Tests on EMPIAR-288 were already using more reasonable settings. Nevertheless here are the results from what you suggested and they are very similar. All parameters except those specified in the table below, par_diam, num_particles and seed, which I set the same for all jobs, were kept default.
Looks like 0.3.20 heavily overfits the model already after the first epoch.
| J351 | topaz==0.2.5 | num_distribute=2 | num_workers=4 |
| J352 | topaz==0.2.5 | num_distribute=1 | num_workers=8 |
| J353 | topaz==0.2.5 | num_distribute=1 | num_workers=16 |
| J354 | topaz==0.3.20 | num_distribute=2 | num_workers=4 |
| J355 | topaz==0.3.20 | num_distribute=1 | num_workers=8 |
| J356 | topaz==0.3.20 | num_distribute=1 | num_workers=16 |
