model with 10k epochs and 0.001 learning rate works much better than

d0ddef9

Jensen-holm commited on Mar 15