AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs

TL;DR

AGoQ introduces novel activation and gradient quantization techniques that significantly reduce memory usage and accelerate training of large language models without sacrificing accuracy.

Contribution

The paper presents AGoQ, a new quantization method with layer-aware activation and 8-bit gradient quantization for efficient LLM training.

Findings

Reduces memory by up to 52% during training.

Achieves up to 1.34× faster training speed.

Maintains comparable accuracy and convergence on LLaMA models.

Abstract

Quantization is a key method for reducing the GPU memory requirement of training large language models (LLMs). Yet, current approaches are ineffective for 4-bit activations and 8-bit gradients, which would easily cause slow convergence or accuracy loss. To address this, we introduce AGoQ, incorporating two new techniques: 1) a layer-aware activation quantization algorithm that allocates appropriate bit-widths for activations of various layers based on their types and pipeline stages to achieve near 4-bit activation storage, and 2) a gradient quantization algorithm that reduces memory usage and shortens communication time by employing 8-bit gradient storage and precision-preserving 8-bit All-Reduce communication. We conduct extensive experiments using different sizes of LLMs on two GPU clusters (up to 64 GPUs), and the experimental results show that our AGoQ reduces the memory by up to 52\% and achieves up to 1.34$\times$ improvement of training speed compared to state-of-the-art training systems Megatron-LM (w/ or w/o ZeRO), COAT and DeepSpeed with 8B to 32B LLaMA models, while achieving convergence loss on pretraining and comparable accuracy on downstream tasks with LLaMA architectures.