SupervisedDatasetBuilder - Tinker Documentation
tinker_cookbook.supervised.SupervisedDatasetBuilder
class tinker_cookbook.supervised.SupervisedDatasetBuilder()
A chz config class that knows how to construct a supervised dataset.
The dataset is usually a chat dataset but does not need to be; it could be raw tokens. Subclasses must implement __call__.
call()
Build and return (train_dataset, eval_dataset).
Returns: tuple[ SupervisedDataset, SupervisedDataset | None] – The training dataset and an optional evaluation dataset.