colab-finetuning

Fine-tune LLMs on Google Colab GPUs directly from openscience. Connects to Colab runtimes via WebSocket bridge for remote training with Unsloth. Supports SFT, GRPO, DPO, vision, and TTS workflows on free T4 to Pro A100 GPUs.

Quality

67%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/ml-training/colab-finetuning/SKILL.md

Quality

Content

68%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 well-organized, actionable, and token-efficient with appropriate progressive disclosure to real reference files. Its main weakness is the Quick Start workflow lacking explicit validation checkpoints for a costly remote-training operation.

Suggestions

Insert a validation checkpoint in Quick Start (e.g., after connecting, run 'Use colab_status detail=gpu' and only proceed to colab_finetune when the connection is healthy).

Link the existing references/bridge-setup.md from the body so all bundle files are discoverable.

Clarify whether the code blocks are openscience CLI tool invocations or pseudocode, and provide at least one copy-paste-ready executable command form.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient with action-oriented tables and tight workflow blocks; only minor repetition of the description and a few explanatory phrases could be trimmed.	4 / 5
Actionability	Provides concrete tool invocations with specific example models and datasets plus exact Colab UI steps, but the code blocks use tool-call pseudo-syntax rather than fully executable shell/python, leaving minor gaps.	4 / 5
Workflow Clarity	Quick Start gives a clear 4-step sequence, but this expensive remote-training operation has no explicit validation checkpoint (e.g., verify connection via colab_status before launching training), capping workflow clarity at 3 per the destructive/batch-operations rule.	3 / 5
Progressive Disclosure	Good section structure with two clearly signaled, one-level-deep references (gpu-tiers.md, troubleshooting.md) that are real files; minor gap is the existing bridge-setup.md reference not being linked from the body.	4 / 5
	Total	15 / 20 Passed

Description

66%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 specific and distinctive with good natural trigger terms, but it omits an explicit 'Use when...' clause, which caps completeness. Adding concrete trigger guidance would raise the overall quality.

Suggestions

Add an explicit 'Use when...' sentence naming natural user triggers (e.g., 'Use when you need to fine-tune a model on free or paid Colab GPUs without a local GPU').

Include common synonyms users say like 'train models', 'Colab notebook', or 'LoRA/QLoRA' to broaden trigger-term coverage.

Tighten the headline so the primary concrete action (fine-tuning) leads, then list supported workflow types in one compact clause.

Dimension	Reasoning	Score
Specificity	Lists several concrete actions ('Fine-tune LLMs', 'Connects to Colab runtimes via WebSocket bridge', 'remote training with Unsloth') plus enumerated workflows, with only minor coverage gaps relative to the comprehensive 5-anchor.	4 / 5
Completeness	Clearly states the 'what' (Colab GPU fine-tuning via WebSocket bridge, SFT/GRPO/DPO/vision/TTS) but lacks an explicit 'Use when...' trigger clause, capping completeness at 3 per the guideline.	3 / 5
Trigger Term Quality	Includes natural terms a user would say ('Fine-tune LLMs', 'Google Colab', 'GPU', 'remote training') but misses some common variations like 'train models', 'notebook', or 'LoRA/QLoRA'.	4 / 5
Distinctiveness Conflict Risk	The Colab + WebSocket bridge + openscience + Unsloth combination is a distinct niche with minimal overlap risk, though it sits close to general cloud-training skills.	4 / 5
	Total	15 / 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
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/colab-finetuning/SKILL.md
Commit: 3a6c3a9

Reviewed: 1 day 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.