When retraining it is tedious to answer all the prompts, and hurts reproducibility. It would be nice if there were support for some kind of configuration language to define a training run. Worth looking around for best practices. Not sure if people are still using https://hydra.cc/ for ML these days
When retraining it is tedious to answer all the prompts, and hurts reproducibility. It would be nice if there were support for some kind of configuration language to define a training run. Worth looking around for best practices. Not sure if people are still using https://hydra.cc/ for ML these days