unsloth-fine-tuning

Fast LLM fine-tuning with Unsloth - 2-5x faster training, 50-80% less VRAM. Use for single-GPU LoRA/QLoRA SFT, GRPO/RL reasoning training, vision/TTS fine-tuning, and GGUF export to Ollama/vLLM/llama.cpp. Supports 300+ models including Llama, Qwen, Gemma, DeepSeek, Mistral, Phi, and gpt-oss.

Quality

73%

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 ./backend/cli/skills/ml-training/unsloth/SKILL.md

Quality

Content

53%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.

A comprehensive, highly actionable skill body with strong code examples and tables, weakened by duplicated link tables, missing dataset definitions in two snippets, and — most seriously — an entire bundle of referenced docs/ files that do not exist.

Suggestions

Bundle the ~30 referenced docs/*.md files (or remove the dangling links) so the Quick Reference and References navigation actually resolves.

Deduplicate the Quick Reference link table and the References section into a single organized index to remove ~60 lines of repeated links.

Add an explicit validation/evaluation checkpoint (e.g., eval on held-out split, verify saved adapter loads and generates sane output) to the SFT and GRPO workflow checklists before the save/deploy step.

Complete the GRPO and vision code snippets by loading `dataset`/`raw_dataset` so they are copy-paste executable.

Dimension	Reasoning	Score
Conciseness	Mostly efficient (tables, checklists, no condescending concept explanations), but the 'Quick Reference' link table (lines 47-73) and 'References' section (lines 593-627) list essentially the same ~30 doc links, and 'Best Practices' reiterates settings already covered in the workflows — clear duplication that could be tightened.	3 / 5
Actionability	Largely copy-paste-ready code with real model names and concrete parameters, plus specific commands (pip, docker, ollama create, vllm serve); minor gaps as the GRPO and vision snippets reference a `dataset`/`raw_dataset` variable that is not loaded in those snippets.	4 / 5
Workflow Clarity	Workflows have clear checklists and sequencing, but the SFT/RL training workflows are batch operations with no explicit validation/evaluation checkpoint before save/deploy (eval is mentioned only separately in Best Practices), so per the batch-operation cap workflow clarity cannot exceed 3.	3 / 5
Progressive Disclosure	Section structure and reference signaling are well organized, but every referenced `docs/*.md` file (~30 of them, across the Quick Reference and References tables) is absent from the bundle — `docs/`, `references/`, `scripts/`, and `assets/` are all empty — so the promised one-level-deep navigation is non-functional and misleading.	2 / 5
	Total	12 / 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.

A strong, specific description that answers both what the skill does and when to use it, with concrete actions and natural trigger terms. Minor room to add file-extension synonyms like '.gguf'.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions across modalities — 'single-GPU LoRA/QLoRA SFT, GRPO/RL reasoning training, vision/TTS fine-tuning, and GGUF export to Ollama/vLLM/llama.cpp' — giving comprehensive coverage of the skill's scope.	5 / 5
Completeness	Explicitly answers both 'what' ('Fast LLM fine-tuning with Unsloth - 2-5x faster training, 50-80% less VRAM') and 'when' via an explicit 'Use for...' trigger clause enumerating concrete use cases.	5 / 5
Trigger Term Quality	Good keyword coverage with natural terms and synonyms ('fine-tuning', 'LoRA/QLoRA', 'GRPO/RL', 'Ollama/vLLM/llama.cpp', plus model families), but a few natural variations/file extensions (e.g. '.gguf', 'train a model') are absent.	4 / 5
Distinctiveness Conflict Risk	Names a specific tool (Unsloth) and distinct algorithm/deployment triggers (GRPO, GGUF, Ollama/vLLM), carving a clear niche with minimal conflict risk; only minor adjacency to a general colab-finetuning skill.	5 / 5
	Total	19 / 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
skill_md_line_count	SKILL.md is long (642 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
relative_links	Relative link issues: 69 missing	Warning

	Total	13 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/unsloth/SKILL.md
Commit: 1d182e9

Reviewed: 3 days 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.