Large-scale language models (LLMs) have demonstrated impressive performance, but their deployment presents challenges due to their significant memory usage. This issue can be alleviated through quantization. In this paper, we identify that the challenge in quantizing activations in LLMs arises from varying ranges across channels, rather than solely the presence of outliers. To address this challenge, we introduce a quantization method called RPTQ, which utilizes a reorder-based approach. By rearranging the channels and quantizing them in clusters, RPTQ effectively mitigates the impact of range differences between channels. To minimize the overhead of the reorder operation, we fuse it into the layer norm operation and weights in linear layers. In our experiments, RPTQ achieved a significant breakthrough by utilizing 3-bit activation in LLMs for the first time, resulting in a substantial reduction in memory usage. For instance, quantizing OPT-175b can lead to a memory consumption reduction of up to 80%.
Virtual humans have gained considerable attention in numerous industries, e.g., entertainment and e-commerce. As a core technology, synthesizing photorealistic face frames from target speech and facial identity has been actively studied with generative adversarial networks. Despite remarkable results of modern talking-face generation models, they often entail high computational burdens, which limit their efficient deployment. This study aims to develop a lightweight model for speech-driven talking-face synthesis. We build a compact generator by removing the residual blocks and reducing the channel width from Wav2Lip, a popular talking-face generator. We also present a knowledge distillation scheme to stably yet effectively train the small-capacity generator without adversarial learning. We reduce the number of parameters and MACs by 28$\times$ while retaining the performance of the original model. Moreover, to alleviate a severe performance drop when converting the whole generator to INT8 precision, we adopt a selective quantization method that uses FP16 for the quantization-sensitive layers and INT8 for the other layers. Using this mixed precision, we achieve up to a 19$\times$ speedup on edge GPUs without noticeably compromising the generation quality.
The recent detection transformer (DETR) has advanced object detection, but its application on resource-constrained devices requires massive computation and memory resources. Quantization stands out as a solution by representing the network in low-bit parameters and operations. However, there is a significant performance drop when performing low-bit quantized DETR (Q-DETR) with existing quantization methods. We find that the bottlenecks of Q-DETR come from the query information distortion through our empirical analyses. This paper addresses this problem based on a distribution rectification distillation (DRD). We formulate our DRD as a bi-level optimization problem, which can be derived by generalizing the information bottleneck (IB) principle to the learning of Q-DETR. At the inner level, we conduct a distribution alignment for the queries to maximize the self-information entropy. At the upper level, we introduce a new foreground-aware query matching scheme to effectively transfer the teacher information to distillation-desired features to minimize the conditional information entropy. Extensive experimental results show that our method performs much better than prior arts. For example, the 4-bit Q-DETR can theoretically accelerate DETR with ResNet-50 backbone by 6.6x and achieve 39.4% AP, with only 2.6% performance gaps than its real-valued counterpart on the COCO dataset.
Contemporary image rescaling aims at embedding a high-resolution (HR) image into a low-resolution (LR) thumbnail image that contains embedded information for HR image reconstruction. Unlike traditional image super-resolution, this enables high-fidelity HR image restoration faithful to the original one, given the embedded information in the LR thumbnail. However, state-of-the-art image rescaling methods do not optimize the LR image file size for efficient sharing and fall short of real-time performance for ultra-high-resolution (e.g., 6K) image reconstruction. To address these two challenges, we propose a novel framework (HyperThumbnail) for real-time 6K rate-distortion-aware image rescaling. Our framework first embeds an HR image into a JPEG LR thumbnail by an encoder with our proposed quantization prediction module, which minimizes the file size of the embedding LR JPEG thumbnail while maximizing HR reconstruction quality. Then, an efficient frequency-aware decoder reconstructs a high-fidelity HR image from the LR one in real time. Extensive experiments demonstrate that our framework outperforms previous image rescaling baselines in rate-distortion performance and can perform 6K image reconstruction in real time.
Recently, the idea of using FP8 as a number format for neural network training has been floating around the deep learning world. Given that most training is currently conducted with entire networks in FP32, or sometimes FP16 with mixed-precision, the step to having some parts of a network run in FP8 with 8-bit weights is an appealing potential speed-up for the generally costly and time-intensive training procedures in deep learning. A natural question arises regarding what this development means for efficient inference on edge devices. In the efficient inference device world, workloads are frequently executed in INT8. Sometimes going even as low as INT4 when efficiency calls for it. In this whitepaper, we compare the performance for both the FP8 and INT formats for efficient on-device inference. We theoretically show the difference between the INT and FP formats for neural networks and present a plethora of post-training quantization and quantization-aware-training results to show how this theory translates to practice. We also provide a hardware analysis showing that the FP formats are somewhere between 50-180% less efficient in terms of compute in dedicated hardware than the INT format. Based on our research and a read of the research field, we conclude that although the proposed FP8 format could be good for training, the results for inference do not warrant a dedicated implementation of FP8 in favor of INT8 for efficient inference. We show that our results are mostly consistent with previous findings but that important comparisons between the formats have thus far been lacking. Finally, we discuss what happens when FP8-trained networks are converted to INT8 and conclude with a brief discussion on the most efficient way for on-device deployment and an extensive suite of INT8 results for many models.
In this paper, we introduce the range of oBERTa language models, an easy-to-use set of language models which allows Natural Language Processing (NLP) practitioners to obtain between 3.8 and 24.3 times faster models without expertise in model compression. Specifically, oBERTa extends existing work on pruning, knowledge distillation, and quantization and leverages frozen embeddings improves distillation and model initialization to deliver higher accuracy on a broad range of transfer tasks. In generating oBERTa, we explore how the highly optimized RoBERTa differs from the BERT for pruning during pre-training and finetuning. We find it less amenable to compression during fine-tuning. We explore the use of oBERTa on seven representative NLP tasks and find that the improved compression techniques allow a pruned oBERTa model to match the performance of BERTbase and exceed the performance of Prune OFA Large on the SQUAD V1.1 Question Answering dataset, despite being 8x and 2x, respectively faster in inference. We release our code, training regimes, and associated model for broad usage to encourage usage and experimentation
This paper considers the problem of distributed estimation in wireless sensor networks (WSN), which is anticipated to support a wide range of applications such as the environmental monitoring, weather forecasting, and location estimation. To this end, we propose a joint model and data driven distributed estimation method by designing the optimal quantizers and fusion center (FC) based on the Bayesian and minimum mean square error (MMSE) criterions. First, universal mean square error (MSE) lower bound for the quantization-based distributed estimation is derived and adopted as the design metric for the quantizers. Then, the optimality of the mean-fusion operation for the FC with MMSE criterion is proved. Next, by exploiting different levels of the statistic information of the desired parameter and observation noise, a joint model and data driven method is proposed to train parts of the quantizer and FC modules as deep neural networks (DNNs), and two loss functions derived from the MMSE criterion are adopted for the sequential training scheme. Furthermore, we extend the above results to the case with multi-bit quantizers, considering both the parallel and one-hot quantization schemes. Finally, simulation results reveal that the proposed method outperforms the state-of-the-art schemes in typical scenarios.
Neural networks have revolutionized many aspects of society but in the era of huge models with billions of parameters, optimizing and deploying them for commercial applications can require significant computational and financial resources. To address these challenges, we introduce the Tetra-AML toolbox, which automates neural architecture search and hyperparameter optimization via a custom-developed black-box Tensor train Optimization algorithm, TetraOpt. The toolbox also provides model compression through quantization and pruning, augmented by compression using tensor networks. Here, we analyze a unified benchmark for optimizing neural networks in computer vision tasks and show the superior performance of our approach compared to Bayesian optimization on the CIFAR-10 dataset. We also demonstrate the compression of ResNet-18 neural networks, where we use 14.5 times less memory while losing just 3.2% of accuracy. The presented framework is generic, not limited by computer vision problems, supports hardware acceleration (such as with GPUs and TPUs) and can be further extended to quantum hardware and to hybrid quantum machine learning models.
Learning rich data representations from unlabeled data is a key challenge towards applying deep learning algorithms in downstream tasks. Several variants of variational autoencoders (VAEs) have been proposed to learn compact data representations by encoding high-dimensional data in a lower dimensional space. Two main classes of VAEs methods may be distinguished depending on the characteristics of the meta-priors that are enforced in the representation learning step. The first class of methods derives a continuous encoding by assuming a static prior distribution in the latent space. The second class of methods learns instead a discrete latent representation using vector quantization (VQ) along with a codebook. However, both classes of methods suffer from certain challenges, which may lead to suboptimal image reconstruction results. The first class suffers from posterior collapse, whereas the second class suffers from codebook collapse. To address these challenges, we introduce a new VAE variant, termed sparse coding-based VAE with learned ISTA (SC-VAE), which integrates sparse coding within variational autoencoder framework. The proposed method learns sparse data representations that consist of a linear combination of a small number of predetermined orthogonal atoms. The sparse coding problem is solved using a learnable version of the iterative shrinkage thresholding algorithm (ISTA). Experiments on two image datasets demonstrate that our model achieves improved image reconstruction results compared to state-of-the-art methods. Moreover, we demonstrate that the use of learned sparse code vectors allows us to perform downstream tasks like image generation and unsupervised image segmentation through clustering image patches.
Object discovery -- separating objects from the background without manual labels -- is a fundamental open challenge in computer vision. Previous methods struggle to go beyond clustering of low-level cues, whether handcrafted (e.g., color, texture) or learned (e.g., from auto-encoders). In this work, we augment the auto-encoder representation learning framework with two key components: motion-guidance and mid-level feature tokenization. Although both have been separately investigated, we introduce a new transformer decoder showing that their benefits can compound thanks to motion-guided vector quantization. We show that our architecture effectively leverages the synergy between motion and tokenization, improving upon the state of the art on both synthetic and real datasets. Our approach enables the emergence of interpretable object-specific mid-level features, demonstrating the benefits of motion-guidance (no labeling) and quantization (interpretability, memory efficiency).
In this paper, we propose binary sparse convolutional networks called BSC-Net for efficient point cloud analysis. We empirically observe that sparse convolution operation causes larger quantization errors than standard convolution. However, conventional network quantization methods directly binarize the weights and activations in sparse convolution, resulting in performance drop due to the significant quantization loss. On the contrary, we search the optimal subset of convolution operation that activates the sparse convolution at various locations for quantization error alleviation, and the performance gap between real-valued and binary sparse convolutional networks is closed without complexity overhead. Specifically, we first present the shifted sparse convolution that fuses the information in the receptive field for the active sites that match the pre-defined positions. Then we employ the differentiable search strategies to discover the optimal opsitions for active site matching in the shifted sparse convolution, and the quantization errors are significantly alleviated for efficient point cloud analysis. For fair evaluation of the proposed method, we empirically select the recently advances that are beneficial for sparse convolution network binarization to construct a strong baseline. The experimental results on Scan-Net and NYU Depth v2 show that our BSC-Net achieves significant improvement upon our srtong baseline and outperforms the state-of-the-art network binarization methods by a remarkable margin without additional computation overhead for binarizing sparse convolutional networks.
As models continue to grow in size, the development of memory optimization methods (MOMs) has emerged as a solution to address the memory bottleneck encountered when training large models. To comprehensively examine the practical value of various MOMs, we have conducted a thorough analysis of existing literature from a systems perspective. Our analysis has revealed a notable challenge within the research community: the absence of standardized metrics for effectively evaluating the efficacy of MOMs. The scarcity of informative evaluation metrics hinders the ability of researchers and practitioners to compare and benchmark different approaches reliably. Consequently, drawing definitive conclusions and making informed decisions regarding the selection and application of MOMs becomes a challenging endeavor. To address the challenge, this paper summarizes the scenarios in which MOMs prove advantageous for model training. We propose the use of distinct evaluation metrics under different scenarios. By employing these metrics, we evaluate the prevailing MOMs and find that their benefits are not universal. We present insights derived from experiments and discuss the circumstances in which they can be advantageous.
Vision transformer emerges as a potential architecture for vision tasks. However, the intense computation and non-negligible delay hinder its application in the real world. As a widespread model compression technique, existing post-training quantization methods still cause severe performance drops. We find the main reasons lie in (1) the existing calibration metric is inaccurate in measuring the quantization influence for extremely low-bit representation, and (2) the existing quantization paradigm is unfriendly to the power-law distribution of Softmax. Based on these observations, we propose a novel Accurate Post-training Quantization framework for Vision Transformer, namely APQ-ViT. We first present a unified Bottom-elimination Blockwise Calibration scheme to optimize the calibration metric to perceive the overall quantization disturbance in a blockwise manner and prioritize the crucial quantization errors that influence more on the final output. Then, we design a Matthew-effect Preserving Quantization for Softmax to maintain the power-law character and keep the function of the attention mechanism. Comprehensive experiments on large-scale classification and detection datasets demonstrate that our APQ-ViT surpasses the existing post-training quantization methods by convincing margins, especially in lower bit-width settings (e.g., averagely up to 5.17% improvement for classification and 24.43% for detection on W4A4). We also highlight that APQ-ViT enjoys versatility and works well on diverse transformer variants.
Deep neural networks have achieved great success in many data processing applications. However, the high computational complexity and storage cost makes deep learning hard to be used on resource-constrained devices, and it is not environmental-friendly with much power cost. In this paper, we focus on low-rank optimization for efficient deep learning techniques. In the space domain, deep neural networks are compressed by low rank approximation of the network parameters, which directly reduces the storage requirement with a smaller number of network parameters. In the time domain, the network parameters can be trained in a few subspaces, which enables efficient training for fast convergence. The model compression in the spatial domain is summarized into three categories as pre-train, pre-set, and compression-aware methods, respectively. With a series of integrable techniques discussed, such as sparse pruning, quantization, and entropy coding, we can ensemble them in an integration framework with lower computational complexity and storage. Besides of summary of recent technical advances, we have two findings for motivating future works: one is that the effective rank outperforms other sparse measures for network compression. The other is a spatial and temporal balance for tensorized neural networks.
Quantization using a small number of bits shows promise for reducing latency and memory usage in deep neural networks. However, most quantization methods cannot readily handle complicated functions such as exponential and square root, and prior approaches involve complex training processes that must interact with floating-point values. This paper proposes a robust method for the full integer quantization of vision transformer networks without requiring any intermediate floating-point computations. The quantization techniques can be applied in various hardware or software implementations, including processor/memory architectures and FPGAs.
Zero-shot quantization (ZSQ) is promising for compressing and accelerating deep neural networks when the data for training full-precision models are inaccessible. In ZSQ, network quantization is performed using synthetic samples, thus, the performance of quantized models depends heavily on the quality of synthetic samples. Nonetheless, we find that the synthetic samples constructed in existing ZSQ methods can be easily fitted by models. Accordingly, quantized models obtained by these methods suffer from significant performance degradation on hard samples. To address this issue, we propose HArd sample Synthesizing and Training (HAST). Specifically, HAST pays more attention to hard samples when synthesizing samples and makes synthetic samples hard to fit when training quantized models. HAST aligns features extracted by full-precision and quantized models to ensure the similarity between features extracted by these two models. Extensive experiments show that HAST significantly outperforms existing ZSQ methods, achieving performance comparable to models that are quantized with real data.
Post-training quantization (PTQ) is a popular method for compressing deep neural networks (DNNs) without modifying their original architecture or training procedures. Despite its effectiveness and convenience, the reliability of PTQ methods in the presence of some extrem cases such as distribution shift and data noise remains largely unexplored. This paper first investigates this problem on various commonly-used PTQ methods. We aim to answer several research questions related to the influence of calibration set distribution variations, calibration paradigm selection, and data augmentation or sampling strategies on PTQ reliability. A systematic evaluation process is conducted across a wide range of tasks and commonly-used PTQ paradigms. The results show that most existing PTQ methods are not reliable enough in term of the worst-case group performance, highlighting the need for more robust methods. Our findings provide insights for developing PTQ methods that can effectively handle distribution shift scenarios and enable the deployment of quantized DNNs in real-world applications.
Mixed-signal neuromorphic processors provide extremely low-power operation for edge inference workloads, taking advantage of sparse asynchronous computation within Spiking Neural Networks (SNNs). However, deploying robust applications to these devices is complicated by limited controllability over analog hardware parameters, as well as unintended parameter and dynamical variations of analog circuits due to fabrication non-idealities. Here we demonstrate a novel methodology for ofDine training and deployment of spiking neural networks (SNNs) to the mixed-signal neuromorphic processor DYNAP-SE2. The methodology utilizes gradient-based training using a differentiable simulation of the mixed-signal device, coupled with an unsupervised weight quantization method to optimize the network's parameters. Parameter noise injection during training provides robustness to the effects of quantization and device mismatch, making the method a promising candidate for real-world applications under hardware constraints and non-idealities. This work extends Rockpool, an open-source deep-learning library for SNNs, with support for accurate simulation of mixed-signal SNN dynamics. Our approach simplifies the development and deployment process for the neuromorphic community, making mixed-signal neuromorphic processors more accessible to researchers and developers.
In this paper, we introduce a new approach, called Posthoc Interpretation via Quantization (PIQ), for interpreting decisions made by trained classifiers. Our method utilizes vector quantization to transform the representations of a classifier into a discrete, class-specific latent space. The class-specific codebooks act as a bottleneck that forces the interpreter to focus on the parts of the input data deemed relevant by the classifier for making a prediction. Our model formulation also enables learning concepts by incorporating the supervision of pretrained annotation models such as state-of-the-art image segmentation models. We evaluated our method through quantitative and qualitative studies involving black-and-white images, color images, and audio. As a result of these studies we found that PIQ generates interpretations that are more easily understood by participants to our user studies when compared to several other interpretation methods in the literature.
Recently, vision transformers (ViTs) have superseded convolutional neural networks in numerous applications, including classification, detection, and segmentation. However, the high computational requirements of ViTs hinder their widespread implementation. To address this issue, researchers have proposed efficient hybrid transformer architectures that combine convolutional and transformer layers with optimized attention computation of linear complexity. Additionally, post-training quantization has been proposed as a means of mitigating computational demands. For mobile devices, achieving optimal acceleration for ViTs necessitates the strategic integration of quantization techniques and efficient hybrid transformer structures. However, no prior investigation has applied quantization to efficient hybrid transformers. In this paper, we discover that applying existing post-training quantization (PTQ) methods for ViTs to efficient hybrid transformers leads to a drastic accuracy drop, attributed to the four following challenges: (i) highly dynamic ranges, (ii) zero-point overflow, (iii) diverse normalization, and (iv) limited model parameters ($<$5M). To overcome these challenges, we propose a new post-training quantization method, which is the first to quantize efficient hybrid ViTs (MobileViTv1, MobileViTv2, Mobile-Former, EfficientFormerV1, EfficientFormerV2). We achieve a significant improvement of 17.73% for 8-bit and 29.75% for 6-bit on average, respectively, compared with existing PTQ methods (EasyQuant, FQ-ViT, PTQ4ViT, and RepQ-ViT)}. We plan to release our code at https://gitlab.com/ones-ai/q-hyvit.