3 ms·Computational Bottlenecks of Training Small-Scale Large Language Models2 points by sandwichsphinx 2y ago