Stella Biderman: How EleutherAI Trains and Releases LLMs

This article from Weights & Biases highlights an interview with Stella Biderman, a lead scientist at Booz Allen Hamilton and Executive Director at EleutherAI. The discussion covers EleutherAI's approach to training and releasing large language models (LLMs). The interview touches upon various aspects of LLM development, including model selection, reinforcement learning, pre-training and fine-tuning strategies, GPU selection, and the importance of public access. The conversation also explores the differences between EleutherAI and other LLM companies, as well as the critical topics of interpretability and memorization.