Description
dataset-curation is an agent skill from the wshobson/agents collection for preparing, formatting and validating datasets used to fine-tune language models, with supervised fine-tuning (SFT) and preference training as the main cases. It takes raw examples, such as demonstrations, preference judgments or task prompts, and turns them into a formatted, packed and validated JSONL dataset. The skill explains how the required format depends on the training method that was chosen, covers the template and packing mechanics behind the most common silent training failures, and sets rules for mixing synthetic data into a dataset without causing the model to collapse. It ends with a dataset card, a record of what the data is and how it was built, which the wider fine-tuning workflow uses as a gate before a training run starts. The skill assumes the training method has already been chosen, so it sits after the routing step in the wshobson fine-tuning sequence and is not a method selector. It is installed with the skills CLI from the wshobson/agents repository. Its skills.sh page showed about 2.4K installs and passing results from three security audits at the time of the scan.…
Highlights
- Turns raw examples into training-ready JSONL
- Formats for SFT and preference training
- Template and packing mechanics
- Rules for mixing in synthetic data
- Dataset card as a pre-training gate
- Installs with the skills CLI
Ratings & Reviews
No ratings yet
Be the first to rate dataset-curation.
People also visit
dataset-curation FAQ
What does this skill produce?
A formatted, packed and validated JSONL dataset, plus a completed dataset card.
What inputs does it need?
Raw examples such as demonstrations, preference judgments or task prompts, plus the training method chosen beforehand.
Does it choose the training method?
No. The skill assumes the method is already decided. Method selection is covered by the separate finetuning-method-selection skill.
What is a dataset card?
A written record of the dataset and how it was built, which the fine-tuning workflow checks before a run starts.
How do I install it?
Run npx skills add https://github.com/wshobson/agents --skill dataset-curation.
What security audits does it have?
skills.sh lists three audits: Gen Agent Trust Hub Pass, Socket Pass and Snyk Pass.
What are the best dataset-curation alternatives?
Popular alternatives to dataset-curation include finetuning-method-selection, trace-to-training-data and fine-tuning-expert. See all dataset-curation alternatives