Historical protein language model experiments

These records use earlier training and evaluation protocols. Their losses are not comparable with the current fixed 15% masking benchmark's bits per masked residue.

See the current research workflow or the historical source data.

Loading historical experiments...