Is it possible to run an experiment starting from the end of precomputed stages, like resume? For example I already ran some preprocessing stages, and some models are trained already, when I am appending new architecture, it reruns everything, but I want to run only the part that is changed or new.

you can use

resume

to resume from any non-failed step of any past run. Hence you could use resume to skip the preprocessing steps and just iterate the training part

Great! Thank you.