An official website of the United States government
Foundation models have superior performance across a wide array of machine learning tasks. The training of these models typically involves model parallelism (MP) to navigate the constraints of GPU memory capacity. However, MP strategies involve transmitting model activations between GPUs, which can hinder training speed in large clusters. Previous research has examined gradient compression in data-parallel contexts, but its applicability in MP settings remains largely unexplored. In this paper, we investigate the unique characteristics of compression in MP and study why strategies from gradient compression might not be directly applicable to MP scenarios. Subsequently, to systematically understand the capabilities and limitations of Model Parallelism Compression, we present a benchmarking framework MCBench. MCBench not only includes four major categories of compression algorithms but also includes several widely used models spanning language and vision tasks on a well-established distributed training framework, Megatron-LM. We initiate the first comprehensive empirical study by using MCBench. Our empirical study encompasses both the fine-tuning and pre-training of FMs. We probe over 200 unique training configurations and present results using 10 widely used datasets. To comprehend the scalability of compression advantages with the expansion of model size and cluster size, we propose a novel cost model designed specifically for training with MP compression. The insights derived from our findings can help direct the future development of new MP compression algorithms for distributed training. Our code is available at https://github.com/uw-mad-dash/MCBench
more »
« less
Training Recommender Systems at Scale: Communication-Efficient Model and Data Parallelism
In this paper, we consider hybrid parallelism—a paradigm that em- ploys both Data Parallelism (DP) and Model Parallelism (MP)—to scale distributed training of large recommendation models. We propose a compression framework called Dynamic Communication Thresholding (DCT) for communication-efficient hybrid training. DCT filters the entities to be communicated across the network through a simple hard-thresholding function, allowing only the most relevant information to pass through. For communication efficient DP, DCT compresses the parameter gradients sent to the parameter server during model synchronization. The threshold is updated only once every few thousand iterations to reduce the computational overhead of compression. For communication efficient MP, DCT incorporates a novel technique to compress the activations and gradients sent across the network during the forward and backward propagation, respectively. This is done by identifying and updating only the most relevant neurons of the neural network for each training sample in the data. We evaluate DCT on publicly available natural language processing and recommender models and datasets, as well as recommendation systems used in production at Facebook. DCT reduces communication by at least 100× and 20× during DP and MP, respectively. The algorithm has been deployed in production, and it improves end-to-end training time for a state-of-the-art industrial recommender model by 37%, without any loss in performance.
more »
« less
- Award ID(s):
- 1703678
- PAR ID:
- 10327111
- Date Published:
- Journal Name:
- KDD '21: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining
- Page Range / eLocation ID:
- 2928 to 2936
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
The deployment of Deep Learning Recommendation Models (DLRMs) involves the parallelization of extra-large embedding tables (EMTs) on multiple GPUs. Existing works overlook the input-dependent behavior of EMTs and parallelize them in a coarse-grained manner, resulting in unbalanced workload distribution and inter-GPU communication. To this end, we propose OPER, an algorithm-system co-design with OPtimality-guided Embedding table parallelization for large-scale Recommendation model training and inference. The core idea of OPER is to explore the connection between DLRM inputs and the efficiency of distributed EMTs, aiming to provide a near-optimal parallelization strategy for EMTs. Specifically, we conduct an in-depth analysis of various types of EMTs parallelism and propose a heuristic search algorithm to efficiently approximate an empirically near-optimal EMT parallelization. Furthermore, we implement a distributed shared memory-based system, which supports the lightweight but complex computation and communication pattern of fine-grained EMT parallelization, effectively converting theoretical improvements into real speedups. Extensive evaluation shows that OPER achieves 2.3× and 4.0× speedup on average in training and inference, respectively, over state-of-the-art DLRM frameworks.more » « less
-
FPGA-based edge servers are used in many applications in smart cities, hospitals, retail, etc. Equipped with heterogeneous FPGA-based accelerator cards, the servers can be implemented with multiple tasks including efficient video prepossessing, machine learning algorithm acceleration, etc. These servers are required to implement inference during the daytime while re-training the model during the night to adapt to new environments, domains, or new users. During the re-training, conventionally, the incoming data are transmitted to the cloud, and then the updated machine learning models will be transferred back to the edge server. Such a process is inefficient and cannot protect users’ privacy, so it is desirable for the models to be directly trained on the edge servers. Deploying convolutional neural network (CNN) training on heterogeneous resource-constrained FPGAs is challenging since it needs to consider both the complex data dependency of the training process and the communication bottleneck among different FPGAs. Previous multi-accelerator training algorithms select optimal scheduling strategies for data parallelism, tensor parallelism, and pipeline parallelism. However, pipeline parallelism cannot deal with batch normalization (BN) which is an essential CNN operator, while purely applying data parallelism and tensor parallelism suffers from resource under-utilization and intensive communication costs. In this work, we propose MTrain, a novel multi-accelerator training scheduling strategy that transfers the training process into a multi-branch workflow, thus independent sub-operations of different branches are executed on different training accelerators in parallelism for better utilization and reduced communication overhead. Experimental results show that we can achieve efficient CNN training on heterogeneous FPGA-based edge servers with 1.07x-2.21x speedup under 15 GB/s peer-to-peer bandwidth compared to the state-of-the-art work.more » « less
-
Model parameter synchronization across GPUs introduces high overheads for data-parallel training at scale. Existing parameter synchronization protocols cannot effectively leverage available network resources in the face of ever increasing hardware heterogeneity. To address this issue, we propose Blink, a collective communication library that dynamically generates optimal communication primitives by packing spanning trees. We propose techniques to minimize the number of trees generated and extend Blink to leverage heterogeneous communication channels for hybrid, and faster, data transfers. Evaluations show that compared to the state-of-the-art (NCCL), Blink can achieve up to 8× faster model synchronization (AllReduce), and reduce end-to-end DNN training time for image classification tasks by up to 40%.more » « less
-
Deep neural networks (DNNs) are becoming increasingly deeper, wider, and non-linear due to the growing demands on prediction accuracy and analysis quality. Training wide and deep neural networks require large amounts of storage resources such as memory because the intermediate activation data must be saved in the memory during forward propagation and then restored for backward propagation. However, state-of-the-art accelerators such as GPUs are only equipped with very limited memory capacities due to hardware design constraints, which significantly limits the maximum batch size and hence performance speedup when training large-scale DNNs. Traditional memory saving techniques either suffer from performance overhead or are constrained by limited interconnect bandwidth or specific interconnect technology. In this paper, we propose a novel memory-efficient CNN training framework (called COMET) that leverages error-bounded lossy compression to significantly reduce the memory requirement for training in order to allow training larger models or to accelerate training. Our framework purposely adopts error-bounded lossy compression with a strict error-controlling mechanism. Specifically, we perform a theoretical analysis on the compression error propagation from the altered activation data to the gradients, and empirically investigate the impact of altered gradients over the training process. Based on these analyses, we optimize the error-bounded lossy compression and propose an adaptive error-bound control scheme for activation data compression. Experiments demonstrate that our proposed framework can significantly reduce the training memory consumption by up to 13.5X over the baseline training and 1.8X over another state-of-the-art compression-based framework, respectively, with little or no accuracy loss.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3447548.3467080
