minicpm5-finetune-unsloth

Fine-tune MiniCPM5-1B or MiniCPM5-2B with unsloth for tight-VRAM single-GPU LoRA / QLoRA. Use when the user wants "unsloth", "FastLanguageModel", QLoRA on a 24 GB consumer GPU, or asks for the smallest VRAM footprint.

Quality

87%

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

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.

A highly actionable, well-structured recipe with copy-paste code and a clear step sequence. Its main weaknesses are a duplicated transformers rationale, an implicit rather than gated validation step, and a dangling reference link.

Suggestions

Fix or remove the broken Reference link to ../../docs/finetune/unsloth.md — the target file does not exist and no bundle directory ships it.

De-duplicate the transformers==4.57.3 / vLLM coexistence rationale, which is explained verbatim in both the top callout and step 1.

Make the Validate step an explicit pass/fail checkpoint (e.g., 'If loss is not decreasing after epoch 1, stop and check LOAD_IN_4BIT / data format') instead of only 'You should see: ...'.

Dimension	Reasoning	Score
Conciseness	Mostly lean with executable code and a compact input table, but the transformers==4.57.3 rationale is restated in both the callout and step 1, and the non-obvious pin context is borderline over-explained.	4 / 5
Actionability	Fully copy-paste ready: complete install commands, a full train_unsloth.py, a parameterized run command, plus inference and merge snippets that cover the common cases.	5 / 5
Workflow Clarity	Steps are clearly sequenced (Install, Train, Run, Validate) with an expected-output checkpoint and a Common pitfalls recovery section, but validation is 'you should see' rather than an explicit pass/fail feedback loop.	4 / 5
Progressive Disclosure	Well-organized single-file overview with clear sections and a one-level reference link, but the sole reference target (../../docs/finetune/unsloth.md) does not exist and no bundle files accompany the skill.	4 / 5
	Total	17 / 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 tight, well-targeted description that clearly states the capability and gives explicit, natural trigger phrases with low conflict risk. Minor room to broaden trigger synonyms and acknowledge the single-task scope.

Dimension	Reasoning	Score
Specificity	Names the concrete capability — 'Fine-tune MiniCPM5-1B or MiniCPM5-2B with unsloth for tight-VRAM single-GPU LoRA / QLoRA' — covering specific models, library, and techniques, though it is a single task rather than a list of several distinct actions.	4 / 5
Completeness	Explicitly answers both 'what' (fine-tune MiniCPM5 variants with unsloth for LoRA/QLoRA) and 'when' ('Use when the user wants...') with concrete trigger phrases, written in third person.	5 / 5
Trigger Term Quality	Strong natural triggers ('unsloth', 'FastLanguageModel', 'QLoRA on a 24 GB consumer GPU', 'smallest VRAM footprint') that users would actually say, but model-name synonyms and a few variations are absent from the trigger clause.	4 / 5
Distinctiveness Conflict Risk	A narrow niche (specific models + unsloth + tight-VRAM single-GPU) with distinct triggers makes conflict with other skills minimal.	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-unsloth/SKILL.md
Commit: 316cfb1

Reviewed: 15 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.