4 ms·
It's not an either or. We're going to leverage the web trained LLMs to bootstrap the specialist models via combination of training token quality classifiers and
by causalmodels 3y ago
It's not an either or. We're going to leverage the web trained LLMs to bootstrap the specialist models via combination of training token quality classifiers and synthetic data generation. Phi-1 is a pretty good example of this.