unsloth

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Quality

36%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Fix and improve this skill with Tessl

tessl review fix ./03-fine-tuning/unsloth/SKILL.md

Quality

Content

22%

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

The body is generic auto-generated boilerplate with placeholder sections and no executable guidance, code, or real workflow for fine-tuning with Unsloth. While it does point to a real references directory, the navigation is poorly signaled and directs users to files that do not exist.

Suggestions

Replace the empty Quick Reference placeholder with a concrete, copy-paste Unsloth fine-tuning snippet (e.g. FastModel load + SFTTrainer setup) so the skill is immediately actionable.

Add a numbered fine-tuning workflow with a validation checkpoint (e.g. train -> check loss/eval -> save adapters), since training is a fragile multi-step process that needs feedback loops.

Fix navigation to list the actual reference files present (index.md, llms.md, llms-txt.md, llms-full.md) with one-line descriptions, and remove references to non-existent files like getting_started, tutorials, api, and guides.

Dimension	Reasoning	Score
Conciseness	The body is short but padded with empty scaffolding ('Quick reference patterns will be added as you use the skill', 'Add helper scripts here', 'Add templates, boilerplate, or example projects here') that adds tokens with no skill-specific value, fitting the level 2 'mostly efficient but includes some unnecessary' anchor rather than the lean level 3.	2 / 3
Actionability	There is no executable code, commands, or concrete examples; the Quick Reference is literally a placeholder ('Quick reference patterns will be added'), and it directs users to non-existent reference files ('getting_started or tutorials', 'api, guides, etc.'), matching the level 1 'vague or abstract; no concrete code/commands' anchor.	1 / 3
Workflow Clarity	No fine-tuning workflow is sequenced and there are no validation checkpoints; the only 'process' is a generic scraper-refresh note under 'Updating', matching the level 1 'steps unclear or missing; no sequence for multi-step tasks' anchor.	1 / 3
Progressive Disclosure	Content is split across a references directory and the body points to llms-txt.md (which exists), but navigation is weak and references non-existent files (getting_started, tutorials, api, guides) while ignoring the other real reference files (index.md, llms.md, llms-full.md), fitting the level 2 'some structure but not clearly signaled' anchor rather than well-organized level 3.	2 / 3
	Total	6 / 12 Passed

Description

50%

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 communicates the Unsloth fine-tuning niche and notable performance benefits, but it reads as a marketing claim rather than a list of concrete actions and omits any explicit trigger ('Use when...') guidance. It is competent but generic, capping completeness and trigger quality at the middle level.

Suggestions

Replace 'Expert guidance' with concrete skill actions, e.g. 'Fine-tune Llama, Mistral, Gemma, and Qwen models with LoRA/QLoRA, optimize memory usage, and run training 2-5x faster.'

Add an explicit trigger clause: 'Use when the user wants to fine-tune, LoRA/QLoRA tune, or reduce memory for training LLM models, or mentions Unsloth.'

Drop benefit metrics unless tied to actions, and lead with what the skill does rather than how fast it is.

Dimension	Reasoning	Score
Specificity	The description names the domain and concrete benefits ('2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization') but 'Expert guidance' is vague and lists benefits rather than concrete skill actions like 'fine-tune', 'train', or 'optimize models'. It is not the level 3 anchor (multiple specific concrete actions) nor the level 1 vague 'Helps with' style.	2 / 3
Completeness	It states what the skill does (fast fine-tuning with Unsloth) but provides no explicit 'when should Claude use it' trigger; per the guidelines a missing 'Use when...' clause caps completeness at 2, which fits the 'Has what, but when is missing or only implied' anchor.	2 / 3
Trigger Term Quality	It includes relevant keywords a user would say ('fine-tuning', 'Unsloth', 'training', 'LoRA', 'QLoRA', 'memory') but lacks common natural phrasings and has no 'Use when...' trigger guidance, matching the level 2 'some relevant keywords but missing common variations' anchor.	2 / 3
Distinctiveness Conflict Risk	Unsloth is a specific niche so it is somewhat distinguishable, but the broad 'fine-tuning' trigger could overlap with general PEFT/fine-tuning skills, matching the level 2 'somewhat specific but could still overlap' anchor rather than a clear level 3 niche.	2 / 3
	Total	8 / 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: Orchestra-Research/AI-Research-SKILLs
Path: 03-fine-tuning/unsloth/SKILL.md
Commit: 773a529

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.