@ElanaPearl I added some fixes on top of your earlier branch - this code worked for pre-training on the 77M set. i think we can close #43
also unpinned the version in the conda file here so that we can get the newest cuda and pytorch builds, though we may want to pin again once we have some stable trained models
(really I should have separated these changes out into different branches)
@ElanaPearl I added some fixes on top of your earlier branch - this code worked for pre-training on the 77M set. i think we can close #43
also unpinned the version in the conda file here so that we can get the newest cuda and pytorch builds, though we may want to pin again once we have some stable trained models
(really I should have separated these changes out into different branches)