NeuZip: Memory-Efficient Training and Inference with Dynamic Compression of Neural Networks
This addresses memory limitations for on-device AI applications, offering a practical solution with significant memory savings, though it is incremental as it builds on existing compression techniques.
The paper tackles the problem of memory constraints in neural network training and inference by introducing NeuZip, a weight compression scheme based on entropy of floating-point numbers, achieving a reduction in memory footprint from 31GB to less than 16GB for training a Llama-3 8B model without performance loss.
The performance of neural networks improves when more parameters are used. However, the model sizes are constrained by the available on-device memory during training and inference. Although applying techniques like quantization can alleviate the constraint, they suffer from performance degradation. In this work, we introduce NeuZip, a new weight compression scheme based on the entropy of floating-point numbers in neural networks. With NeuZip, we are able to achieve memory-efficient training and inference without sacrificing performance. Notably, we significantly reduce the memory footprint of training a Llama-3 8B model from 31GB to less than 16GB, while keeping the training dynamics fully unchanged. In inference, our method can reduce memory usage by more than half while maintaining near-lossless performance. Our code is publicly available.