CtrlK
BlogDocsLog inGet started
Tessl Logo

minicpm5-finetune-trl

Fine-tune MiniCPM5-1B with bare-metal TRL + PEFT, including assistant-only loss via a chat-template patch. Use when the user wants minimal Python, no YAML, full control, or asks for "TRL", "SFTTrainer", "PEFT", "LoraConfig", "assistant_only_loss".

68

Quality

83%

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.

A highly actionable, well-structured recipe with copy-paste code and a clear sequenced workflow including a validation step. Minor conciseness trims and a more explicit validation feedback loop would push it to the top anchor.

DimensionReasoningScore

Conciseness

Lean recipe-style body dominated by executable code with tight, justified comments (the 🔑 markers and chat-template rationale earn their place); a few explanatory prose lines and inline comments could still be trimmed, so it is efficient but not maximally lean.

4 / 5

Actionability

Fully copy-paste-ready: complete install commands, a full train_lora.py script, the train invocation, inference code, and merge code — specific examples cover the common train/validate/infer/merge cases.

5 / 5

Workflow Clarity

Clearly sequenced numbered steps (Install → Patch → Train → Validate) with an explicit validation section showing expected output, but the validation is an expected-output confirmation rather than an explicit validate→fix→retry feedback loop, leaving a minor checkpoint gap.

4 / 5

Progressive Disclosure

Well-organized into labeled sections (Required input, Steps, Inference, Merge, Pitfalls) with a single clearly-signaled one-level reference (docs/finetune/trl.md); no bundle files exist so content is appropriately self-contained, though there is no split beyond the one link.

4 / 5

Total

17

/

20

Passed

Description

83%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, specific description that clearly states both capability and trigger conditions with concrete library/class names. Minor overlap risk with sibling fine-tuning skills and slight synonym gaps keep trigger and distinctiveness scores just below max.

DimensionReasoningScore

Specificity

Names the exact domain (fine-tuning MiniCPM5-1B) and several concrete techniques ('bare-metal TRL + PEFT', 'assistant-only loss via a chat-template patch'), but it is fundamentally one task with implementation detail rather than multiple distinct actions, so it sits below the comprehensive 5 anchor.

4 / 5

Completeness

Explicitly answers both what ('Fine-tune MiniCPM5-1B with bare-metal TRL + PEFT, including assistant-only loss via a chat-template patch') and when ('Use when the user wants minimal Python, no YAML, full control, or asks for...') with concrete trigger phrases.

5 / 5

Trigger Term Quality

Includes natural request phrases ('minimal Python', 'no YAML', 'full control') plus the library/class names a user would actually say ('TRL', 'SFTTrainer', 'PEFT', 'LoraConfig', 'assistant_only_loss'); good coverage but missing common synonyms like 'LoRA fine-tuning' or 'SFT'.

4 / 5

Distinctiveness Conflict Risk

The 'bare-metal', 'assistant_only_loss', and 'chat-template patch' framing carves a clear niche, but trigger terms like 'PEFT', 'LoraConfig', and 'SFTTrainer' overlap with generic TRL/LoRA fine-tuning skills, leaving minor conflict risk.

4 / 5

Total

17

/

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.

Validation15 / 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.