NVIDIA integrates Muon and advanced optimizers into Megatron to enhance large-scale LLM training with near-parity throughput to AdamW. (Read More)