Turn raw examples into training-ready datasets.
What it does
Prepare raw training examples into a formatted, packed, validated JSONL dataset and a completed dataset card.
When to use it
Use when converting raw data into training formats, applying chat templates, configuring packing, generating synthetic data, or writing a dataset card.
How to use it
Give it raw examples and a training-method routing decision; it returns a formatted, packed, validated JSONL dataset and a completed dataset card.
What you provide
- the selected training method
- Your files
What you need · 3
The documented SFT workflow uses current TRL to apply chat templates, pack examples, and mask loss to assistant spans.
The documented template-application code uses AutoTokenizer from transformers to load the selected model's tokenizer.
The workflow assumes this skill has already selected the training method and supplied a routing decision.
Access · 0
No additional actions listed.