The as-trained Stage-2 corpus for Artemis1-13B (Schneewolf Labs'
Qwen3-VL-ViT-grafted-onto-A2/Mistral vision-language flagship).
Extends schneewolflabs/ArtemisMix-v1
with 14,816 non-reasoning creative-writing rows derived from
schneewolflabs/Athanorlite-DPO
(DPO collapsed to SFT, chosen only, bare assistant content so the A2
chat template renders an empty
block — i.e. thinking-off
mode). The bucket balances v1's reasoning-heavy L4 with a… See the full description on the dataset page:
https://huggingface.co/datasets/schneewolflabs/ArtemisMix-v1.1.
