Fetching the paper…

Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism · Around