MegaScale: Scaling Large Language Model Training to More Than 10,000 GPUs

TL;DR

MegaScale demonstrates a full-stack system design for training large language models on over 10,000 GPUs, achieving high efficiency and stability through innovative system and algorithm co-design, diagnostics, and fault tolerance.

Contribution

This work introduces MegaScale, a scalable, efficient, and stable system for training LLMs at unprecedented GPU scale, with novel diagnostics and optimization techniques.

Findings

Achieved 55.2% Model FLOPs Utilization on 12,288 GPUs.

Improved MFU by 1.34x over Megatron-LM.

Developed diagnostic tools for stability and fault tolerance.

Abstract

We present the design, implementation and engineering experience in building and deploying MegaScale, a production system for training large language models (LLMs) at the scale of more than 10,000 GPUs. Training LLMs at this scale brings unprecedented challenges to training efficiency and stability. We take a full-stack approach that co-designs the algorithmic and system components across model block and optimizer design, computation and communication overlapping, operator optimization, data pipeline, and network performance tuning. Maintaining high efficiency throughout the training process (i.e., stability) is an important consideration in production given the long extent of LLM training jobs. Many hard stability issues only emerge at large scale, and in-depth observability is the key to address them. We develop a set of diagnosis tools to monitor system components and events deep in the stack, identify root causes, and derive effective techniques to achieve fault tolerance and mitigate stragglers. MegaScale achieves 55.2% Model FLOPs Utilization (MFU) when training a 175B LLM model on 12,288 GPUs, improving the MFU by 1.34x compared to Megatron-LM. We share our operational experience in identifying and fixing failures and stragglers. We hope by articulating the problems and sharing our experience from a systems perspective, this work can inspire future LLM systems research.