Initial results from Pythia-[160M, 410M, 1B] fine tuned on v1 of the chemrxiv dataset for 1 epoch with no fine tuning or work on cleaning the dataset

— = Random performance

STEM: hendrycks-[subject]

Screenshot 2023-04-20 at 10.46.22.png

Screenshot 2023-04-20 at 10.46.33.png

Screenshot 2023-04-20 at 10.46.43.png

Screenshot 2023-04-20 at 10.47.05.png

Screenshot 2023-04-20 at 10.45.56.png

Screenshot 2023-04-20 at 10.47.13.png

Screenshot 2023-04-20 at 10.46.11.png

Screenshot 2023-04-20 at 10.46.54.png

Screenshot 2023-04-19 at 17.18.34.png

LAMBADA

Screenshot 2023-04-20 at 10.44.53.png

Screenshot 2023-04-19 at 17.37.01.png