Trained on MS MARCO using the a similar strategy as https://huggingface.co/cross-encoder/ms-marco-MiniLM-L12-v2, except with the Ettin base models