Rockmate: an Efficient, Fast, Automatic and Generic Tool for Re-materialization in PyTorch

TL;DR

Rockmate is an automatic, efficient tool for controlling memory usage during PyTorch model training by re-computing activations, applicable to various model architectures, with minimal overhead.

Contribution

It introduces Rockmate, a novel method that automatically rewrites models into block structures for efficient memory management during training.

Findings

Reduces activation memory by a factor of 2 to 5

Achieves speed comparable to Rotor and efficiency similar to Checkmate

Maintains low overhead of 10-20% during training

Abstract

We propose Rockmate to control the memory requirements when training PyTorch DNN models. Rockmate is an automatic tool that starts from the model code and generates an equivalent model, using a predefined amount of memory for activations, at the cost of a few re-computations. Rockmate automatically detects the structure of computational and data dependencies and rewrites the initial model as a sequence of complex blocks. We show that such a structure is widespread and can be found in many models in the literature (Transformer based models, ResNet, RegNets,...). This structure allows us to solve the problem in a fast and efficient way, using an adaptation of Checkmate (too slow on the whole model but general) at the level of individual blocks and an adaptation of Rotor (fast but limited to sequential models) at the level of the sequence itself. We show through experiments on many models that Rockmate is as fast as Rotor and as efficient as Checkmate, and that it allows in many cases to obtain a significantly lower memory consumption for activations (by a factor of 2 to 5) for a rather negligible overhead (of the order of 10% to 20%). Rockmate is open source and available at https://github.com/topal-team/rockmate.