Batch size variation

[odunkel]

Hey,

thanks for your nice work!
Did you explore various batch sizes for training the sliders? Did you see improvements for higher batch sizes? Do you have an intuition to what extent this, e.g., reduces the required number of training steps? Otherwise, I will have to perform this analysis myself... ;)

Thanks

[rohitgandikota]

Hi, we haven't done a principled analysis to answer this. If you find anything interesting, please share it here. Thanks!