Closed lvwerra closed 1 year ago
The csv file is missing?
Indeed, added it here: https://huggingface.co/datasets/bigcode/scaling-laws-exp/tree/main
Just pushed a fix for the optimization issues in the scaling laws. Fitted function now looks as follows:
Probably better make x-axis logscale
Notebook to replicate Chinchilla scaling laws. Follows steps in D.2 (however for pass@k and its inverse rather than loss).