Please note that both Megatron-LM and DeepSpeed have Pipeline Parallelism and BF16 Optimizer implementations, but we used the ones from DeepSpeed as they are integrated with ZeRO. Megatron-DeepSpeed implements 3D Parallelism to allow huge models to train in a very efficient way. Let’s briefly discuss the 3D components. DataParallel (DP) - the same setup is replicated multiple times, and each being
