Training of modern large neural networks (NN) requires a combination of parallelization strategies encompassing data, model, or optimizer sharding. When strategies increase in complexity, it becomes necessary for partitioning tools to be 1) expressive, allowing the composition of simpler strategies, and 2) predictable to estimate performance analytically. We present PartIR, our design for a NN partitioning system. PartIR is focused on an incremental approach to rewriting and is hardware-and-runtime agnostic. We present a simple but powerful API for composing sharding strategies and a simulator to validate them. The process is driven by high-level programmer-issued partitioning tactics, which can be both manual and automatic. Importantly, the tactics are specified separately from the model code, making them easy to change. We evaluate PartIR on several different models to demonstrate its predictability, expressibility, and ability to reach peak performance..
@article{arxiv.2401.11202,
title = {PartIR: Composing SPMD Partitioning Strategies for Machine Learning},
author = {Sami Alabed and Daniel Belov and Bart Chrzaszcz and Juliana Franco and Dominik Grewe and Dougal Maclaurin and James Molloy and Tom Natan and Tamara Norman and Xiaoyue Pan and Adam Paszke and Norman A. Rink and Michael Schaarschmidt and Timur Sitdikov and Agnieszka Swietlik and Dimitrios Vytiniotis and Joel Wee},
journal= {arXiv preprint arXiv:2401.11202},
year = {2025}
}