As communication systems advance towards the future 6G era, the incorporation of large-scale antenna arrays in base stations (BSs) presents challenges such as increased hardware costs and energy consumption. To address these issues, the use of one-bit analog-to-digital converters (ADCs)/digital-to-analog converters (DACs) has gained significant attentions. This paper focuses on one-bit multiple-input multiple-output (MIMO) detection in an uplink multiuser transmission scenario where the BS employs one-bit ADCs. One-bit quantization retains only the sign information and loses the amplitude information, which poses a unique challenge in the corresponding detection problem. The maximum-likelihood (ML) formulation of one-bit MIMO detection has a challenging likelihood function that hinders the application of many high-performance detectors developed for classic MIMO detection (under high-resolution ADCs). While many approximate methods for the ML detection problem have been studied, it lacks an efficient global algorithm. This paper fills this gap by proposing an efficient branch-and-bound algorithm, which is guaranteed to find the global solution of the one-bit ML MIMO detection problem. Additionally, a new amplitude retrieval (AR) detection approach is developed, incorporating explicit amplitude variables into the problem formulation. The AR approach yields simpler objective functions that enable the development of efficient algorithms offering both global and approximate solutions. The paper also contributes to the computational complexity analysis of both ML and AR detection problems. Extensive simulations are conducted to demonstrate the effectiveness and efficiency of the proposed formulations and algorithms.
Vision Transformer (ViT) has become one of the most prevailing fundamental backbone networks in the computer vision community. Despite the high accuracy, deploying it in real applications raises critical challenges including the high computational cost and inference latency. Recently, the post-training quantization (PTQ) technique has emerged as a promising way to enhance ViT's efficiency. Nevertheless, existing PTQ approaches for ViT suffer from the inflexible quantization on the post-Softmax and post-GELU activations that obey the power-law-like distributions. To address these issues, we propose a novel non-uniform quantizer, dubbed the Adaptive Logarithm AdaLog (AdaLog) quantizer. It optimizes the logarithmic base to accommodate the power-law-like distribution of activations, while simultaneously allowing for hardware-friendly quantization and de-quantization. By employing the bias reparameterization, the AdaLog quantizer is applicable to both the post-Softmax and post-GELU activations. Moreover, we develop an efficient Fast Progressive Combining Search (FPCS) strategy to determine the optimal logarithm base for AdaLog, as well as the scaling factors and zero points for the uniform quantizers. Extensive experimental results on public benchmarks demonstrate the effectiveness of our approach for various ViT-based architectures and vision tasks including classification, object detection, and instance segmentation. Code is available at https://github.com/GoatWu/AdaLog.
This paper introduces LiNR, LinkedIn's large-scale, GPU-based retrieval system. LiNR supports a billion-sized index on GPU models. We discuss our experiences and challenges in creating scalable, differentiable search indexes using TensorFlow and PyTorch at production scale. In LiNR, both items and model weights are integrated into the model binary. Viewing index construction as a form of model training, we describe scaling our system for large indexes, incorporating full scans and efficient filtering. A key focus is on enabling attribute-based pre-filtering for exhaustive GPU searches, addressing the common challenge of post-filtering in KNN searches that often reduces system quality. We further provide multi-embedding retrieval algorithms and strategies for tackling cold start issues in retrieval. Our advancements in supporting larger indexes through quantization are also discussed. We believe LiNR represents one of the industry's first Live-updated model-based retrieval indexes. Applied to out-of-network post recommendations on LinkedIn Feed, LiNR has contributed to a 3% relative increase in professional daily active users. We envisage LiNR as a step towards integrating retrieval and ranking into a single GPU model, simplifying complex infrastructures and enabling end-to-end optimization of the entire differentiable infrastructure through gradient descent.
Network quantization generally converts full-precision weights and/or activations into low-bit fixed-point values in order to accelerate an inference process. Recent approaches to network quantization further discretize the gradients into low-bit fixed-point values, enabling an efficient training. They typically set a quantization interval using a min-max range of the gradients or adjust the interval such that the quantization error for entire gradients is minimized. In this paper, we analyze the quantization error of gradients for the low-bit fixed-point training, and show that lowering the error for large-magnitude gradients boosts the quantization performance significantly. Based on this, we derive an upper bound of quantization error for the large gradients in terms of the quantization interval, and obtain an optimal condition for the interval minimizing the quantization error for large gradients. We also introduce an interval update algorithm that adjusts the quantization interval adaptively to maintain a small quantization error for large gradients. Experimental results demonstrate the effectiveness of our quantization method for various combinations of network architectures and bit-widths on various tasks, including image classification, object detection, and super-resolution.
Most existing Low-light Image Enhancement (LLIE) methods either directly map Low-Light (LL) to Normal-Light (NL) images or use semantic or illumination maps as guides. However, the ill-posed nature of LLIE and the difficulty of semantic retrieval from impaired inputs limit these methods, especially in extremely low-light conditions. To address this issue, we present a new LLIE network via Generative LAtent feature based codebook REtrieval (GLARE), in which the codebook prior is derived from undegraded NL images using a Vector Quantization (VQ) strategy. More importantly, we develop a generative Invertible Latent Normalizing Flow (I-LNF) module to align the LL feature distribution to NL latent representations, guaranteeing the correct code retrieval in the codebook. In addition, a novel Adaptive Feature Transformation (AFT) module, featuring an adjustable function for users and comprising an Adaptive Mix-up Block (AMB) along with a dual-decoder architecture, is devised to further enhance fidelity while preserving the realistic details provided by codebook prior. Extensive experiments confirm the superior performance of GLARE on various benchmark datasets and real-world data. Its effectiveness as a preprocessing tool in low-light object detection tasks further validates GLARE for high-level vision applications. Code is released at https://github.com/LowLevelAI/GLARE.
Class-incremental continual learning is an important area of research, as static deep learning methods fail to adapt to changing tasks and data distributions. In previous works, promising results were achieved using replay and compressed replay techniques. In the field of regular replay, GDumb achieved outstanding results but requires a large amount of memory. This problem can be addressed by compressed replay techniques. The goal of this work is to evaluate compressed replay in the pipeline of GDumb. We propose FETCH, a two-stage compression approach. First, the samples from the continual datastream are encoded by the early layers of a pre-trained neural network. Second, the samples are compressed before being stored in the episodic memory. Following GDumb, the remaining classification head is trained from scratch using only the decompressed samples from the reply memory. We evaluate FETCH in different scenarios and show that this approach can increase accuracy on CIFAR10 and CIFAR100. In our experiments, simple compression methods (e.g., quantization of tensors) outperform deep autoencoders. In the future, FETCH could serve as a baseline for benchmarking compressed replay learning in constrained memory scenarios.
Binary Neural Networks (BNNs) enable efficient deep learning by saving on storage and computational costs. However, as the size of neural networks continues to grow, meeting computational requirements remains a challenge. In this work, we propose a new form of quantization to tile neural network layers with sequences of bits to achieve sub-bit compression of binary-weighted neural networks. The method learns binary vectors (i.e. tiles) to populate each layer of a model via aggregation and reshaping operations. During inference, the method reuses a single tile per layer to represent the full tensor. We employ the approach to both fully-connected and convolutional layers, which make up the breadth of space in most neural architectures. Empirically, the approach achieves near fullprecision performance on a diverse range of architectures (CNNs, Transformers, MLPs) and tasks (classification, segmentation, and time series forecasting) with up to an 8x reduction in size compared to binary-weighted models. We provide two implementations for Tiled Bit Networks: 1) we deploy the model to a microcontroller to assess its feasibility in resource-constrained environments, and 2) a GPU-compatible inference kernel to facilitate the reuse of a single tile per layer in memory.
Transformer models have revolutionized AI tasks, but their large size hinders real-world deployment on resource-constrained and latency-critical edge devices. While binarized Transformers offer a promising solution by significantly reducing model size, existing approaches suffer from algorithm-hardware mismatches with limited co-design exploration, leading to suboptimal performance on edge devices. Hence, we propose a co-design method for efficient end-to-end edge deployment of Transformers from three aspects: algorithm, hardware, and joint optimization. First, we propose BMT, a novel hardware-friendly binarized Transformer with optimized quantization methods and components, and we further enhance its model accuracy by leveraging the weighted ternary weight splitting training technique. Second, we develop a streaming processor mixed binarized Transformer accelerator, namely BAT, which is equipped with specialized units and scheduling pipelines for efficient inference of binarized Transformers. Finally, we co-optimize the algorithm and hardware through a design space exploration approach to achieve a global trade-off between accuracy, latency, and robustness for real-world deployments. Experimental results show our co-design achieves up to 2.14-49.37x throughput gains and 3.72-88.53x better energy efficiency over state-of-the-art Transformer accelerators, enabling efficient end-to-end edge deployment.
Modern recurrent layers are emerging as a promising path toward edge deployment of foundation models, especially in the context of large language models (LLMs). Compressing the whole input sequence in a finite-dimensional representation enables recurrent layers to model long-range dependencies while maintaining a constant inference cost for each token and a fixed memory requirement. However, the practical deployment of LLMs in resource-limited environments often requires further model compression, such as quantization and pruning. While these techniques are well-established for attention-based models, their effects on recurrent layers remain underexplored. In this preliminary work, we focus on post-training quantization for recurrent LLMs and show that Mamba models exhibit the same pattern of outlier channels observed in attention-based LLMs. We show that the reason for the difficulty of quantizing SSMs is caused by activation outliers, similar to those observed in transformer-based LLMs. We report baseline results for post-training quantization of Mamba that do not take into account the activation outliers and suggest first steps for outlier-aware quantization.
Crossbar-based in-memory computing (IMC) has emerged as a promising platform for hardware acceleration of deep neural networks (DNNs). However, the energy and latency of IMC systems are dominated by the large overhead of the peripheral analog-to-digital converters (ADCs). To address such ADC bottleneck, here we propose to implement stochastic processing of array-level partial sums (PS) for efficient IMC. Leveraging the probabilistic switching of spin-orbit torque magnetic tunnel junctions, the proposed PS processing eliminates the costly ADC, achieving significant improvement in energy and area efficiency. To mitigate accuracy loss, we develop PS-quantization-aware training that enables backward propagation across stochastic PS. Furthermore, a novel scheme with an inhomogeneous sampling length of the stochastic conversion is proposed. When running ResNet20 on the CIFAR-10 dataset, our architecture-to-algorithm co-design demonstrates up to 16x, 8x, and 10x improvement in energy, latency, and area, respectively, compared to IMC with standard ADC. Our optimized design configuration using stochastic PS achieved 130x (24x) improvement in Energy-Delay-Product compared to IMC with full precision ADC (sparse low-bit ADC), while maintaining near-software accuracy at various benchmark classification tasks.
Rapid advancements in GPU computational power has outpaced memory capacity and bandwidth growth, creating bottlenecks in Large Language Model (LLM) inference. Post-training quantization is the leading method for addressing memory-related bottlenecks in LLM inference, but it suffers from significant performance degradation below 4-bit precision. This paper addresses these challenges by investigating the pretraining of low-bitwidth models specifically Ternary Language Models (TriLMs) as an alternative to traditional floating-point models (FloatLMs) and their post-training quantized versions (QuantLMs). We present Spectra LLM suite, the first open suite of LLMs spanning multiple bit-widths, including FloatLMs, QuantLMs, and TriLMs, ranging from 99M to 3.9B parameters trained on 300B tokens. Our comprehensive evaluation demonstrates that TriLMs offer superior scaling behavior in terms of model size (in bits). Surprisingly, at scales exceeding one billion parameters, TriLMs consistently outperform their QuantLM and FloatLM counterparts for a given bit size across various benchmarks. Notably, the 3.9B parameter TriLM matches the performance of the FloatLM 3.9B across all benchmarks, despite having fewer bits than FloatLM 830M. Overall, this research provides valuable insights into the feasibility and scalability of low-bitwidth language models, paving the way for the development of more efficient LLMs. To enhance understanding of low-bitwidth models, we are releasing 500+ intermediate checkpoints of the Spectra suite at https://github.com/NolanoOrg/SpectraSuite.
The increasing scale of Transformer models has led to an increase in their pre-training computational requirements. While quantization has proven to be effective after pre-training and during fine-tuning, applying quantization in Transformers during pre-training has remained largely unexplored at scale for language modeling. This study aims to explore the impact of quantization for efficient pre-training of Transformers, with a focus on linear layer components. By systematically applying straightforward linear quantization to weights, activations, gradients, and optimizer states, we assess its effects on model efficiency, stability, and performance during training. By offering a comprehensive recipe of effective quantization strategies to be applied during the pre-training of Transformers, we promote high training efficiency from scratch while retaining language modeling ability. Code is available at https://github.com/chandar-lab/EfficientLLMs.
Vision-Language Large Models (VLMs) recently become primary backbone of AI, due to the impressive performance. However, their expensive computation costs, i.e., throughput and delay, impede potentials in the real-world scenarios. To achieve acceleration for VLMs, most existing methods focus on the model perspective: pruning, distillation, quantization, but completely overlook the data-perspective redundancy. To fill the overlook, this paper pioneers the severity of data redundancy, and designs one plug-and-play Turbo module guided by information degree to prune inefficient tokens from visual or textual data. In pursuit of efficiency-performance trade-offs, information degree takes two crucial factors into consideration: mutual redundancy and semantic value. Concretely, the former evaluates data duplication between sequential tokens; while the latter evaluates each token by its contribution to the overall semantics. As a result, tokens with high information degree carry less redundancy and stronger semantics. For VLMs' calculation, Turbo works as a user-friendly plug-in that sorts data referring to information degree, utilizing only top-level ones to save costs. Its advantages are multifaceted, e.g., being generally compatible to various VLMs across understanding and generation, simple use without re-training and trivial engineering efforts. On multiple VLMs benchmarks, we fully experiment to demonstrate the good acceleration of Turbo, under negligible performance drop.
Recently, the field of Image Coding for Machines (ICM) has garnered heightened interest and significant advances thanks to the rapid progress of learning-based techniques for image compression and analysis. Previous studies often require training separate codecs to support various bitrate levels, machine tasks, and networks, thus lacking both flexibility and practicality. To address these challenges, we propose a rate-distortion-cognition controllable versatile image compression, which method allows the users to adjust the bitrate (i.e., Rate), image reconstruction quality (i.e., Distortion), and machine task accuracy (i.e., Cognition) with a single neural model, achieving ultra-controllability. Specifically, we first introduce a cognition-oriented loss in the primary compression branch to train a codec for diverse machine tasks. This branch attains variable bitrate by regulating quantization degree through the latent code channels. To further enhance the quality of the reconstructed images, we employ an auxiliary branch to supplement residual information with a scalable bitstream. Ultimately, two branches use a `$\beta x + (1 - \beta) y$' interpolation strategy to achieve a balanced cognition-distortion trade-off. Extensive experiments demonstrate that our method yields satisfactory ICM performance and flexible Rate-Distortion-Cognition controlling.
Quantization has become increasingly pivotal in addressing the steadily increasing computational and memory requirements of Deep Neural Networks (DNNs). By reducing the number of bits used to represent weights and activations (typically from 32-bit floating-point to 16-bit or 8-bit integers), quantization reduces the memory footprint, energy consumption, and execution time of DNN models. However, traditional quantization methods typically focus on the inference of DNNs, while the training process still relies on floating-point operations. To date, only one work in the literature has addressed integer-only training for Multi-Layer Perceptron (MLP) architectures. This work introduces NITRO-D, a new framework for training arbitrarily deep integer-only Convolutional Neural Networks (CNNs) that operate entirely in the integer-only domain for both training and inference. NITRO-D is the first framework in the literature enabling the training of integer-only CNNs without the need to introduce a quantization scheme. Specifically, NITRO-D introduces a novel architecture integrating multiple integer local-loss blocks, which include the proposed NITRO Scaling Layer and the NITRO-ReLU activation function. Additionally, it introduces a novel integer-only learning algorithm derived from Local Error Signals (LES), utilizing IntegerSGD, an optimizer specifically designed to operate in an integer-only context. NITRO-D is implemented in an open-source Python library. Extensive experimental evaluations demonstrate its effectiveness across several state-of-the-art image recognition datasets. Results show significant performance improvements from 2.47% to 5.96% for integer-only MLP architectures over the state-of-the-art solution, and the capability of training integer-only CNN architectures with minimal accuracy degradation from -0.15% to -4.22% compared to floating-point LES.
Recently, video diffusion models (VDMs) have garnered significant attention due to their notable advancements in generating coherent and realistic video content. However, processing multiple frame features concurrently, coupled with the considerable model size, results in high latency and extensive memory consumption, hindering their broader application. Post-training quantization (PTQ) is an effective technique to reduce memory footprint and improve computational efficiency. Unlike image diffusion, we observe that the temporal features, which are integrated into all frame features, exhibit pronounced skewness. Furthermore, we investigate significant inter-channel disparities and asymmetries in the activation of video diffusion models, resulting in low coverage of quantization levels by individual channels and increasing the challenge of quantization. To address these issues, we introduce the first PTQ strategy tailored for video diffusion models, dubbed QVD. Specifically, we propose the High Temporal Discriminability Quantization (HTDQ) method, designed for temporal features, which retains the high discriminability of quantized features, providing precise temporal guidance for all video frames. In addition, we present the Scattered Channel Range Integration (SCRI) method which aims to improve the coverage of quantization levels across individual channels. Experimental validations across various models, datasets, and bit-width settings demonstrate the effectiveness of our QVD in terms of diverse metrics. In particular, we achieve near-lossless performance degradation on W8A8, outperforming the current methods by 205.12 in FVD.
With the commercialization of large language models (LLMs), weight-activation quantization has emerged to compress and accelerate LLMs, achieving high throughput while reducing inference costs. However, existing post-training quantization (PTQ) techniques for quantizing weights and activations of LLMs still suffer from non-negligible accuracy drops, especially on massive multitask language understanding. To address this issue, we propose Low-Rank Quantization (LRQ) $-$ a simple yet effective post-training weight quantization method for LLMs that reconstructs the outputs of an intermediate Transformer block by leveraging low-rank weight-scaling matrices, replacing the conventional full weight-scaling matrices that entail as many learnable scales as their associated weights. Thanks to parameter sharing via low-rank structure, LRQ only needs to learn significantly fewer parameters while enabling the individual scaling of weights, thus boosting the generalization capability of quantized LLMs. We show the superiority of LRQ over prior LLM PTQ works under (i) $8$-bit weight and per-tensor activation quantization, (ii) $4$-bit weight and $8$-bit per-token activation quantization, and (iii) low-bit weight-only quantization schemes. Our code is available at \url{https://github.com/onliwad101/FlexRound_LRQ} to inspire LLM researchers and engineers.
Deep Learning Architectures employ heavy computations and bulk of the computational energy is taken up by the convolution operations in the Convolutional Neural Networks. The objective of our proposed work is to reduce the energy consumption and size of CNN for using machine learning techniques in edge computing on ubiquitous computing devices. We propose Systematic Quality Scalable Design Methodology consisting of Quality Scalable Quantization on a higher abstraction level and Quality Scalable Multipliers at lower abstraction level. The first component consists of parameter compression where we approximate representation of values in filters of deep learning models by encoding in 3 bits. A shift and scale based on-chip decoding hardware is proposed which can decode these 3-bit representations to recover approximate filter values. The size of the DNN model is reduced this way and can be sent over a communication channel to be decoded on the edge computing devices. This way power is reduced by limiting data bits by approximation. In the second component we propose a quality scalable multiplier which reduces the number of partial products by converting numbers in canonic sign digit representations and further approximating the number by reducing least significant bits. These quantized CNNs provide almost same ac-curacy as network with original weights with little or no fine-tuning. The hardware for the adaptive multipliers utilize gate clocking for reducing energy consumption during multiplications. The proposed methodology greatly reduces the memory and power requirements of DNN models making it a feasible approach to deploy Deep Learning on edge computing. The experiments done on LeNet and ConvNets show an increase upto 6% of zeros and memory savings upto 82.4919% while keeping the accuracy near the state of the art.
Recently proposed methods for 1-bit and 1.58-bit quantization aware training investigate the performance and behavior of these methods in the context of large language models, finding state-of-the-art performance for models with more than 3B parameters. In this work, we investigate 1.58-bit quantization for small language and vision models ranging from 100K to 48M parameters. We introduce a variant of BitNet b1.58, which allows to rely on the median rather than the mean in the quantization process. Through extensive experiments we investigate the performance of 1.58-bit models obtained through quantization aware training. We further investigate the robustness of 1.58-bit quantization-aware training to changes in the learning rate and regularization through weight decay, finding different patterns for small language and vision models than previously reported for large language models. Our results showcase that 1.58-bit quantization-aware training provides state-of-the-art performance for small language models when doubling hidden layer sizes and reaches or even surpasses state-of-the-art performance for small vision models of identical size. Ultimately, we demonstrate that 1.58-bit quantization-aware training is a viable and promising approach also for training smaller deep learning networks, facilitating deployment of such models in low-resource use-cases and encouraging future research.
The utilization of AIoT technology has become a crucial trend in modern poultry management, offering the potential to optimize farming operations and reduce human workloads. This paper presents a real-time and compact edge-AI enabled detector designed to identify chickens and their healthy statuses using frames captured by a lightweight and intelligent camera equipped with an edge-AI enabled CMOS sensor. To ensure efficient deployment of the proposed compact detector within the memory-constrained edge-AI enabled CMOS sensor, we employ a FCOS-Lite detector leveraging MobileNet as the backbone. To mitigate the issue of reduced accuracy in compact edge-AI detectors without incurring additional inference costs, we propose a gradient weighting loss function as classification loss and introduce CIOU loss function as localization loss. Additionally, we propose a knowledge distillation scheme to transfer valuable information from a large teacher detector to the proposed FCOS-Lite detector, thereby enhancing its performance while preserving a compact model size. Experimental results demonstrate the proposed edge-AI enabled detector achieves commendable performance metrics, including a mean average precision (mAP) of 95.1$\%$ and an F1-score of 94.2$\%$, etc. Notably, the proposed detector can be efficiently deployed and operates at a speed exceeding 20 FPS on the edge-AI enabled CMOS sensor, achieved through int8 quantization. That meets practical demands for automated poultry health monitoring using lightweight intelligent cameras with low power consumption and minimal bandwidth costs.