Estimations for memory and compute requirements for training large language models.
No changes between version 2 and version 2
