PyTorchKR Papers

최신 논문

총 5,576편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2306.07030

Resource Efficient Neural Networks Using Hessian Based Pruning

Neural network pruning is a practical way for reducing the size of trained models and the number of floating-point operations. One way of pruning is to use the relative Hessian trace to calculate sensitivity of each channel, as compared to the more common magnitude pruning approach. However, the stochastic approach used to estimate the Hessian trace needs to iterate over many times before it can converge. This can be time-consuming when used for larger models with many millions of parameters. To address this problem, we modify the existing approach by estimating the Hessian trace using FP16 precision instead of FP32. We test the modified approach (EHAP) on ResNet-32/ResNet-56/WideResNet-28-8 trained on CIFAR10/CIFAR100 image classification tasks and achieve faster computation of the Hessian trace. Specifically, our modified approach can achieve speed ups ranging from 17% to as much as 44% during our experiments on different combinations of model architectures and GPU devices. Our modified approach also takes up around 40% less GPU memory when pruning ResNet-32 and ResNet-56 models, which allows for a larger Hessian batch size to be used for estimating the Hessian trace. Meanwhile, we also present the results of pruning using both FP16 and FP32 Hessian trace calculation and show that there are no noticeable accuracy differences between the two. Overall, it is a simple and effective way to compute the relative Hessian trace faster without sacrificing on pruned model performance. We also present a full pipeline using EHAP and quantization aware training (QAT), using INT8 QAT to compress the network further after pruning. In particular, we use symmetric quantization for the weights and asymmetric quantization for the activations.

arXiv:2306.06000

S$^{3}$: Increasing GPU Utilization during Generative Inference for Higher Throughput

Generating texts with a large language model (LLM) consumes massive amounts of memory. Apart from the already-large model parameters, the key/value (KV) cache that holds information about previous tokens in a sequence can grow to be even larger than the model itself. This problem is exacerbated in one of the current LLM serving frameworks which reserves the maximum sequence length of memory for the KV cache to guarantee generating a complete sequence as they do not know the output sequence length. This restricts us to use a smaller batch size leading to lower GPU utilization and above all, lower throughput. We argue that designing a system with a priori knowledge of the output sequence can mitigate this problem. To this end, we propose S$^{3}$, which predicts the output sequence length, schedules generation queries based on the prediction to increase device resource utilization and throughput, and handle mispredictions. Our proposed method achieves 6.49$\times$ throughput over those systems that assume the worst case for the output sequence length.

arXiv:2306.04655

Modulation Classification Through Deep Learning Using Resolution Transformed Spectrograms

Modulation classification is an essential step of signal processing and has been regularly applied in the field of tele-communication. Since variations of frequency with respect to time remains a vital distinction among radio signals having different modulation formats, these variations can be used for feature extraction by converting 1-D radio signals into frequency domain. In this paper, we propose a scheme for Automatic Modulation Classification (AMC) using modern architectures of Convolutional Neural Networks (CNN), through generating spectrum images of eleven different modulation types. Additionally, we perform resolution transformation of spectrograms that results up to 99.61% of computational load reduction and 8x faster conversion from the received I/Q data. This proposed AMC is implemented on CPU and GPU, to recognize digital as well as analogue signal modulation schemes on signals. The performance is evaluated on existing CNN models including SqueezeNet, Resnet-50, InceptionResnet-V2, Inception-V3, VGG-16 and Densenet-201. Best results of 91.2% are achieved in presence of AWGN and other noise impairments in the signals, stating that the transformed spectrogram-based AMC has good classification accuracy as the spectral features are highly discriminant, and CNN based models have capability to extract these high-dimensional features. The spectrograms were created under different SNRs ranging from 5 to 30db with a step size of 5db to observe the experimental results at various SNR levels. The proposed methodology is efficient to be applied in wireless communication networks for real-time applications.

arXiv:2306.04645

Special Session: Approximation and Fault Resiliency of DNN Accelerators

Deep Learning, and in particular, Deep Neural Network (DNN) is nowadays widely used in many scenarios, including safety-critical applications such as autonomous driving. In this context, besides energy efficiency and performance, reliability plays a crucial role since a system failure can jeopardize human life. As with any other device, the reliability of hardware architectures running DNNs has to be evaluated, usually through costly fault injection campaigns. This paper explores the approximation and fault resiliency of DNN accelerators. We propose to use approximate (AxC) arithmetic circuits to agilely emulate errors in hardware without performing fault injection on the DNN. To allow fast evaluation of AxC DNN, we developed an efficient GPU-based simulation framework. Further, we propose a fine-grain analysis of fault resiliency by examining fault propagation and masking in networks

arXiv:2306.05035공식 코드 · Papers with Code

Does Long-Term Series Forecasting Need Complex Attention and Extra Long Inputs?

As Transformer-based models have achieved impressive performance on various time series tasks, Long-Term Series Forecasting (LTSF) tasks have also received extensive attention in recent years. However, due to the inherent computational complexity and long sequences demanding of Transformer-based methods, its application on LTSF tasks still has two major issues that need to be further investigated: 1) Whether the sparse attention mechanism designed by these methods actually reduce the running time on real devices; 2) Whether these models need extra long input sequences to guarantee their performance? The answers given in this paper are negative. Therefore, to better copy with these two issues, we design a lightweight Period-Attention mechanism (Periodformer), which renovates the aggregation of long-term subseries via explicit periodicity and short-term subseries via built-in proximity. Meanwhile, a gating mechanism is embedded into Periodformer to regulate the influence of the attention module on the prediction results. Furthermore, to take full advantage of GPUs for fast hyperparameter optimization (e.g., finding the suitable input length), a Multi-GPU Asynchronous parallel algorithm based on Bayesian Optimization (MABO) is presented. MABO allocates a process to each GPU via a queue mechanism, and then creates multiple trials at a time for asynchronous parallel search, which greatly reduces the search time. Compared with the state-of-the-art methods, the prediction error of Periodformer reduced by 13% and 26% for multivariate and univariate forecasting, respectively. In addition, MABO reduces the average search time by 46% while finding better hyperparameters. As a conclusion, this paper indicates that LTSF may not need complex attention and extra long input sequences. The code has been open sourced on Github.

arXiv:2306.04988공식 코드 · Papers with Code

StreetSurf: Extending Multi-view Implicit Surface Reconstruction to Street Views

We present a novel multi-view implicit surface reconstruction technique, termed StreetSurf, that is readily applicable to street view images in widely-used autonomous driving datasets, such as Waymo-perception sequences, without necessarily requiring LiDAR data. As neural rendering research expands rapidly, its integration into street views has started to draw interests. Existing approaches on street views either mainly focus on novel view synthesis with little exploration of the scene geometry, or rely heavily on dense LiDAR data when investigating reconstruction. Neither of them investigates multi-view implicit surface reconstruction, especially under settings without LiDAR data. Our method extends prior object-centric neural surface reconstruction techniques to address the unique challenges posed by the unbounded street views that are captured with non-object-centric, long and narrow camera trajectories. We delimit the unbounded space into three parts, close-range, distant-view and sky, with aligned cuboid boundaries, and adapt cuboid/hyper-cuboid hash-grids along with road-surface initialization scheme for finer and disentangled representation. To further address the geometric errors arising from textureless regions and insufficient viewing angles, we adopt geometric priors that are estimated using general purpose monocular models. Coupled with our implementation of efficient and fine-grained multi-stage ray marching strategy, we achieve state of the art reconstruction quality in both geometry and appearance within only one to two hours of training time with a single RTX3090 GPU for each street view sequence. Furthermore, we demonstrate that the reconstructed implicit surfaces have rich potential for various downstream tasks, including ray tracing and LiDAR simulation.

arXiv:2306.04039

Revisiting Neural Retrieval on Accelerators

Retrieval finds a small number of relevant candidates from a large corpus for information retrieval and recommendation applications. A key component of retrieval is to model (user, item) similarity, which is commonly represented as the dot product of two learned embeddings. This formulation permits efficient inference, commonly known as Maximum Inner Product Search (MIPS). Despite its popularity, dot products cannot capture complex user-item interactions, which are multifaceted and likely high rank. We hence examine non-dot-product retrieval settings on accelerators, and propose \textit{mixture of logits} (MoL), which models (user, item) similarity as an adaptive composition of elementary similarity functions. This new formulation is expressive, capable of modeling high rank (user, item) interactions, and further generalizes to the long tail. When combined with a hierarchical retrieval strategy, \textit{h-indexer}, we are able to scale up MoL to 100M corpus on a single GPU with latency comparable to MIPS baselines. On public datasets, our approach leads to uplifts of up to 77.3\% in hit rate (HR). Experiments on a large recommendation surface at Meta showed strong metric gains and reduced popularity bias, validating the proposed approach's performance and improved generalization.

arXiv:2306.03725

Towards Memory-Efficient Training for Extremely Large Output Spaces -- Learning with 500k Labels on a Single Commodity GPU

In classification problems with large output spaces (up to millions of labels), the last layer can require an enormous amount of memory. Using sparse connectivity would drastically reduce the memory requirements, but as we show below, it can result in much diminished predictive performance of the model. Fortunately, we found that this can be mitigated by introducing a penultimate layer of intermediate size. We further demonstrate that one can constrain the connectivity of the sparse layer to be uniform, in the sense that each output neuron will have the exact same number of incoming connections. This allows for efficient implementations of sparse matrix multiplication and connection redistribution on GPU hardware. Via a custom CUDA implementation, we show that the proposed approach can scale to datasets with 670,000 labels on a single commodity GPU with only 4GB memory.

arXiv:2306.03413공식 코드 · Papers with Code

DVIS: Decoupled Video Instance Segmentation Framework

Video instance segmentation (VIS) is a critical task with diverse applications, including autonomous driving and video editing. Existing methods often underperform on complex and long videos in real world, primarily due to two factors. Firstly, offline methods are limited by the tightly-coupled modeling paradigm, which treats all frames equally and disregards the interdependencies between adjacent frames. Consequently, this leads to the introduction of excessive noise during long-term temporal alignment. Secondly, online methods suffer from inadequate utilization of temporal information. To tackle these challenges, we propose a decoupling strategy for VIS by dividing it into three independent sub-tasks: segmentation, tracking, and refinement. The efficacy of the decoupling strategy relies on two crucial elements: 1) attaining precise long-term alignment outcomes via frame-by-frame association during tracking, and 2) the effective utilization of temporal information predicated on the aforementioned accurate alignment outcomes during refinement. We introduce a novel referring tracker and temporal refiner to construct the \textbf{D}ecoupled \textbf{VIS} framework (\textbf{DVIS}). DVIS achieves new SOTA performance in both VIS and VPS, surpassing the current SOTA methods by 7.3 AP and 9.6 VPQ on the OVIS and VIPSeg datasets, which are the most challenging and realistic benchmarks. Moreover, thanks to the decoupling strategy, the referring tracker and temporal refiner are super light-weight (only 1.69\% of the segmenter FLOPs), allowing for efficient training and inference on a single GPU with 11G memory. The code is available at \href{https://github.com/zhang-tao-whu/DVIS}{https://github.com/zhang-tao-whu/DVIS}.

arXiv:2306.01811

DVFO: Learning-Based DVFS for Energy-Efficient Edge-Cloud Collaborative Inference

Due to limited resources on edge and different characteristics of deep neural network (DNN) models, it is a big challenge to optimize DNN inference performance in terms of energy consumption and end-to-end latency on edge devices. In addition to the dynamic voltage frequency scaling (DVFS) technique, the edge-cloud architecture provides a collaborative approach for efficient DNN inference. However, current edge-cloud collaborative inference methods have not optimized various compute resources on edge devices. Thus, we propose DVFO, a novel DVFS-enabled edge-cloud collaborative inference framework, which co-optimizes DVFS and offloading parameters via deep reinforcement learning (DRL). Specifically, DVFO automatically co-optimizes 1) the CPU, GPU and memory frequencies of edge devices, and 2) the feature maps to be offloaded to cloud servers. In addition, it leverages a thinking-while-moving concurrent mechanism to accelerate the DRL learning process, and a spatial-channel attention mechanism to extract DNN feature maps of secondary importance for workload offloading. This approach improves inference performance for different DNN models under various edge-cloud network conditions. Extensive evaluations using two datasets and six widely-deployed DNN models on three heterogeneous edge devices show that DVFO significantly reduces the energy consumption by 33% on average, compared to state-of-the-art schemes. Moreover, DVFO achieves up to 28.6%-59.1% end-to-end latency reduction, while maintaining accuracy within 1% loss on average.

arXiv:2306.03078공식 코드 · Papers with Code

SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression

Recent advances in large language model (LLM) pretraining have led to high-quality LLMs with impressive abilities. By compressing such LLMs via quantization to 3-4 bits per parameter, they can fit into memory-limited devices such as laptops and mobile phones, enabling personalized use. However, quantization down to 3-4 bits per parameter usually leads to moderate-to-high accuracy losses, especially for smaller models in the 1-10B parameter range, which are well-suited for edge deployments. To address this accuracy issue, we introduce the Sparse-Quantized Representation (SpQR), a new compressed format and quantization technique which enables for the first time near-lossless compression of LLMs across model scales, while reaching similar compression levels to previous methods. SpQR works by identifying and isolating outlier weights, which cause particularly-large quantization errors, and storing them in higher precision, while compressing all other weights to 3-4 bits, and achieves relative accuracy losses of less than 1% in perplexity for highly-accurate LLaMA and Falcon LLMs. This makes it possible to run 33B parameter LLM on a single 24 GB consumer GPU without any performance degradation at 15% speedup thus making powerful LLMs available to consumer without any downsides. SpQR comes with efficient algorithms for both encoding weights into its format, as well as decoding them efficiently at runtime. Specifically, we provide an efficient GPU inference algorithm for SpQR which yields faster inference than 16-bit baselines at similar accuracy, while enabling memory compression gains of more than 4x.

arXiv:2306.00283

Autism Disease Detection Using Transfer Learning Techniques: Performance Comparison Between Central Processing Unit vs Graphics Processing Unit Functions for Neural Networks

Neural network approaches are machine learning methods that are widely used in various domains, such as healthcare and cybersecurity. Neural networks are especially renowned for their ability to deal with image datasets. During the training process with images, various fundamental mathematical operations are performed in the neural network. These operations include several algebraic and mathematical functions, such as derivatives, convolutions, and matrix inversions and transpositions. Such operations demand higher processing power than what is typically required for regular computer usage. Since CPUs are built with serial processing, they are not appropriate for handling large image datasets. On the other hand, GPUs have parallel processing capabilities and can provide higher speed. This paper utilizes advanced neural network techniques, such as VGG16, Resnet50, Densenet, Inceptionv3, Xception, Mobilenet, XGBOOST VGG16, and our proposed models, to compare CPU and GPU resources. We implemented a system for classifying Autism disease using face images of autistic and non-autistic children to compare performance during testing. We used evaluation matrices such as Accuracy, F1 score, Precision, Recall, and Execution time. It was observed that GPU outperformed CPU in all tests conducted. Moreover, the performance of the neural network models in terms of accuracy increased on GPU compared to CPU.

arXiv:2306.00272

Accelerated Fingerprint Enhancement: A GPU-Optimized Mixed Architecture Approach

This document presents a preliminary approach to latent fingerprint enhancement, fundamentally designed around a mixed Unet architecture. It combines the capabilities of the Resnet-101 network and Unet encoder, aiming to form a potentially powerful composite. This combination, enhanced with attention mechanisms and forward skip connections, is intended to optimize the enhancement of ridge and minutiae features in fingerprints. One innovative element of this approach includes a novel Fingerprint Enhancement Gabor layer, specifically designed for GPU computations. This illustrates how modern computational resources might be harnessed to expedite enhancement. Given its potential functionality as either a CNN or Transformer layer, this Gabor layer could offer improved agility and processing speed to the system. However, it is important to note that this approach is still in the early stages of development and has not yet been fully validated through rigorous experiments. As such, it may require additional time and testing to establish its robustness and usability in the field of latent fingerprint enhancement. This includes improvements in processing speed, enhancement adaptability with distinct latent fingerprint types, and full validation in experimental approaches such as open-set (identification 1:N) and open-set validation, fingerprint quality evaluation, among others.

arXiv:2306.00637공식 코드 · Papers with Code

Wuerstchen: An Efficient Architecture for Large-Scale Text-to-Image Diffusion Models

We introduce W\"urstchen, a novel architecture for text-to-image synthesis that combines competitive performance with unprecedented cost-effectiveness for large-scale text-to-image diffusion models. A key contribution of our work is to develop a latent diffusion technique in which we learn a detailed but extremely compact semantic image representation used to guide the diffusion process. This highly compressed representation of an image provides much more detailed guidance compared to latent representations of language and this significantly reduces the computational requirements to achieve state-of-the-art results. Our approach also improves the quality of text-conditioned image generation based on our user preference study. The training requirements of our approach consists of 24,602 A100-GPU hours - compared to Stable Diffusion 2.1's 200,000 GPU hours. Our approach also requires less training data to achieve these results. Furthermore, our compact latent representations allows us to perform inference over twice as fast, slashing the usual costs and carbon footprint of a state-of-the-art (SOTA) diffusion model significantly, without compromising the end performance. In a broader comparison against SOTA models our approach is substantially more efficient and compares favorably in terms of image quality. We believe that this work motivates more emphasis on the prioritization of both performance and computational accessibility.

arXiv:2305.19911공식 코드 · Papers with Code

Neuron to Graph: Interpreting Language Model Neurons at Scale

Advances in Large Language Models (LLMs) have led to remarkable capabilities, yet their inner mechanisms remain largely unknown. To understand these models, we need to unravel the functions of individual neurons and their contribution to the network. This paper introduces a novel automated approach designed to scale interpretability techniques across a vast array of neurons within LLMs, to make them more interpretable and ultimately safe. Conventional methods require examination of examples with strong neuron activation and manual identification of patterns to decipher the concepts a neuron responds to. We propose Neuron to Graph (N2G), an innovative tool that automatically extracts a neuron's behaviour from the dataset it was trained on and translates it into an interpretable graph. N2G uses truncation and saliency methods to emphasise only the most pertinent tokens to a neuron while enriching dataset examples with diverse samples to better encompass the full spectrum of neuron behaviour. These graphs can be visualised to aid researchers' manual interpretation, and can generate token activations on text for automatic validation by comparison with the neuron's ground truth activations, which we use to show that the model is better at predicting neuron activation than two baseline methods. We also demonstrate how the generated graph representations can be flexibly used to facilitate further automation of interpretability research, by searching for neurons with particular properties, or programmatically comparing neurons to each other to identify similar neurons. Our method easily scales to build graph representations for all neurons in a 6-layer Transformer model using a single Tesla T4 GPU, allowing for wide usability. We release the code and instructions for use at https://github.com/alexjfoote/Neuron2Graph.

arXiv:2305.19982

Adam Accumulation to Reduce Memory Footprints of both Activations and Gradients for Large-scale DNN Training

Running out of GPU memory has become a main bottleneck for large-scale DNN training. How to reduce the memory footprint during training has received intensive research attention. We find that previous gradient accumulation reduces activation memory but fails to be compatible with gradient memory reduction due to a contradiction between preserving gradients and releasing gradients. To address this issue, we propose a novel optimizer accumulation method for Adam, named Adam Accumulation (AdamA), which enables reducing both activation and gradient memory. Specifically, AdamA directly integrates gradients into optimizer states and accumulates optimizer states over micro-batches, so that gradients can be released immediately after use. We mathematically and experimentally demonstrate AdamA yields the same convergence properties as Adam. Evaluated on transformer-based models, AdamA achieves up to 23% memory reduction compared to gradient accumulation with less than 2% degradation in training throughput. Notably, AdamA can work together with memory reduction methods for optimizer states to fit 1.26x~3.14x larger models over PyTorch and DeepSpeed baseline on GPUs with different memory capacities.

arXiv:2305.18483

Bringing regularized optimal transport to lightspeed: a splitting method adapted for GPUs

We present an efficient algorithm for regularized optimal transport. In contrast to previous methods, we use the Douglas-Rachford splitting technique to develop an efficient solver that can handle a broad class of regularizers. The algorithm has strong global convergence guarantees, low per-iteration cost, and can exploit GPU parallelization, making it considerably faster than the state-of-the-art for many problems. We illustrate its competitiveness in several applications, including domain adaptation and learning of generative models.

arXiv:2305.18513

SlimFit: Memory-Efficient Fine-Tuning of Transformer-based Models Using Training Dynamics

Transformer-based models, such as BERT and ViT, have achieved state-of-the-art results across different natural language processing (NLP) and computer vision (CV) tasks. However, these models are extremely memory intensive during their fine-tuning process, making them difficult to deploy on GPUs with limited memory resources. To address this issue, we introduce a new tool called SlimFit that reduces the memory requirements of these models by dynamically analyzing their training dynamics and freezing less-contributory layers during fine-tuning. The layers to freeze are chosen using a runtime inter-layer scheduling algorithm. SlimFit adopts quantization and pruning for particular layers to balance the load of dynamic activations and to minimize the memory footprint of static activations, where static activations refer to those that cannot be discarded regardless of freezing. This allows SlimFit to freeze up to 95% of layers and reduce the overall on-device GPU memory usage of transformer-based models such as ViT and BERT by an average of 2.2x, across different NLP and CV benchmarks/datasets such as GLUE, SQuAD 2.0, CIFAR-10, CIFAR-100 and ImageNet with an average degradation of 0.2% in accuracy. For such NLP and CV tasks, SlimFit can reduce up to 3.1x the total on-device memory usage with an accuracy degradation of only up to 0.4%. As a result, while fine-tuning of ViT on ImageNet and BERT on SQuAD 2.0 with a batch size of 128 requires 3 and 2 32GB GPUs respectively, SlimFit enables their fine-tuning on a single 32GB GPU without any significant accuracy degradation.

arXiv:2305.18356

RT-kNNS Unbound: Using RT Cores to Accelerate Unrestricted Neighbor Search

The problem of identifying the k-Nearest Neighbors (kNNS) of a point has proven to be very useful both as a standalone application and as a subroutine in larger applications. Given its far-reaching applicability in areas such as machine learning and point clouds, extensive research has gone into leveraging GPU acceleration to solve this problem. Recent work has shown that using Ray Tracing cores in recent GPUs to accelerate kNNS is much more efficient compared to traditional acceleration using shader cores. However, the existing translation of kNNS to a ray tracing problem imposes a constraint on the search space for neighbors. Due to this, we can only use RT cores to accelerate fixed-radius kNNS, which requires the user to set a search radius a priori and hence can miss neighbors. In this work, we propose TrueKNN, the first unbounded RT-accelerated neighbor search. TrueKNN adopts an iterative approach where we incrementally grow the search space until all points have found their k neighbors. We show that our approach is orders of magnitude faster than existing approaches and can even be used to accelerate fixed-radius neighbor searches.

arXiv:2305.18691공식 코드 · Papers with Code

Edge-MoE: Memory-Efficient Multi-Task Vision Transformer Architecture with Task-level Sparsity via Mixture-of-Experts

Computer vision researchers are embracing two promising paradigms: Vision Transformers (ViTs) and Multi-task Learning (MTL), which both show great performance but are computation-intensive, given the quadratic complexity of self-attention in ViT and the need to activate an entire large MTL model for one task. M$^3$ViT is the latest multi-task ViT model that introduces mixture-of-experts (MoE), where only a small portion of subnetworks ("experts") are sparsely and dynamically activated based on the current task. M$^3$ViT achieves better accuracy and over 80% computation reduction but leaves challenges for efficient deployment on FPGA. Our work, dubbed Edge-MoE, solves the challenges to introduce the first end-to-end FPGA accelerator for multi-task ViT with a collection of architectural innovations, including (1) a novel reordering mechanism for self-attention, which requires only constant bandwidth regardless of the target parallelism; (2) a fast single-pass softmax approximation; (3) an accurate and low-cost GELU approximation; (4) a unified and flexible computing unit that is shared by almost all computational layers to maximally reduce resource usage; and (5) uniquely for M$^3$ViT, a novel patch reordering method to eliminate memory access overhead. Edge-MoE achieves 2.24x and 4.90x better energy efficiency comparing with GPU and CPU, respectively. A real-time video demonstration is available online, along with our open-source code written using High-Level Synthesis.