I'm particularly looking forward to updates based on Kontext weights, or the direct open-sourcing of the trained model. Thank you very much!