arXiv subjects

Shigang Li

Publications and source records attributed to Shigang Li.

30 records · Page 2Linked to original sources

arXiv2023

Efficient Quantized Sparse Matrix Operations on Tensor Cores

The exponentially growing model size drives the continued success of deep learning, but it brings prohibitive computation and memory cost. From the algorithm perspective, model sparsification and quantization have been studied to alleviate the problem. From the architecture perspective, hardware vendors provide Tensor cores for acceleration. However, it is very challenging to gain practical speedups from sparse, low-precision matrix operations on Tensor cores, because of the strict requirements for data layout and lack of support for efficiently manipulating the low-precision integers. We propose Magicube, a high-performance sparse-matrix library for low-precision integers on Tensor cores. Magicube supports SpMM and SDDMM, two major sparse operations in deep learning with mixed precision. Experimental results on an NVIDIA A100 GPU show that Magicube achieves on average 1.44x (up to 2.37x) speedup over the vendor-optimized library for sparse kernels, and 1.43x speedup over the state-of-the-art with a comparable accuracy for end-to-end sparse Transformer inference.

cs.DC↗

arXiv2023

An End-to-End Network for Upright Adjustment of Panoramic Images

Nowadays, panoramic images can be easily obtained by panoramic cameras. However, when the panoramic camera orientation is tilted, a non-upright panoramic image will be captured. Existing upright adjustment models focus on how to estimate more accurate camera orientation, and attribute image reconstruction to offline or post-processing tasks. To this end, we propose an online end-to-end network for upright adjustment. Our network is designed to reconstruct the image while finding the angle. Our network consists of three modules: orientation estimation, LUT online generation, and upright reconstruction. Direction estimation estimates the tilt angle of the panoramic image. Then, a converter block with upsampling function is designed to generate angle to LUT. This module can output corresponding online LUT for different input angles. Finally, a lightweight generative adversarial network (GAN) aims to generate upright images from shallow features. The experimental results show that in terms of angles, we have improved the accuracy of small angle errors. In terms of image reconstruction, In image reconstruction, we have achieved the first real-time online upright reconstruction of panoramic images using deep learning networks.

cs.CV↗

arXiv2022

HammingMesh: A Network Topology for Large-Scale Deep Learning

Numerous microarchitectural optimizations unlocked tremendous processing power for deep neural networks that in turn fueled the AI revolution. With the exhaustion of such optimizations, the growth of modern AI is now gated by the performance of training systems, especially their data movement. Instead of focusing on single accelerators, we investigate data-movement characteristics of large-scale training at full system scale. Based on our workload analysis, we design HammingMesh, a novel network topology that provides high bandwidth at low cost with high job scheduling flexibility. Specifically, HammingMesh can support full bandwidth and isolation to deep learning training jobs with two dimensions of parallelism. Furthermore, it also supports high global bandwidth for generic traffic. Thus, HammingMesh will power future large-scale deep learning systems with extreme bandwidth requirements.

cs.DC↗

arXiv2022

A Data-Centric Optimization Framework for Machine Learning

Rapid progress in deep learning is leading to a diverse set of quickly changing models, with a dramatically growing demand for compute. However, as frameworks specialize performance optimization to patterns in popular networks, they implicitly constrain novel and diverse models that drive progress in research. We empower deep learning researchers by defining a flexible and user-customizable pipeline for optimizing training of arbitrary deep neural networks, based on data movement minimization. The pipeline begins with standard networks in PyTorch or ONNX and transforms computation through progressive lowering. We define four levels of general-purpose transformations, from local intra-operator optimizations to global data movement reduction. These operate on a data-centric graph intermediate representation that expresses computation and data movement at all levels of abstraction, including expanding basic operators such as convolutions to their underlying computations. Central to the design is the interactive and introspectable nature of the pipeline. Every part is extensible through a Python API, and can be tuned interactively using a GUI. We demonstrate competitive performance or speedups on ten different networks, with interactive optimizations discovering new opportunities in EfficientNet.

cs.LG↗

arXiv2022

Asynchronous Decentralized SGD with Quantized and Local Updates

Decentralized optimization is emerging as a viable alternative for scalable distributed machine learning, but also introduces new challenges in terms of synchronization costs. To this end, several communication-reduction techniques, such as non-blocking communication, quantization, and local steps, have been explored in the decentralized setting. Due to the complexity of analyzing optimization in such a relaxed setting, this line of work often assumes \emph{global} communication rounds, which require additional synchronization. In this paper, we consider decentralized optimization in the simpler, but harder to analyze, \emph{asynchronous gossip} model, in which communication occurs in discrete, randomly chosen pairings among nodes. Perhaps surprisingly, we show that a variant of SGD called \emph{SwarmSGD} still converges in this setting, even if \emph{non-blocking communication}, \emph{quantization}, and \emph{local steps} are all applied \emph{in conjunction}, and even if the node data distributions and underlying graph topology are both \emph{heterogenous}. Our analysis is based on a new connection with multi-dimensional load-balancing processes. We implement this algorithm and deploy it in a super-computing environment, showing that it can outperform previous decentralized methods in terms of end-to-end training time, and that it can even rival carefully-tuned large-batch SGD for certain tasks.

cs.LG↗

arXiv2021

Data Movement Is All You Need: A Case Study on Optimizing Transformers

Transformers are one of the most important machine learning workloads today. Training one is a very compute-intensive task, often taking days or weeks, and significant attention has been given to optimizing transformers. Despite this, existing implementations do not efficiently utilize GPUs. We find that data movement is the key bottleneck when training. Due to Amdahl's Law and massive improvements in compute performance, training has now become memory-bound. Further, existing frameworks use suboptimal data layouts. Using these insights, we present a recipe for globally optimizing data movement in transformers. We reduce data movement by up to 22.91% and overall achieve a 1.30x performance improvement over state-of-the-art frameworks when training a BERT encoder layer and 1.19x for the entire BERT. Our approach is applicable more broadly to optimizing deep neural networks, and offers insight into how to tackle emerging performance bottlenecks.

cs.LG↗

arXiv2021

Flare: Flexible In-Network Allreduce

The allreduce operation is one of the most commonly used communication routines in distributed applications. To improve its bandwidth and to reduce network traffic, this operation can be accelerated by offloading it to network switches, that aggregate the data received from the hosts, and send them back the aggregated result. However, existing solutions provide limited customization opportunities and might provide suboptimal performance when dealing with custom operators and data types, with sparse data, or when reproducibility of the aggregation is a concern. To deal with these problems, in this work we design a flexible programmable switch by using as a building block PsPIN, a RISC-V architecture implementing the sPIN programming model. We then design, model, and analyze different algorithms for executing the aggregation on this architecture, showing performance improvements compared to state-of-the-art approaches.

cs.DC↗

arXiv2020

Deep Learning for Post-Processing Ensemble Weather Forecasts

Quantifying uncertainty in weather forecasts is critical, especially for predicting extreme weather events. This is typically accomplished with ensemble prediction systems, which consist of many perturbed numerical weather simulations, or trajectories, run in parallel. These systems are associated with a high computational cost and often involve statistical post-processing steps to inexpensively improve their raw prediction qualities. We propose a mixed model that uses only a subset of the original weather trajectories combined with a post-processing step using deep neural networks. These enable the model to account for non-linear relationships that are not captured by current numerical models or post-processing methods. Applied to global data, our mixed models achieve a relative improvement in ensemble forecast skill (CRPS) of over 14%. Furthermore, we demonstrate that the improvement is larger for extreme weather events on select case studies. We also show that our post-processing can use fewer trajectories to achieve comparable results to the full ensemble. By using fewer trajectories, the computational costs of an ensemble prediction system can be reduced, allowing it to run at higher resolution and produce more accurate forecasts.

cs.LG↗

arXiv2020

The static parallel distribution algorithms for hybrid density-functional calculations in HONPAS package

Hybrid density-functional calculation is one of the most commonly adopted electronic structure theory used in computational chemistry and materials science because of its balance between accuracy and computational cost. Recently, we have developed a novel scheme called NAO2GTO to achieve linear scaling (Order-N) calculations for hybrid density-functionals. In our scheme, the most time-consuming step is the calculation of the electron repulsion integrals (ERIs) part. So how to create an even distribution of these ERIs in parallel implementation is an issue of particular importance. Here, we present two static scalable distributed algorithms for the ERIs computation. Firstly, the ERIs are distributed over ERIs shell pairs. Secondly, the ERIs is distributed over ERIs shell quartets. In both algorithms, the calculation of ERIs is independent of each other, so the communication time is minimized. We show our speedup results to demonstrate the performance of these static parallel distributed algorithms in the Hefei Order-N packages for \textit{ab initio} simulations (HONPAS).

physics.comp-ph↗

arXiv2020

The Scalability for Parallel Machine Learning Training Algorithm: Dataset Matters

To gain a better performance, many researchers put more computing resource into an application. However, in the AI area, there is still a lack of a successful large-scale machine learning training application: The scalability and performance reproducibility of parallel machine learning training algorithm are limited and there are a few pieces of research focusing on why these indexes are limited but there are very few research efforts explaining the reasons in essence. In this paper, we propose that the sample difference in dataset plays a more prominent role in parallel machine learning algorithm scalability. Dataset characters can measure sample difference. These characters include the variance of the sample in a dataset, sparsity, sample diversity and similarity in sampling sequence. To match our proposal, we choose four kinds of parallel machine learning training algorithms as our research objects: (1) Asynchronous parallel SGD algorithm (Hogwild! algorithm) (2) Parallel model average SGD algorithm (Mini-batch SGD algorithm) (3) Decenterilization optimization algorithm, (4) Dual Coordinate Optimization (DADM algorithm). These algorithms cover different types of machine learning optimization algorithms. We present the analysis of their convergence proof and design experiments. Our results show that the characters datasets decide the scalability of the machine learning algorithm. What is more, there is an upper bound of parallel scalability for machine learning algorithms.

cs.DC↗

arXiv2019

Predicting Weather Uncertainty with Deep Convnets

Modern weather forecast models perform uncertainty quantification using ensemble prediction systems, which collect nonparametric statistics based on multiple perturbed simulations. To provide accurate estimation, dozens of such computationally intensive simulations must be run. We show that deep neural networks can be used on a small set of numerical weather simulations to estimate the spread of a weather forecast, significantly reducing computational cost. To train the system, we both modify the 3D U-Net architecture and explore models that incorporate temporal data. Our models serve as a starting point to improve uncertainty quantification in current real-time weather forecasting systems, which is vital for predicting extreme events.

cs.LG↗

arXiv2017

Asynchronous COMID: the theoretic basis for transmitted data sparsification tricks on Parameter Server

Asynchronous FTRL-proximal and L2 norm done at server are two widely used tricks in Parameters Server which is an implement of delayed SGD. Their commonness is leaving parts of updating computation on server which reduces the burden of network via making transmitted data sparse. But above tricks' convergences are not well-proved. In this paper, based on above commonness, we propose a more general algorithm named as asynchronous COMID and prove its convergence. We prove that asynchronous FTRL-proximal and L2 norm done at server are applications of asynchronous COMID, which demonstrates the convergences of above two tricks. Then, we conduct experiments to verify theoretical results. Experimental results show that compared with delayed SGD on Parameters Server, asynchronous COMID reduces the burden of the network without any harm on the mathematical convergence speed and final output.

cs.DC↗