CtrlK
BlogDocsLog inGet started
Tessl Logo

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".

73

Quality

92%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

82%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 body is a well-structured, actionable router that efficiently maps user intent to a specific framework sub-skill and provides a solid post-training sanity check. It is concise and sequenced, with only minor tightening possible on repeated phrasing and on making the cookbook/sub-skill references resolvable.

Suggestions

De-duplicate the 'every framework has at least one gotcha' phrasing — state it once (e.g. in section 3) and reference it rather than repeating across sections 1, 3, and 4.

Make the external references resolvable: either link the sub-skills by a concrete path/slug (e.g. `./minicpm5-finetune-llamafactory/SKILL.md`) or confirm `docs/finetune/` exists and link a specific cookbook file rather than a bare directory.

Tighten section 4's inference script into a callable snippet (parameterize $BASE_MODEL/$OUTPUT_DIR consistently with the rest of the skill) so the sanity check reads as a single step rather than an inline block.

DimensionReasoningScore

Conciseness

Mostly lean — tables, a decision matrix, and a single executable sanity-check script with no concept padding — but the 'every framework has at least one MiniCPM5-specific gotcha' phrasing repeats across sections 1, 3 and the inference-check note, which could be tightened.

4 / 5

Actionability

Provides a concrete decision matrix mapping user intent to a named skill to invoke, decision shortcuts, a per-framework gotcha table, and a copy-paste-ready bash/python sanity-check script — fully executable guidance covering the common routing cases.

5 / 5

Workflow Clarity

Clear five-step sequence (gather inputs → pick via matrix → invoke sub-skill → run sanity check → deploy) with a concrete validation checkpoint in section 4 and an error-recovery pointer ('If it's gibberish... check the framework's gotcha row'), but most per-framework validation is deferred to the sub-skills rather than explicit here.

4 / 5

Progressive Disclosure

Well-structured as a router that points to five sub-skills by name and to a cookbook directory, keeping the body short; however the only external link ([docs/finetune/]) does not resolve within the bundle and sub-skills are referenced by name rather than as resolvable paths, leaving minor navigation gaps.

4 / 5

Total

17

/

20

Passed

Description

100%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 precise, action-oriented, and tightly scoped to MiniCPM5 fine-tuning framework selection, with rich bilingual trigger phrases. It cleanly answers both what the skill does and when to invoke it, with low conflict risk against the sibling framework skills.

DimensionReasoningScore

Specificity

Names the concrete actions — 'Pick the right fine-tuning framework' and 'route to a framework-specific cookbook skill' — and enumerates the supported regimes (SFT / LoRA / DPO / continue-pretrain), giving comprehensive coverage of what the skill does.

5 / 5

Completeness

Explicitly answers both 'what' (pick framework, route to a cookbook skill) and 'when' via a clear 'Use when the user wants to...' clause with concrete trigger phrases, matching the score-5 anchor.

5 / 5

Trigger Term Quality

Comprehensive natural triggers including English and Chinese variants ('fine-tune MiniCPM5', 'train MiniCPM5', 'MiniCPM5 微调', 'LoRA MiniCPM5', '继续训练 MiniCPM5') plus the conditional trigger 'has not yet committed to a specific framework'.

5 / 5

Distinctiveness Conflict Risk

Scoped to MiniCPM5-1B and explicitly limited to users who 'have not yet committed to a specific framework', carving a distinct router niche with minimal overlap against the framework-specific sibling skills.

5 / 5

Total

20

/

20

Passed

Validation

93%

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

CriteriaDescriptionResult

relative_links

Relative link issues: 1 suspicious

Warning

Total

15

/

16

Passed

Repository
OpenBMB/MiniCPM
Reviewed

Table of Contents

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.