minicpm5-finetune-ms-swift

Fine-tune MiniCPM5-1B with ms-swift (ModelScope's SFT / DPO / KTO / ORPO toolkit). Use when the user mentions "ms-swift", "swift sft", "swift rlhf", or wants ModelScope-native training. The two mandatory flags `--model_type llama --template chatml` are baked in.

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

90%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 tight, highly actionable skill body with complete copy-paste commands and a sensible install→train→validate→merge flow. Its only real weaknesses are a missing error-recovery feedback loop in the validation step and a broken outbound reference link.

Suggestions

Turn the validation step into an explicit feedback loop: state what to do if loss does not decrease (e.g., check data format, lower learning rate, re-run), raising workflow_clarity to 5.

Fix or remove the broken Reference link — `docs/finetune/ms_swift.md` does not exist in the bundle; either ship the file or drop the link so navigation is trustworthy.

Consider a one-line note on the Multi-GPU / RLHF examples' "..." indicating the omitted flags are the same LoRA block from step 2, so the variants are unambiguously executable.

Dimension	Reasoning	Score
Conciseness	The body is lean: no explanations of what fine-tuning, LoRA, or ChatML are; every section (input table, install, train, validate, merge, pitfalls) earns its place. Time-sensitive version notes are appropriately confined to a deprecation callout and the pitfalls section rather than the main flow.	5 / 5
Actionability	Provides copy-paste-ready commands for the primary case — a complete `swift sft` invocation with all hyperparameters, plus install, export/merge, and pitfalls with concrete fixes. Secondary variant examples use "..." but that ellipsis is explicitly justified ("Same flag surface, just swap the trainer").	5 / 5
Workflow Clarity	Clear numbered sequence (Install → Train → Validate → Merge) with a validation step showing expected loss decrease and adapter path. Not a 5 because validation is observational rather than an explicit validate→fix→retry feedback loop, leaving a minor checkpoint gap.	4 / 5
Progressive Disclosure	Well-organized with clear section headers and self-contained content (no bundle files present). Not a 5 because the single outbound reference `docs/finetune/ms_swift.md` does not resolve to a real file, undermining easy navigation.	4 / 5
	Total	18 / 20 Passed

Description

87%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-targeted description that explicitly states both capability and activation triggers using the tool's natural CLI vocabulary. It is highly distinctive and unlikely to misfire; the only minor gap is that the listed actions are all variants of one operation.

Dimension	Reasoning	Score
Specificity	Quotes a specific model and lists several concrete training modes ("SFT / DPO / KTO / ORPO toolkit"), matching anchor 4 (several specific actions). Not a 5 because all actions are variants of fine-tuning rather than a comprehensive set of distinct operations.	4 / 5
Completeness	Explicitly answers what ("Fine-tune MiniCPM5-1B with ms-swift ... SFT / DPO / KTO / ORPO toolkit") and when ("Use when the user mentions ...") with concrete trigger phrases, matching anchor 5.	5 / 5
Trigger Term Quality	Quotes natural user phrases "ms-swift", "swift sft", "swift rlhf" plus "ModelScope-native training", giving good keyword coverage. Not a 5 because "ModelScope-native training" is a manufactured phrase rather than a naturally spoken synonym set.	4 / 5
Distinctiveness Conflict Risk	The triggers are specific to a named CLI tool and model (ms-swift, MiniCPM5-1B), creating a clear niche with minimal overlap risk against other training skills like LLaMA-Factory.	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-ms-swift/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.