LLM & Text Generation1 min read
I trained GPT-2 with full-batch gradient descent so you don't have to
LessWrong
Every AI story we track on Deep Learning Training Methods: 1 story so far, each summarized in our own words and linked to the publisher that reported it.