A small, internal demonstration dataset used by Vision Ops to showcase how
processed datasets are managed on HuggingFace.
This dataset is in Alpaca format (instruction, input, output) and is
split into three partitions so that the train / validation / test boundary is
explicit and reproducible.