personads/earlyberts-seed0
Updated
🐤 BERT pre-training checkpoints used for analyzing early learning dynamics in "The Subspace Chronicles" (Müller-Eberstein et al., 2023).
Note Seed 0 | Steps 10–40,000
Note Seed 1 | Steps 10–40,000
Note Seed 2 | Steps 10–40,000
Note Seed 3 | Steps 10–40,000
Note Seed 4 | Steps 10–40,000