Coop: Memory is not a Commodity
This addresses memory inefficiency in deep learning frameworks for researchers and practitioners, representing an incremental improvement over existing tensor rematerialization techniques.
The paper tackles memory fragmentation in tensor rematerialization for deep neural network training by proposing Coop, a method that co-optimizes tensor allocation and rematerialization, achieving up to 2× memory savings and significant reductions in compute overhead, search latency, and fragmentation compared to state-of-the-art baselines.
Tensor rematerialization allows the training of deep neural networks (DNNs) under limited memory budgets by checkpointing the models and recomputing the evicted tensors as needed. However, the existing tensor rematerialization techniques overlook the memory system in deep learning frameworks and implicitly assume that free memory blocks at different addresses are identical. Under this flawed assumption, discontiguous tensors are evicted, among which some are not used to allocate the new tensor. This leads to severe memory fragmentation and increases the cost of potential rematerializations. To address this issue, we propose to evict tensors within a sliding window to ensure all evictions are contiguous and are immediately used. Furthermore, we proposed cheap tensor partitioning and recomputable in-place to further reduce the rematerialization cost by optimizing the tensor allocation. We named our method Coop as it is a co-optimization of tensor allocation and tensor rematerialization. We evaluated Coop on eight representative DNNs. The experimental results demonstrate that Coop achieves up to $2\times$ memory saving and hugely reduces compute overhead, search latency, and memory fragmentation compared to the state-of-the-art baselines.