DataStates-LLM: Lazy Asynchronous Checkpointing for Large Language Models

TL;DR

This paper introduces DataStates-LLM, a lazy asynchronous checkpointing method that significantly reduces I/O overheads during large language model training, enabling faster and more scalable checkpointing at high frequencies.

Contribution

It proposes a novel multi-level lazy asynchronous checkpointing approach that leverages tensor immutability to minimize I/O interference during LLM training.

Findings

Up to 48× faster checkpointing performance.

Achieved 2.2× reduction in total training time.

Effective at scales up to 180 GPUs.

Abstract

LLMs have seen rapid adoption in all domains. They need to be trained on high-end high-performance computing (HPC) infrastructures and ingest massive amounts of input data. Unsurprisingly, at such a large scale, unexpected events (e.g., failures of components, instability of the software, undesirable learning patterns, etc.), are frequent and typically impact the training in a negative fashion. Thus, LLMs need to be checkpointed frequently so that they can be rolled back to a stable state and subsequently fine-tuned. However, given the large sizes of LLMs, a straightforward checkpointing solution that directly writes the model parameters and optimizer state to persistent storage (e.g., a parallel file system), incurs significant I/O overheads. To address this challenge, in this paper we study how to reduce the I/O overheads for enabling fast and scalable checkpointing for LLMs that can be applied at high frequency (up to the granularity of individual iterations) without significant impact on the training process. Specifically, we introduce a lazy asynchronous multi-level approach that takes advantage of the fact that the tensors making up the model and optimizer state shards remain immutable for extended periods of time, which makes it possible to copy their content in the background with minimal interference during the training process. We evaluate our approach at scales of up to 180 GPUs using different model sizes, parallelism settings, and checkpointing frequencies. The results show up to 48$\times$ faster checkpointing and 2.2$\times$ faster end-to-end training runtime compared with the state-of-art checkpointing approaches.