This is the second half of the Astral 1.5 Post-Training dataset collection designed to be used with the C-RLFT training method detailed in the "OPENCHAT: ADVANCING OPEN-SOURCE LANGUAGE MODELS WITH MIXED-QUALITY DATA" paper. The SFT portion can be found here
Dataset Description
This dataset merges two datasets sourced from the the original Astral 1 post-training SFT dataset and 1000 Kimi K2 Thinking examples.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/LucidityAI/Astral-1.5-Post-Training-Dataset-CRLFT.
