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

66%

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

65%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A concise, well-organized skill body with strong token efficiency and clear sequencing, held back by pseudo-prose command examples and missing validation checkpoints for batch training. Wiring up the orphaned bridge-setup.md reference would tighten progressive disclosure.

Suggestions

Convert tool invocations to copy-paste ready commands (e.g. a fenced block with the literal CLI call) instead of 'Use <tool> with: ...' prose.

Add an explicit validation step before training — e.g. run `colab_status detail=gpu` and confirm a GPU is connected before calling colab_finetune.

Reference references/bridge-setup.md from the Quick Start (or remove the unused file) so the bridge setup detail is one level deep instead of inlined.

Dimension	Reasoning	Score
Conciseness	The body is lean and dense — tables for GPU tiers and tools, terse workflow blocks — and assumes Claude's competence without explaining what LLMs, fine-tuning, or QLoRA are.	3 / 3
Actionability	Tool invocations are named and parameterized ('Use colab_finetune tool with: workflow: "sft"...') but rendered as prose pseudocode rather than copy-paste executable commands, leaving them concrete but not fully executable.	2 / 3
Workflow Clarity	The 4-step Quick Start is clearly sequenced, but launching a long/batch training job has no validation checkpoint (e.g. verifying GPU/connection via colab_status before colab_finetune), capping this at 2.	2 / 3
Progressive Disclosure	Two references are clearly linked one level deep (gpu-tiers.md, troubleshooting.md), but bridge-setup.md exists in the bundle and is never referenced, and core setup steps are inlined in SKILL.md rather than split out.	2 / 3
	Total	9 / 12 Passed

Description

67%

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 specific, well-differentiated description with concrete actions and a clear niche, weakened only by the absence of an explicit 'Use when' trigger and slightly jargon-heavy phrasing. Adding natural trigger terms would lift completeness and trigger quality.

Suggestions

Add an explicit trigger clause, e.g. 'Use when you need to fine-tune a model but have no local GPU, or the user asks to train on Colab/free GPUs.'

Soften product jargon ('openscience', 'WebSocket bridge') with natural phrasings a user would actually say, like 'train a model on Colab'.

Mirror common variations ('Colab', 'free GPU', 'LoRA/QLoRA fine-tuning') to broaden trigger coverage.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'Fine-tune LLMs', 'Connects to Colab runtimes via WebSocket bridge', 'Supports SFT, GRPO, DPO, vision, and TTS workflows' — matching the top anchor for naming specific actions.	3 / 3
Completeness	The 'what' is thorough, but there is no explicit 'Use when...' trigger clause, so completeness is capped at 2 per the missing-trigger guideline.	2 / 3
Trigger Term Quality	Relevant keywords exist ('Google Colab', 'GPU', 'fine-tune', 'Unsloth') but the phrasing leans on product/jargon ('openscience', 'WebSocket bridge') and omits common natural variations like 'train a model on Colab'.	2 / 3
Distinctiveness Conflict Risk	The niche is distinct — remote Colab GPU training via a WebSocket bridge from the openscience CLI — and unlikely to trigger for unrelated skills.	3 / 3
	Total	10 / 12 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: 52845c3

Reviewed: about 11 hours 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.