Unicron: Economizing Self-Healing LLM Training at Scale

TL;DR

Unicron is a workload manager that enhances large-scale language model training by reducing failure recovery costs and improving efficiency through real-time error detection and dynamic reconfiguration.

Contribution

It introduces a novel self-healing system that optimizes failure recovery in large-scale LLM training, addressing diverse failure scenarios with cost-aware strategies.

Findings

Up to 1.9x training efficiency improvement

Significant reduction in failure recovery costs

Effective real-time error detection without extra overhead

Abstract

Training large-scale language models is increasingly critical in various domains, but it is hindered by frequent failures, leading to significant time and economic costs. Current failure recovery methods in cloud-based settings inadequately address the diverse and complex scenarios that arise, focusing narrowly on erasing downtime for individual tasks without considering the overall cost impact on a cluster. We introduce Unicron, a workload manager designed for efficient self-healing in large-scale language model training. Unicron optimizes the training process by minimizing failure-related costs across multiple concurrent tasks within a cluster. Its key features include in-band error detection for real-time error identification without extra overhead, a dynamic cost-aware plan generation mechanism for optimal reconfiguration, and an efficient transition strategy to reduce downtime during state changes. Deployed on a 128-GPU distributed cluster, Unicron demonstrates up to a 1.9x improvement in training efficiency over state-of-the-art methods, significantly reducing failure recovery costs and enhancing the reliability of large-scale language model training.