# Coop: Memory is not a Commodity

arXiv: https://arxiv.org/abs/2311.00591 · HTML: https://mlsanity.com/paper/2311.00591/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 37 references that are in the map.
- First posted on arXiv 1 Nov 2023.

## Builds on (most cited first, 22 of 22)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift](https://mlsanity.com/paper/1502.03167/) (2015)
- [Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding](https://mlsanity.com/paper/2205.11487/) (2022)
- [Zero-Shot Text-to-Image Generation](https://mlsanity.com/paper/2102.12092/) (2021)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [Reformer: The Efficient Transformer](https://mlsanity.com/paper/2001.04451/) (2020)
- [Group Normalization](https://mlsanity.com/paper/1803.08494/) (2018)
- [ZeRO: Memory Optimizations Toward Training Trillion Parameter Models](https://mlsanity.com/paper/1910.02054/) (2019)
- [Automatic differentiation in machine learning: a survey](https://mlsanity.com/paper/1502.05767/) (2015)
- [Training Deep Nets with Sublinear Memory Cost](https://mlsanity.com/paper/1604.06174/) (2016)
- [Reducing Activation Recomputation in Large Transformer Models](https://mlsanity.com/paper/2205.05198/) (2022)
- [The Reversible Residual Network: Backpropagation Without Storing Activations](https://mlsanity.com/paper/1707.04585/) (2017)
- [MXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems](https://mlsanity.com/paper/1512.01274/) (2015)
- [Beyond Data and Model Parallelism for Deep Neural Networks](https://mlsanity.com/paper/1807.05358/) (2018)
- [Checkmate: Breaking the Memory Wall with Optimal Tensor Rematerialization](https://mlsanity.com/paper/1910.02653/) (2019)
- [Memory-Efficient Backpropagation Through Time](https://mlsanity.com/paper/1606.03401/) (2016)
- [ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training](https://mlsanity.com/paper/2104.14129/) (2021)
- [Dynamic Tensor Rematerialization](https://mlsanity.com/paper/2006.09616/) (2020)
- [OneFlow: Redesign the Distributed Deep Learning Framework from Scratch](https://mlsanity.com/paper/2110.15032/) (2021)
- [Koka: Programming with Row Polymorphic Effect Types](https://mlsanity.com/paper/1406.2061/) (2014)

## Built on by (most cited first, 3 of 3)

- [MEMO: Fine-grained Tensor Management For Ultra-long Context LLM Training](https://mlsanity.com/paper/2407.12117/) (2024)
- [Deterministic Regime Switching and Feasibility Inversion in Dynamic Tensor Rematerialization](https://mlsanity.com/paper/2609.31250/) (2026)
- [CarbonScaling: Extending Neural Scaling Laws for Carbon Footprint in Large Language Models](https://mlsanity.com/paper/2508.06524/) (2025)

## Abstract (arXiv)

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× memory saving and hugely reduces compute overhead, search latency, and memory fragmentation compared to the state-of-the-art baselines.