اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدTraining Recommender Systems at Scale: Communication-Efficient Model and Data Parallelism
108
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
In this paper, we consider hybrid parallelism -- a paradigm that employs 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 $100times$ and $20times$ 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.
قيم البحث
اقرأ أيضاً
We present Distributed Equivalent Substitution (DES) training, a novel distributed training framework for large-scale recommender systems with dynamic sparse features. DES introduces fully synchronous training to large-scale recommendation system for
the first time by reducing communication, thus making the training of commercial recommender systems converge faster and reach better CTR. DES requires much less communication by substituting the weights-rich operators with the computationally equivalent sub-operators and aggregating partial results instead of transmitting the huge sparse weights directly through the network. Due to the use of synchronous training on large-scale Deep Learning Recommendation Models (DLRMs), DES achieves higher AUC(Area Under ROC). We successfully apply DES training on multiple popular DLRMs of industrial scenarios. Experiments show that our implementation outperforms the state-of-the-art PS-based training framework, achieving up to 68.7% communication savings and higher throughput compared to other PS-based recommender systems.
Scalable training of large models (like BERT and GPT-3) requires careful optimization rooted in model design, architecture, and system capabilities. From a system standpoint, communication has become a major bottleneck, especially on commodity system
s with standard TCP interconnects that offer limited network bandwidth. Communication compression is an important technique to reduce training time on such systems. One of the most effective methods is error-compensated compression, which offers robust convergence speed even under 1-bit compression. However, state-of-the-art error compensation techniques only work with basic optimizers like SGD and momentum SGD, which are linearly dependent on the gradients. They do not work with non-linear gradient-based optimizers like Adam, which offer state-of-the-art convergence efficiency and accuracy for models like BERT. In this paper, we propose 1-bit Adam that reduces the communication volume by up to $5times$, offers much better scalability, and provides the same convergence speed as uncompressed Adam. Our key finding is that Adams variance (non-linear term) becomes stable (after a warmup phase) and can be used as a fixed precondition for the rest of the training (compression phase). Experiments on up to 256 GPUs show that 1-bit Adam enables up to $3.3times$ higher throughput for BERT-Large pre-training and up to $2.9times$ higher throughput for SQuAD fine-tuning. In addition, we provide theoretical analysis for our proposed work.
To train large models (like BERT and GPT-3) with hundreds or even thousands of GPUs, the communication has become a major bottleneck, especially on commodity systems with limited-bandwidth TCP interconnects network. On one side large-batch optimizati
on such as LAMB algorithm was proposed to reduce the number of communications. On the other side, communication compression algorithms such as 1-bit SGD and 1-bit Adam help to reduce the volume of each communication. However, we find that simply using one of the techniques is not sufficient to solve the communication challenge, especially on low-bandwidth Ethernet networks. Motivated by this we aim to combine the power of large-batch optimization and communication compression, but we find that existing compression strategies cannot be directly applied to LAMB due to its unique adaptive layerwise learning rates. To this end, we design a new communication-efficient algorithm, 1-bit LAMB, which introduces a novel way to support adaptive layerwise learning rates even when communication is compressed. In addition, we introduce a new system implementation for compressed communication using the NCCL backend of PyTorch distributed, which improves both usability and performance compared to existing MPI-based implementation. For BERT-Large pre-training task with batch sizes from 8K to 64K, our evaluations on up to 256 GPUs demonstrate that 1-bit LAMB with NCCL-based backend is able to achieve up to 4.6x communication volume reduction, up to 2.8x end-to-end speedup (in terms of number of training samples per second), and the same convergence speed (in terms of number of pre-training samples to reach the same accuracy on fine-tuning tasks) compared to uncompressed LAMB.
Communication of model updates between client nodes and the central aggregating server is a major bottleneck in federated learning, especially in bandwidth-limited settings and high-dimensional models. Gradient quantization is an effective way of red
ucing the number of bits required to communicate each model update, albeit at the cost of having a higher error floor due to the higher variance of the stochastic gradients. In this work, we propose an adaptive quantization strategy called AdaQuantFL that aims to achieve communication efficiency as well as a low error floor by changing the number of quantization levels during the course of training. Experiments on training deep neural networks show that our method can converge in much fewer communicated bits as compared to fixed quantization level setups, with little or no impact on training and test accuracy.
Federated learning (FL) is a distributed learning methodology that allows multiple nodes to cooperatively train a deep learning model, without the need to share their local data. It is a promising solution for telemonitoring systems that demand inten
sive data collection, for detection, classification, and prediction of future events, from different locations while maintaining a strict privacy constraint. Due to privacy concerns and critical communication bottlenecks, it can become impractical to send the FL updated models to a centralized server. Thus, this paper studies the potential of hierarchical FL in IoT heterogeneous systems and propose an optimized solution for user assignment and resource allocation on multiple edge nodes. In particular, this work focuses on a generic class of machine learning models that are trained using gradient-descent-based schemes while considering the practical constraints of non-uniformly distributed data across different users. We evaluate the proposed system using two real-world datasets, and we show that it outperforms state-of-the-art FL solutions. In particular, our numerical results highlight the effectiveness of our approach and its ability to provide 4-6% increase in the classification accuracy, with respect to hierarchical FL schemes that consider distance-based user assignment. Furthermore, the proposed approach could significantly accelerate FL training and reduce communication overhead by providing 75-85% reduction in the communication rounds between edge nodes and the centralized server, for the same model accuracy.
الأسئلة المقترحة
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
