separate loading of train and eval data using param:eval_mode.
combining args.dataset and args.custom_data_path into one option
loading pajama and refinedweb using dataset option (since they both are included into this repo in a fixed location)
tests done in notebook
next step - separate train and eval loading to save time by avoiding lengthy train loading in eval stage.
updated, cleaner version of https://github.com/Vahe1994/SpQR/pull/20
separate loading of train and eval data using param:eval_mode. combining args.dataset and args.custom_data_path into one option loading
pajama
andrefinedweb
using dataset option (since they both are included into this repo in a fixed location) tests done in notebooknext step - separate train and eval loading to save time by avoiding lengthy train loading in eval stage.