# Accelerating Large Language Model Training with Hybrid GPU-based Compression

arXiv: https://arxiv.org/abs/2409.02423 · HTML: https://mlsanity.com/paper/2409.02423/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 19 of its 44 references that are in the map.
- First posted on arXiv 4 Sep 2024.

## Builds on (most cited first, 19 of 19)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://mlsanity.com/paper/2101.00027/) (2020)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [Mixed Precision Training](https://mlsanity.com/paper/1710.03740/) (2017)
- [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://mlsanity.com/paper/2204.06745/) (2022)
- [signSGD: Compressed Optimisation for Non-Convex Problems](https://mlsanity.com/paper/1802.04434/) (2018)
- [Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model](https://mlsanity.com/paper/2201.11990/) (2022)
- [Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training](https://mlsanity.com/paper/1712.01887/) (2017)
- [8-bit Optimizers via Block-wise Quantization](https://mlsanity.com/paper/2110.02861/) (2021)
- [ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning](https://mlsanity.com/paper/2104.07857/) (2021)
- [Demystifying Parallel and Distributed Deep Learning: An In-Depth Concurrency Analysis](https://mlsanity.com/paper/1802.09941/) (2018)
- [TeraPipe: Token-Level Pipeline Parallelism for Training Large-Scale Language Models](https://mlsanity.com/paper/2102.07988/) (2021)
- [ZeRO++: Extremely Efficient Collective Communication for Giant Model Training](https://mlsanity.com/paper/2306.10209/) (2023)
- [cuSZ: An Efficient GPU-Based Error-Bounded Lossy Compression Framework for Scientific Data](https://mlsanity.com/paper/2007.09625/) (2020)
- [Does compressing activations help model parallel training?](https://mlsanity.com/paper/2301.02654/) (2023)
- [MCR-DL: Mix-and-Match Communication Runtime for Deep Learning](https://mlsanity.com/paper/2303.08374/) (2023)

## Built on by (most cited first, 2 of 2)

- [TACO: Efficient Communication Compression of Intermediate Tensors for Scalable Tensor-Parallel LLM Training](https://mlsanity.com/paper/2604.24088/) (2026)
- [Scaling Large Language Model Training on Frontier with Low-Bandwidth Partitioning](https://mlsanity.com/paper/2501.04266/) (2025)

## Abstract (arXiv)

Data Parallelism (DP), Tensor Parallelism (TP), and Pipeline Parallelism (PP) are the three strategies widely adopted to enable fast and efficient Large Language Model (LLM) training. However, these approaches rely on data-intensive communication routines to collect, aggregate, and re-distribute gradients, activations, and other important model information, which pose significant overhead. Co-designed with GPU-based compression libraries, MPI libraries have been proven to reduce message size significantly, and leverage interconnect bandwidth, thus increasing training efficiency while maintaining acceptable accuracy. In this work, we investigate the efficacy of compression-assisted MPI collectives under the context of distributed LLM training using 3D parallelism and ZeRO optimizations. We scaled up to 192 V100 GPUs on the Lassen supercomputer. First, we enabled a na"ive compression scheme across all collectives and observed a 22.5% increase in TFLOPS per GPU and a 23.6% increase in samples per second for GPT-NeoX-20B training. Nonetheless, such a strategy ignores the sparsity discrepancy among messages communicated in each parallelism degree, thus introducing more errors and causing degradation in training loss. Therefore, we incorporated hybrid compression settings toward each parallel dimension and adjusted the compression intensity accordingly. Given their low-rank structure (arXiv:2301.02654), we apply aggressive compression on gradients when performing DP All-reduce. We adopt milder compression to preserve precision while communicating activations, optimizer states, and model parameters in TP and PP. Using the adjusted hybrid compression scheme, we demonstrate a 17.3% increase in TFLOPS per GPU and a 12.7% increase in samples per second while reaching baseline loss convergence.