Speeding up Model Loading with fastsafetensors

TL;DR

This paper introduces fastsafetensors, a Python library that significantly accelerates loading large pre-trained models by optimizing tensor deserialization and data transfer processes, achieving up to 7.5x speedup.

Contribution

The work presents a novel deserialization approach for safetensors files that improves model loading speed through direct device memory instantiation and low-level I/O optimizations.

Findings

Achieves 4.8x to 7.5x faster model loading times

Effective for models with up to 176 billion parameters

Improves I/O and tensor preprocessing efficiency

Abstract

The rapid increases in model parameter sizes introduces new challenges in pre-trained model loading. Currently, machine learning code often deserializes each parameter as a tensor object in host memory before copying it to device memory. We found that this approach underutilized storage throughput and significantly slowed down loading large models with a widely-used model file formats, safetensors. In this work, we present fastsafetensors, a Python library designed to optimize the deserialization of tensors in safetensors files. Our approach first copies groups of on-disk parameters to device memory, where they are directly instantiated as tensor objects. This design enables further optimization in low-level I/O and high-level tensor preprocessing, including parallelized copying, peer-to-peer DMA, and GPU offloading. Experimental results show performance improvements of 4.8x to 7.5x in loading models such as Llama (7, 13, and 70 billion parameters), Falcon (40 billion parameters), and the Bloom (176 billion parameters).