Accelerating Large Language Model Training with Hybrid GPU-based Compression

• URL: http://arxiv.org/abs/2409.02423v1
• Date: Wed, 4 Sep 2024 04:05:30 GMT
• Title: Accelerating Large Language Model Training with Hybrid GPU-based Compression
• Authors: Lang Xu, Quentin Anthony, Qinghua Zhou, Nawras Alnaasan, Radha R. Gulhane, Aamir Shafi, Hari Subramoni, Dhabaleswar K. Panda,
• Abstract summary: MPI libraries have been proven to reduce message size significantly and leverage interconnect bandwidth. We investigate the efficacy of compression-assisted MPI collectives under the context of distributed Large Language Model (LLM) training.
• Score: 3.204387803072905
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• LoRC: Low-Rank Compression for LLMs KV Cache with a Progressive Compression Strategy [59.1298692559785]
  Key-Value ( KV) cache is crucial component in serving transformer-based autoregressive large language models (LLMs) Existing approaches to mitigate this issue include: (1) efficient attention variants integrated in upcycling stages; (2) KV cache compression at test time; and (3) KV cache compression at test time. We propose a low-rank approximation of KV weight matrices, allowing plug-in integration with existing transformer-based LLMs without model retraining. Our method is designed to function without model tuning in upcycling stages or task-specific profiling in test stages.
  arXiv  Detail & Related papers  (2024-10-04T03:10:53Z)
• LoCo: Low-Bit Communication Adaptor for Large-scale Model Training [63.040522637816906]
  Low-bit communication often degrades training quality due to compression information loss. We propose Low-bit Communication Adaptor (LoCo), which compensates local local GPU nodes before, without compromising quality. Experimental results show that across moving large-scale training model frameworks like Megatron-LM and PyTorchs FSDP, LoCo significantly improves compression communication efficiency.
  arXiv  Detail & Related papers  (2024-07-05T13:01:36Z)
• Accelerating Communication in Deep Learning Recommendation Model Training with Dual-Level Adaptive Lossy Compression [10.233937665979694]
  DLRM is a state-of-the-art recommendation system model that has gained widespread adoption across various industry applications. A significant bottleneck in this process is the time-consuming all-to-all communication required to collect embedding data from all devices. We introduce a method that employs error-bounded lossy compression to reduce the communication data size and accelerate DLRM training.
  arXiv  Detail & Related papers  (2024-07-05T05:55:18Z)
• Accelerating Distributed Deep Learning using Lossless Homomorphic Compression [17.654138014999326]
  We introduce a novel compression algorithm that effectively merges worker-level compression with in-network aggregation. We show up to a 6.33$times$ improvement in aggregation throughput and a 3.74$times$ increase in per-iteration training speed.
  arXiv  Detail & Related papers  (2024-02-12T09:57:47Z)
• GraVAC: Adaptive Compression for Communication-Efficient Distributed DL Training [0.0]
  Distributed data-parallel (DDP) training improves overall application throughput as multiple devices train on a subset of data and aggregate updates to produce a globally shared model. GraVAC is a framework to dynamically adjust compression factor throughout training by evaluating model progress and assessing information loss associated with compression. As opposed to using a static compression factor, GraVAC reduces end-to-end training time for ResNet101, VGG16 and LSTM by 4.32x, 1.95x and 6.67x respectively.
  arXiv  Detail & Related papers  (2023-05-20T14:25:17Z)
• Optimus-CC: Efficient Large NLP Model Training with 3D Parallelism Aware Communication Compression [8.591088380355252]
  We present Optimus-CC, a fast and scalable distributed training framework for large NLP models with aggressive communication compression. We propose techniques to avoid the model quality drop that comes from the compression. We demonstrate our solution on a GPU cluster, and achieve superior speedup from the baseline state-of-the-art solutions for distributed training without sacrificing the model quality.
  arXiv  Detail & Related papers  (2023-01-24T06:07:55Z)
• Compressed Communication for Distributed Training: Adaptive Methods and System [13.244482588437972]
  Communication overhead severely hinders the scalability of distributed machine learning systems. Recently, there has been a growing interest in using gradient compression to reduce the communication overhead. In this paper, we first introduce a novel adaptive gradient method with gradient compression.
  arXiv  Detail & Related papers  (2021-05-17T13:41:47Z)
• An Efficient Statistical-based Gradient Compression Technique for Distributed Training Systems [77.88178159830905]
  Sparsity-Inducing Distribution-based Compression (SIDCo) is a threshold-based sparsification scheme that enjoys similar threshold estimation quality to deep gradient compression (DGC) Our evaluation shows SIDCo speeds up training by up to 41:7%, 7:6%, and 1:9% compared to the no-compression baseline, Topk, and DGC compressors, respectively.
  arXiv  Detail & Related papers  (2021-01-26T13:06:00Z)
• Training Recommender Systems at Scale: Communication-Efficient Model and Data Parallelism [56.78673028601739]
  We propose a compression framework called Dynamic Communication Thresholding (DCT) for communication-efficient hybrid training. DCT reduces communication by at least $100times$ and $20times$ during DP and MP, respectively. It improves end-to-end training time for a state-of-the-art industrial recommender model by 37%, without any loss in performance.
  arXiv  Detail & Related papers  (2020-10-18T01:44:42Z)
• Sparse Communication for Training Deep Networks [56.441077560085475]
  Synchronous gradient descent (SGD) is the most common method used for distributed training of deep learning models. In this algorithm, each worker shares its local gradients with others and updates the parameters using the average gradients of all workers. We study several compression schemes and identify how three key parameters affect the performance.
  arXiv  Detail & Related papers  (2020-09-19T17:28:11Z)
• PowerGossip: Practical Low-Rank Communication Compression in Decentralized Deep Learning [62.440827696638664]
  We introduce a simple algorithm that directly compresses the model differences between neighboring workers. Inspired by the PowerSGD for centralized deep learning, this algorithm uses power steps to maximize the information transferred per bit.
  arXiv  Detail & Related papers  (2020-08-04T09:14:52Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.