llm-finetuning

LLM fine-tuning expert for LoRA, QLoRA, dataset preparation, and training optimization

Quality

72%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./crates/openfang-skills/bundled/llm-finetuning/SKILL.md

Quality

Content

72%

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

The body is well-structured and highly actionable with concrete hyperparameters and library calls, but it carries some redundant conceptual padding and lacks an explicit validation feedback loop for a long-running training operation.

Suggestions

Tighten or remove the opening paragraph and the more generic Key Principles (e.g. dataset-quality and base-model axioms Claude already knows) to improve token efficiency.

Add an explicit validation feedback loop in the workflow, e.g. after training: evaluate on a held-out set, and if validation loss rises or metrics drop, reduce epochs/lr and retrain before merging adapters.

Dimension	Reasoning	Score
Conciseness	The Techniques and Common Patterns sections are lean and earn their place with specific values and library calls, but the opening paragraph restates the description and several "Key Principles" (dataset quality over quantity, start with a strong base model) are general ML wisdom Claude already knows, so it is mostly efficient but could be tightened.	2 / 3
Actionability	It gives concrete, specific guidance throughout: LoRA rank r=8 to 64 with alpha 2x rank on q_proj/v_proj, the PEFT workflow (get_peft_model(), merge_and_unload()), learning rate 1e-5 to 2e-4 with cosine schedule, warmup 0.03-0.1, and task metrics (ROUGE, exact match) — actionable even without code.	3 / 3
Workflow Clarity	The PEFT workflow (load base, create config, get_peft_model(), train, save/load adapters) is sequenced, but fine-tuning is a long-running batch operation with no explicit validate→fix→retry checkpoint, so per the feedback-loops note workflow clarity is capped at 2.	2 / 3
Progressive Disclosure	The body is under 50 lines with no need for external references and is organized into well-labeled sections (Key Principles, Techniques, Common Patterns, Pitfalls), so per the simple-skills note it scores 3 on organization alone.	3 / 3
	Total	10 / 12 Passed

Description

72%

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 on-niche and uses strong natural trigger terms, but it reads as an area list rather than concrete actions and lacks an explicit "Use when..." clause, which caps completeness.

Suggestions

Add an explicit trigger clause, e.g. "Use when the user wants to fine-tune or adapt an LLM with LoRA/QLoRA, prepare an instruction-tuning dataset, or optimize training hyperparameters."

Reframe abstract areas as concrete verbs, e.g. "Configure LoRA/QLoRA adapters, prepare instruction-response datasets, and optimize training hyperparameters for large language models."

Dimension	Reasoning	Score
Specificity	The description names the domain and several areas ("LoRA, QLoRA, dataset preparation, and training optimization") but frames them as abstract areas under "expert for" rather than listing multiple concrete actions like extract/fill/merge, matching the score-2 anchor of naming domain and some actions but not comprehensive.	2 / 3
Completeness	It clearly states what the skill does but contains no "Use when..." clause or equivalent explicit trigger guidance, so per the judging guidelines completeness is capped at 2 (has what, when missing/implied).	2 / 3
Trigger Term Quality	"LLM fine-tuning", "LoRA", and "QLoRA" are exactly the natural terms a user would say when they need this skill, giving good coverage of common phrasings; it is not merely generic jargon nor missing common variations.	3 / 3
Distinctiveness Conflict Risk	Parameter-efficient LLM fine-tuning via LoRA/QLoRA is a clear niche with distinct triggers unlikely to conflict with unrelated skills, matching the score-3 anchor.	3 / 3
	Total	10 / 12 Passed

Validation

100%

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 — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: RightNow-AI/openfang
Path: crates/openfang-skills/bundled/llm-finetuning/SKILL.md
Commit: acf2587

Reviewed: about 7 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.