Training MLM models without softmax distribution

Data: wtorek, 07.03.2023, godz. 11:00-12:00

Prelegent: Karol Kaczmarek (UAM/Applica)

Abstrakt: I would like to present an alternative way to train MLM models without the softmax distribution to predict the masked tokens.

Miejsce: B1-7/8 oraz online