An official website of the United States government
Large-scale machine learning (ML) models are increasingly being used in critical domains like education, lending, recruitment, healthcare, criminal justice, etc. However, the training, deployment, and utilization of these models demand substantial computational resources. To decrease computation and memory costs, machine learning models with sparse weight matrices are widely used in the literature. Among sparse models, those with special sparse structures (e.g., models with block-wise sparse weight matrices) fit better with the hardware accelerators and can decrease the memory and computation costs during the inference. Unfortunately, while there are several efficient training methods, none of them are designed to train a block-wise sparse model efficiently. As a result, the current methods for training block-wise sparse models start with full and dense models leading to inefficient training. In this work, we focus on training models with block-wise sparse matrices and propose an efficient training algorithm to decrease both computation and memory costs during training and inference. In addition, we will show that our proposed method enables us to efficiently find the right block size for the sparsity pattern during the training process. Our extensive empirical and theoretical analyses show that our algorithms can decrease the computation and memory costs significantly without a performance drop compared to baselines.
more »
« less
Memory-Efficient Pipeline-Parallel DNN Training
Many state-of-the-art ML results have been obtained by scaling up the number of parameters in existing models. However, parameters and activations for such large models often do not fit in the memory of a single accelerator device; this means that it is necessary to distribute training of large models over multiple accelerators. In this work, we propose PipeDream-2BW, a system that supports memory-efficient pipeline parallelism. PipeDream-2BW uses a novel pipelining and weight gradient coalescing strategy, combined with the double buffering of weights, to ensure high throughput, low memory footprint, and weight update semantics similar to data parallelism. In addition, PipeDream-2BW automatically partitions the model over the available hardware resources, while respecting hardware constraints such as memory capacities of accelerators and interconnect topologies. PipeDream-2BW can accelerate the training of large GPT and BERT language models by up to 20x with similar final model accuracy.
more »
« less
- Award ID(s):
- 1651570
- PAR ID:
- 10327318
- Date Published:
- Journal Name:
- Proceedings of Machine Learning Research
- Volume:
- 139
- ISSN:
- 2640-3498
- Page Range / eLocation ID:
- 7937-7947
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Large language models have led to state-of-the-art accuracies across several tasks. However, training these models efficiently is challenging because: a) GPU memory capacity is limited, making it impossible to fit large models on even a multi-GPU server, and b) the number of compute operations required can result in unrealistically long training times. Consequently, new methods of model parallelism such as tensor and pipeline parallelism have been proposed. Unfortunately, naive usage of these methods leads to scaling issues at thousands of GPUs. In this paper, we show how tensor, pipeline, and data parallelism can be composed to scale to thousands of GPUs. We propose a novel interleaved pipelining schedule that can improve throughput by 10+% with memory footprint comparable to existing approaches. Our approach allows us to perform training iterations on a model with 1 trillion parameters at 502 petaFLOP/s on 3072 GPUs (per-GPU throughput of 52% of theoretical peak).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
-
Open Modification Search (OMS) is a promising algorithm for mass spectrometry analysis that enables the discovery of modified peptides. However, OMS encounters challenges as it exponentially extends the search scope. Existing OMS accelerators either have limited parallelism or struggle to scale effectively with growing data volumes. In this work, we introduce an OMS accelerator utilizing multi-level-cell (MLC) RRAM memory to enhance storage capacity by 3x. Through in-memory computing, we achieve up to 77x faster data processing with two to three orders of magnitude better energy efficiency. Testing was done on a fabricated MLC RRAM chip. We leverage hyperdimensional computing to tolerate up to 10% memory errors while delivering massive parallelism in hardware.more » « less
-
null (Ed.)PIM (processing-in-memory) based hardware accelerators have shown great potentials in addressing the computation and memory access intensity of modern CNNs (convolutional neural networks). While adopting NVM (non-volatile memory) helps to further mitigate the storage and energy consumption overhead, adopting quantization, e.g., shift-based quantization, helps to tradeoff the computation overhead and the accuracy loss, integrating both NVM and quantization in hardware accelerators leads to sub-optimal acceleration. In this paper, we exploit the natural shift property of DWM (domain wall memory) to devise DWMAcc, a DWM-based accelerator with asymmetrical storage of weight and input data, to speed up the inference phase of shift-based CNNs. DWMAcc supports flexible shift operations to enable fast processing with low performance and area overhead. We then optimize it with zero-sharing , input-reuse , and weight-share schemes. Our experimental results show that, on average, DWMAcc achieves 16.6× performance improvement and 85.6× energy consumption reduction over a state-of-the-art SRAM based design.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
