We covered this in some of our earlier courses – lower batch sizes provide more updates, which should give better results for a fixed # epochs.

→ View original post on X — @jeremyphoward