minicpm5-finetune-gguf-lora

Fine-tune MiniCPM5-1B into a LoRA adapter and convert it to a GGUF adapter that loads directly into llama.cpp / llama-server and the MiniCPM Desk Pet app's custom-LoRA upload. Use when the user wants "GGUF LoRA", "convert LoRA to GGUF", "convert_lora_to_gguf", a custom persona/skin for the desktop pet, "桌宠自定义 LoRA", "上传 LoRA 到桌宠", or asks how to take a trained adapter and run it on a GGUF base.

Quality

91%

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

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 high-quality, actionable conversion workflow with executable commands at every step and explicit gotcha/troubleshooting guidance. Validation checkpoints are present but the retry loop is loose, and the body is somewhat long for a single-file skill with no bundle to offload into.

Suggestions

Tighten the retry path: in step 4, add an explicit 'if unchanged → verify --base matched the trained base in step 2, then re-run step 2 and re-test' loop rather than only pointing back.

Trim prose in 'Common pitfalls' and 'When NOT to use' to bullet-style diagnostics so each pitfall maps to one fix without surrounding narration.

Consider extracting the troubleshooting matrix ('Common pitfalls' + the step-2 gotcha) into a references file to shorten the SKILL.md overview and exercise true progressive disclosure.

Dimension	Reasoning	Score
Conciseness	Mostly lean and purposeful — the ASCII pipeline, the `--base` gotcha, and the pitfalls earn their tokens — but the 'Common pitfalls' and 'When NOT to use' prose could be tightened in places without losing meaning.	4 / 5
Actionability	Copy-paste-ready, executable commands throughout: `git clone`, `pip install -r requirements.txt`, the full `convert_lora_to_gguf.py` invocation with flags, `huggingface-cli download`, and both `llama-cli`/`llama-server` test invocations covering the common cases.	5 / 5
Workflow Clarity	Five clearly sequenced steps with two validation checkpoints (step 3 metadata sanity-check, step 4 'test before shipping' with a 'if output is identical to base → adapter didn't load' diagnostic), but the validate→fix→retry loop points back to step 2 rather than forming a tight local retry.	4 / 5
Progressive Disclosure	Well-organized single-file structure (Pipeline overview, Required input table, Steps, Common pitfalls, When NOT to use, Reference) with clearly signaled one-level cross-skill references; no bundle files exist so all content is appropriately inline, though it runs longer than the simple-skill threshold.	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.

A strong, third-person description that states concrete capabilities and pairs them with an explicit, multi-trigger 'Use when' clause spanning English, Chinese, and function-name phrasings. It is both comprehensive and concise with no fluff.

Dimension	Reasoning	Score
Specificity	Names the specific model (MiniCPM5-1B) and multiple concrete actions — 'Fine-tune...into a LoRA adapter', 'convert it to a GGUF adapter that loads directly into llama.cpp / llama-server and the MiniCPM Desk Pet app's custom-LoRA upload' — covering the full train→convert→load path.	5 / 5
Completeness	Explicitly answers both what (fine-tune a LoRA adapter and convert it to GGUF) and when via a clearly marked 'Use when the user wants...' clause with concrete trigger phrases.	5 / 5
Trigger Term Quality	Comprehensive natural-term coverage including synonyms, the literal function name, and Chinese variants: 'GGUF LoRA', 'convert LoRA to GGUF', 'convert_lora_to_gguf', 'custom persona/skin for the desktop pet', '桌宠自定义 LoRA', '上传 LoRA 到桌宠'.	5 / 5
Distinctiveness Conflict Risk	Occupies a clear niche — PEFT→GGUF LoRA conversion for llama.cpp and the Desk Pet app — with distinct, specific triggers that are unlikely to fire for sibling training or deploy skills.	5 / 5
	Total	20 / 20 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: OpenBMB/MiniCPM
Path: skills/minicpm5-finetune-gguf-lora/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.