minicpm5-finetune

Pick the right fine-tuning framework for a MiniCPM5-1B base checkpoint and route to a framework-specific cookbook skill. Use when the user wants to SFT / LoRA / DPO / continue-pretrain MiniCPM5 and has not yet committed to a specific framework, or when they say "fine-tune MiniCPM5", "train MiniCPM5", "MiniCPM5 微调", "LoRA MiniCPM5", "继续训练 MiniCPM5".

Quality

91%

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

Quality

Content

92%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 well-engineered router skill body: concrete, executable, and clearly sequenced with a validation feedback loop. Its only blemish is minor repetition of the 'gotcha' framing between the intro and section 3.

Suggestions

Remove the duplicate 'every framework has at least one MiniCPM5-specific gotcha' sentence from either the intro or section 3 to tighten conciseness.

Since the bundle ships no references/scripts/assets and docs/finetune/ is not present in-skill, consider noting that the sub-skills live in the sibling skill collection so the relative paths are unambiguous to a fresh reader.

Dimension	Reasoning	Score
Conciseness	The body is dense and assumes Claude knows what SFT/LoRA/DPO are, with concrete tables and a copy-paste sanity-check snippet. Minor over-explanation remains: the phrase 'every framework has at least one MiniCPM5-specific gotcha' is restated in both the intro paragraph and section 3, which could be trimmed.	4 / 5
Actionability	Fully executable guidance: a decision matrix mapping user intent to a specific skill to invoke, decision shortcuts, an exact gotcha table with concrete flags ('template: empty', '--model_type llama --template chatml', 'transformers==4.57.3'), and a complete copy-paste bash/python sanity-check block covering the common case.	5 / 5
Workflow Clarity	A clear numbered sequence (gather input → pick framework → read sub-skill → sanity check → deploy adapter) with an explicit validation step in section 4 and a feedback loop ('If it's gibberish or empty, training broke ... check the framework's gotcha row above') for error recovery.	5 / 5
Progressive Disclosure	A clean router overview pointing one level deep to five sub-skills and the docs/finetune/ cookbooks, with each reference clearly signaled via a decision matrix and an explicit 'Read the picked sub-skill in full before running' instruction; content is appropriately split rather than inlined.	5 / 5
	Total	19 / 20 Passed

Description

90%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, well-scoped router description that clearly states what it does and when to use it, with rich natural trigger terms in both English and Chinese. Its only weakness is that the skill's own actions (pick, route) are inherently limited, capping the specificity of capabilities.

Suggestions

To lift specificity, name the concrete routing outputs more explicitly (e.g. 'selects one of five framework skills and invokes it'), so the action set reads as more than just 'pick and route'.

Dimension	Reasoning	Score
Specificity	Names the domain ('fine-tuning framework for a MiniCPM5-1B base checkpoint') and 1-2 concrete actions ('Pick the right fine-tuning framework', 'route to a framework-specific cookbook skill'), but the skill's own actions are limited to picking and routing rather than performing training. The SFT/LoRA/DPO/continue-pretrain enumeration describes the user's goal, not distinct actions the skill performs, so coverage is not comprehensive — anchor 3 fits better than 4.	3 / 5
Completeness	Explicitly answers both 'what' (pick the right framework, route to a framework-specific cookbook skill) and 'when' via an explicit 'Use when...' clause with concrete trigger phrases, matching the anchor 5 example pattern.	5 / 5
Trigger Term Quality	Comprehensive natural triggers including synonyms and both English and Chinese phrasings: 'fine-tune MiniCPM5', 'train MiniCPM5', 'MiniCPM5 微调', 'LoRA MiniCPM5', '继续训练 MiniCPM5', plus the SFT/LoRA/DPO/continue-pretrain vocabulary users would naturally say.	5 / 5
Distinctiveness Conflict Risk	A tightly scoped niche (MiniCPM5-1B fine-tuning framework routing) with distinct, model-specific triggers; minimal overlap risk with other skills since the triggers are pinned to a specific model.	5 / 5
	Total	18 / 20 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
relative_links	Relative link issues: 1 suspicious	Warning

	Total	15 / 16 Passed

Repository: OpenBMB/MiniCPM
Path: skills/minicpm5-finetune/SKILL.md
Commit: 719e4fc

Reviewed: 17 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.