allenai/Olmo-3-7B-Instruct-SFT further finetuned using full SFT on a 10k sample of allenai/WildChat.

We also train four variants, see subfolders: project dataset along "Balanced and measured approaches" and "Individuality and personalization", and train on only top (50-100) or bottom (0-50) half of the dataset.

For the exact 10k dataset used, see data_hf.csv in repo.

Open-instruct (https://github.com/allenai/open-instruct), same training setup as in Olmo-3 (https://arxiv.org/abs/2512.13961).