A compact ShareGPT-style dataset for training models to generate runnable Three.js HTML scenes from short object-building prompts. This revision contains a reasoning-augmented subset of the dataset. For each completed row, the assistant response has a generated reasoning trace prepended to the original gpt message value.

	Contents

Path Split Rows Description

train.parquet train 708 ShareGPT conversations with reasoning traces prepended… See the full description on the dataset page: https://huggingface.co/datasets/Akicou/Threen-PT-ShareGPT.