 / 5
Trigger Term Quality	Natural keywords like "fine-tune", "LLMs", "LoRA", and "GPU memory" are present, but common synonyms such as PEFT, QLoRA, and adapters are missing, placing it just below comprehensive coverage.	4 / 5
Distinctiveness Conflict Risk	The description is somewhat specific to LoRA fine-tuning but could still overlap with general fine-tuning skills, with no explicit trigger phrase to disambiguate.	3 / 5
	Total	13 / 20 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: 6680afb

Reviewed: about 3 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.