salesforce
About trained model
I have thoroughly reviewed both your paper and the associated code. In the paper, it is mentioned that the model was trained on nine tasks. However, upon examining the code, it appears that the model is capable of being trained on twelve tasks, including a zero-shot task which is currently commented out.
Could you please clarify if the unicontrol.ckpt and unicontrol_v1.1.ckpt models were trained using the nine datasets specified in your paper, or if there were additional datasets involved?
The very initial ckpt (released on May 25, 2023, https://huggingface.co/Salesforce/UniControl/blob/main/unicontrol.ckpt) is trained by the nine tasks. The current opensourced ckpts (including unicontrol.ckpt and unicontrol_v1.1.ckpt) are all trained by 11 tasks to demonstrate better capacity.
