umd-psycholing / lm-syntactic-generalization

0 stars 0 forks source link

GRNN surprisal fixes #5

Closed sathvikn closed 9 months ago

sathvikn commented 9 months ago

Add a notebook that works w our eval scripts' outputs to make plots. Fix GRNN surprisal to compute logprobs with base 2, and add start/stop tokens for GPT and GRNN (consistent w/LMZoo and reference implementations)