This dataset contains 1,339,649 complete conversations from
mlabonne/open-perfectblend
with assistant responses regenerated by Qwen3-4B. It is published as one
default/train split and does not expose internal source subdivisions.
id (string): contiguous dataset row ID from 0.
conversations (list): complete ShareGPT messages with from set to
human or gpt and the original text in value.
source (string): always… See the full description on the dataset page:
https://huggingface.co/datasets/alice1001/open-perfectblend-qwen3-4b-regen.
