The rapid rise in demand for training large neural network architectures has brought into focus the need for partitioning strategies, for example by using data, model, or pipeline parallelism. Implementing these methods is increasingly supported through program primitives, but identifying efficient partitioning strategies requires expensive experimentation and expertise. We present the prototype of an automated partitioner that seamlessly integrates into existing compilers and existing user workflows. Our partitioner enables SPMD-style parallelism that encompasses data parallelism and parameter/activation sharding. Through a combination of inductive tactics and search in a platform-independent partitioning IR, automap can recover expert partitioning strategies such as Megatron sharding for transformer layers.
@article{arxiv.2112.02958,
title = {Automap: Towards Ergonomic Automated Parallelism for ML Models},
author = {Michael Schaarschmidt and Dominik Grewe and Dimitrios Vytiniotis and Adam Paszke and Georg Stefan Schmid and Tamara Norman and James Molloy and Jonathan Godwin and Norman Alexander Rink and Vinod Nair and Dan Belov},
journal= {arXiv preprint arXiv:2112.02958},
year = {2021}
}