Post-Training Quantization (PTQ) is pivotal for deploying large language models (LLMs) within resource-limited settings by significantly reducing resource demands. However, existing PTQ strategies underperform at low bit levels < 3 bits due to the significant difference between the quantized and original weights. To enhance the quantization performance at low bit widths, we introduce a Mixed-precision Graph Neural PTQ (MG-PTQ) approach, employing a graph neural network (GNN) module to capture dependencies among weights and adaptively assign quantization bit-widths. Through the information propagation of the GNN module, our method more effectively captures dependencies among target weights, leading to a more accurate assessment of weight importance and optimized allocation of quantization strategies. Extensive experiments on the WikiText2 and C4 datasets demonstrate that our MG-PTQ method outperforms previous state-of-the-art PTQ method GPTQ, setting new benchmarks for quantization performance under low-bit conditions.
MRI imputation aims to synthesize the missing modality from one or more available ones, which is highly desirable since it reduces scanning costs and delivers comprehensive MRI information to enhance clinical diagnosis. In this paper, we propose a unified model, CodeBrain, designed to adapt to various brain MRI imputation scenarios. The core design lies in casting various inter-modality transformations as a full-modality code prediction task. To this end, CodeBrain is trained in two stages: Reconstruction and Code Prediction. First, in the Reconstruction stage, we reconstruct each MRI modality, which is mapped into a shared latent space followed by a scalar quantization. Since such quantization is lossy and the code is low dimensional, another MRI modality belonging to the same subject is randomly selected to generate common features to supplement the code and boost the target reconstruction. In the second stage, we train another encoder by a customized grading loss to predict the full-modality codes from randomly masked MRI samples, supervised by the corresponding quantized codes generated from the first stage. In this way, the inter-modality transformation is achieved by mapping the instance-specific codes in a finite scalar space. We evaluated the proposed CodeBrain model on two public brain MRI datasets (i.e., IXI and BraTS 2023). Extensive experiments demonstrate that our CodeBrain model achieves superior imputation performance compared to four existing methods, establishing a new state of the art for unified brain MRI imputation. Codes will be released.
Diffusion models (DMs) have emerged as promising approaches for sequential recommendation due to their strong ability to model data distributions and generate high-quality items. Existing work typically adds noise to the next item and progressively denoises it guided by the user's interaction sequence, generating items that closely align with user interests. However, we identify two key issues in this paradigm. First, the sequences are often heterogeneous in length and content, exhibiting noise due to stochastic user behaviors. Using such sequences as guidance may hinder DMs from accurately understanding user interests. Second, DMs are prone to data bias and tend to generate only the popular items that dominate the training dataset, thus failing to meet the personalized needs of different users. To address these issues, we propose Distinguished Quantized Guidance for Diffusion-based Sequence Recommendation (DiQDiff), which aims to extract robust guidance to understand user interests and generate distinguished items for personalized user interests within DMs. To extract robust guidance, DiQDiff introduces Semantic Vector Quantization (SVQ) to quantize sequences into semantic vectors (e.g., collaborative signals and category interests) using a codebook, which can enrich the guidance to better understand user interests. To generate distinguished items, DiQDiff personalizes the generation through Contrastive Discrepancy Maximization (CDM), which maximizes the distance between denoising trajectories using contrastive loss to prevent biased generation for different users. Extensive experiments are conducted to compare DiQDiff with multiple baseline models across four widely-used datasets. The superior recommendation performance of DiQDiff against leading approaches demonstrates its effectiveness in sequential recommendation tasks.
The Mamba model, utilizing a structured state-space model (SSM), offers linear time complexity and demonstrates significant potential. Vision Mamba (ViM) extends this framework to vision tasks by incorporating a bidirectional SSM and patch embedding, surpassing Transformer-based models in performance. While model quantization is essential for efficient computing, existing works have focused solely on the original Mamba model and have not been applied to ViM. Additionally, they neglect quantizing the SSM layer, which is central to Mamba and can lead to substantial error propagation by naive quantization due to its inherent structure. In this paper, we focus on the post-training quantization (PTQ) of ViM. We address the issues with three core techniques: 1) a k-scaled token-wise quantization method for linear and convolutional layers, 2) a reparameterization technique to simplify hidden state quantization, and 3) a factor-determining method that reduces computational overhead by integrating operations. Through these methods, the error caused by PTQ can be mitigated. Experimental results on ImageNet-1k demonstrate only a 0.8-1.2\% accuracy degradation due to PTQ, highlighting the effectiveness of our approach.
Quantizing deep neural networks ,reducing the precision (bit-width) of their computations, can remarkably decrease memory usage and accelerate processing, making these models more suitable for large-scale medical imaging applications with limited computational resources. However, many existing methods studied "fake quantization", which simulates lower precision operations during inference, but does not actually reduce model size or improve real-world inference speed. Moreover, the potential of deploying real 3D low-bit quantization on modern GPUs is still unexplored. In this study, we introduce a real post-training quantization (PTQ) framework that successfully implements true 8-bit quantization on state-of-the-art (SOTA) 3D medical segmentation models, i.e., U-Net, SegResNet, SwinUNETR, nnU-Net, UNesT, TransUNet, ST-UNet,and VISTA3D. Our approach involves two main steps. First, we use TensorRT to perform fake quantization for both weights and activations with unlabeled calibration dataset. Second, we convert this fake quantization into real quantization via TensorRT engine on real GPUs, resulting in real-world reductions in model size and inference latency. Extensive experiments demonstrate that our framework effectively performs 8-bit quantization on GPUs without sacrificing model performance. This advancement enables the deployment of efficient deep learning models in medical imaging applications where computational resources are constrained. The code and models have been released, including U-Net, TransUNet pretrained on the BTCV dataset for abdominal (13-label) segmentation, UNesT pretrained on the Whole Brain Dataset for whole brain (133-label) segmentation, and nnU-Net, SegResNet, SwinUNETR and VISTA3D pretrained on TotalSegmentator V2 for full body (104-label) segmentation. https://github.com/hrlblab/PTQ.
Deploying Large Language Models (LLMs) on edge devices is increasingly important, as it eliminates reliance on network connections, reduces expensive API calls, and enhances user privacy. However, on-device deployment is challenging due to the limited computational resources of edge devices. In particular, the key bottleneck stems from memory bandwidth constraints related to weight loading. Weight-only quantization effectively reduces memory access, yet often induces significant accuracy degradation. Recent efforts to incorporate sub-branches have shown promise for mitigating quantization errors, but these methods either lack robust optimization strategies or rely on suboptimal objectives. To address these gaps, we propose FeedBack Quantization (FBQuant), a novel approach inspired by negative feedback mechanisms in automatic control. FBQuant inherently ensures that the reconstructed weights remain bounded by the quantization process, thereby reducing the risk of overfitting. To further offset the additional latency introduced by sub-branches, we develop an efficient CUDA kernel that decreases 60\% of extra inference time. Comprehensive experiments demonstrate the efficiency and effectiveness of FBQuant across various LLMs. Notably, for 3-bit Llama2-7B, FBQuant improves zero-shot accuracy by 1.2\%.
The growing computational demands of training large language models (LLMs) necessitate more efficient methods. Quantized training presents a promising solution by enabling low-bit arithmetic operations to reduce these costs. While FP8 precision has demonstrated feasibility, leveraging FP4 remains a challenge due to significant quantization errors and limited representational capacity. This work introduces the first FP4 training framework for LLMs, addressing these challenges with two key innovations: a differentiable quantization estimator for precise weight updates and an outlier clamping and compensation strategy to prevent activation collapse. To ensure stability, the framework integrates a mixed-precision training scheme and vector-wise quantization. Experimental results demonstrate that our FP4 framework achieves accuracy comparable to BF16 and FP8, with minimal degradation, scaling effectively to 13B-parameter LLMs trained on up to 100B tokens. With the emergence of next-generation hardware supporting FP4, our framework sets a foundation for efficient ultra-low precision training.
Key-Value (KV) cache facilitates efficient large language models (LLMs) inference by avoiding recomputation of past KVs. As the batch size and context length increase, the oversized KV caches become a significant memory bottleneck, highlighting the need for efficient compression. Existing KV quantization rely on fine-grained quantization or the retention of a significant portion of high bit-widths caches, both of which compromise compression ratio and often fail to maintain robustness at extremely low average bit-widths. In this work, we explore the potential of rotation technique for 2-bit KV quantization and propose RotateKV, which achieves accurate and robust performance through the following innovations: (i) Outlier-Aware Rotation, which utilizes channel-reordering to adapt the rotations to varying channel-wise outlier distributions without sacrificing the computational efficiency of the fast Walsh-Hadamard transform (FWHT); (ii) Pre-RoPE Grouped-Head Rotation, which mitigates the impact of rotary position embedding (RoPE) on proposed outlier-aware rotation and further smooths outliers across heads; (iii) Attention-Sink-Aware Quantization, which leverages the massive activations to precisely identify and protect attention sinks. RotateKV achieves less than 0.3 perplexity (PPL) degradation with 2-bit quantization on WikiText-2 using LLaMA-2-13B, maintains strong CoT reasoning and long-context capabilities, with less than 1.7\% degradation on GSM8K, outperforming existing methods even at lower average bit-widths. RotateKV also showcases a 3.97x reduction in peak memory usage, supports 5.75x larger batch sizes, and achieves a 2.32x speedup in decoding stage.
This paper introduces EdgeMLOps, a framework leveraging Cumulocity IoT and thin-edge.io for deploying and managing machine learning models on resource-constrained edge devices. We address the challenges of model optimization, deployment, and lifecycle management in edge environments. The framework's efficacy is demonstrated through a visual quality inspection (VQI) use case where images of assets are processed on edge devices, enabling real-time condition updates within an asset management system. Furthermore, we evaluate the performance benefits of different quantization methods, specifically static and dynamic signed-int8, on a Raspberry Pi 4, demonstrating significant inference time reductions compared to FP32 precision. Our results highlight the potential of EdgeMLOps to enable efficient and scalable AI deployments at the edge for industrial applications.
We solve the global asymptotic stability problem of an unstable reaction-diffusion Partial Differential Equation (PDE) subject to input delay and state quantization developing a switched predictor-feedback law. To deal with the input delay, we reformulate the problem as an actuated transport PDE coupled with the original reaction-diffusion PDE. Then, we design a quantized predictor-based feedback mechanism that employs a dynamic switching strategy to adjust the quantization range and error over time. The stability of the closed-loop system is proven properly combining backstepping with a small-gain approach and input-to-state stability techniques, for deriving estimates on solutions, despite the quantization effect and the system's instability. We also extend this result to the input quantization case.
Dual-functional radar-communication (DFRC) signal design has received much attention lately. We consider the scenario of one-bit massive multi-input multi-output (MIMO) wherein one-bit DACs are employed for the sake of saving hardware costs. Specifically, a spatial Sigma-Delta $(\Sigma\Delta)$ modulation scheme is proposed for one-bit MIMO-DFRC waveform design. Unlike the existing approaches which require large-scale binary optimization, the proposed scheme performs $\Sigma\Delta$ modulation on a continuous-valued DFRC signal. The subsequent waveform design is formulated as a constrained least square problem, which can be efficiently solved. Moreover, we leverage quantization noise for radar probing purposes, rather than treating it as unwanted noise. Numerical results demonstrate that the proposed scheme performs well in both radar probing and downlink precoding.
Over-the-air (OTA) computation has emerged as a promising technique for efficiently aggregating data from massive numbers of wireless devices. OTA computations can be performed by analog or digital communications. Analog OTA systems are often constrained by limited function adaptability and their reliance on analog amplitude modulation. On the other hand, digital OTA systems may face limitations such as high computational complexity and limited adaptability to varying network configurations. To address these challenges, this paper proposes a novel digital OTA computation system with a channel-aware constellation design for demodulation mappers. The proposed system dynamically adjusts the constellation based on the channel conditions of participating nodes, enabling reliable computation of various functions. By incorporating channel randomness into the constellation design, the system prevent overlap of constellation points, reduces computational complexity, and mitigates excessive transmit power consumption under poor channel conditions. Numerical results demonstrate that the system achieves reliable NMSE performance across a range of scenarios, offering valuable insights into the choice of signal processing methods and weighting strategies under varying computation point configurations, node counts, and quantization levels. This work advances the state of digital OTA computation by addressing critical challenges in scalability, transmit power consumption, and function adaptability.
Vision-language models (VLMs) show remarkable performance in multimodal tasks. However, excessively long multimodal inputs lead to oversized Key-Value (KV) caches, resulting in significant memory consumption and I/O bottlenecks. Previous KV quantization methods for Large Language Models (LLMs) may alleviate these issues but overlook the attention saliency differences of multimodal tokens, resulting in suboptimal performance. In this paper, we investigate the attention-aware token saliency patterns in VLM and propose AKVQ-VL. AKVQ-VL leverages the proposed Text-Salient Attention (TSA) and Pivot-Token-Salient Attention (PSA) patterns to adaptively allocate bit budgets. Moreover, achieving extremely low-bit quantization requires effectively addressing outliers in KV tensors. AKVQ-VL utilizes the Walsh-Hadamard transform (WHT) to construct outlier-free KV caches, thereby reducing quantization difficulty. Evaluations of 2-bit quantization on 12 long-context and multimodal tasks demonstrate that AKVQ-VL maintains or even improves accuracy, outperforming LLM-oriented methods. AKVQ-VL can reduce peak memory usage by 2.13x, support up to 3.25x larger batch sizes and 2.46x throughput.
The emergence of Large Language Models (LLMs) such as GPT-4, LLaMA, and BERT has transformed artificial intelligence, enabling advanced capabilities across diverse applications. While parameter-efficient fine-tuning (PEFT) techniques like LoRA offer computationally efficient adaptations of these models, their practical deployment often assumes centralized data and training environments. However, real-world scenarios frequently involve distributed, privacy-sensitive datasets that require decentralized solutions. Federated learning (FL) addresses data privacy by coordinating model updates across clients, but it is typically based on centralized aggregation through a parameter server, which can introduce bottlenecks and communication constraints. Decentralized learning, in contrast, eliminates this dependency by enabling direct collaboration between clients, improving scalability and efficiency in distributed environments. Despite its advantages, decentralized LLM fine-tuning remains underexplored. In this work, we propose \texttt{Dec-LoRA}, an algorithm for decentralized fine-tuning of LLMs based on low-rank adaptation (LoRA). Through extensive experiments on BERT and LLaMA-2 models, we evaluate \texttt{Dec-LoRA}'s performance in handling data heterogeneity and quantization constraints, enabling scalable, privacy-preserving LLM fine-tuning in decentralized settings.
Effective image tokenization is crucial for both multi-modal understanding and generation tasks due to the necessity of the alignment with discrete text data. To this end, existing approaches utilize vector quantization (VQ) to project pixels onto a discrete codebook and reconstruct images from the discrete representation. However, compared with the continuous latent space, the limited discrete codebook space significantly restrict the representational ability of these image tokenizers. In this paper, we propose GaussianToken: An Effective Image Tokenizer with 2D Gaussian Splatting as a solution. We first represent the encoded samples as multiple flexible featured 2D Gaussians characterized by positions, rotation angles, scaling factors, and feature coefficients. We adopt the standard quantization for the Gaussian features and then concatenate the quantization results with the other intrinsic Gaussian parameters before the corresponding splatting operation and the subsequent decoding module. In general, GaussianToken integrates the local influence of 2D Gaussian distribution into the discrete space and thus enhances the representation capability of the image tokenizer. Competitive reconstruction performances on CIFAR, Mini-ImageNet, and ImageNet-1K demonstrate the effectiveness of our framework. Our code is available at: https://github.com/ChrisDong-THU/GaussianToken.
Diffusion models have gained significant popularity in image generation tasks. However, generating high-quality content remains notably slow because it requires running model inference over many time steps. To accelerate these models, we propose to aggressively quantize both weights and activations, while simultaneously promoting significant activation sparsity. We further observe that the stated sparsity pattern varies among different channels and evolves across time steps. To support this quantization and sparsity scheme, we present a novel diffusion model accelerator featuring a heterogeneous mixed-precision dense-sparse architecture, channel-last address mapping, and a time-step-aware sparsity detector for efficient handling of the sparsity pattern. Our 4-bit quantization technique demonstrates superior generation quality compared to existing 4-bit methods. Our custom accelerator achieves 6.91x speed-up and 51.5% energy reduction compared to traditional dense accelerators.
Resource-constrained edge deployments demand AI solutions that balance high performance with stringent compute, memory, and energy limitations. In this survey, we present a comprehensive overview of the primary strategies for accelerating deep learning models under such constraints. First, we examine model compression techniques-pruning, quantization, tensor decomposition, and knowledge distillation-that streamline large models into smaller, faster, and more efficient variants. Next, we explore Neural Architecture Search (NAS), a class of automated methods that discover architectures inherently optimized for particular tasks and hardware budgets. We then discuss compiler and deployment frameworks, such as TVM, TensorRT, and OpenVINO, which provide hardware-tailored optimizations at inference time. By integrating these three pillars into unified pipelines, practitioners can achieve multi-objective goals, including latency reduction, memory savings, and energy efficiency-all while maintaining competitive accuracy. We also highlight emerging frontiers in hierarchical NAS, neurosymbolic approaches, and advanced distillation tailored to large language models, underscoring open challenges like pre-training pruning for massive networks. Our survey offers practical insights, identifies current research gaps, and outlines promising directions for building scalable, platform-independent frameworks to accelerate deep learning models at the edge.
We present an end-to-end workflow for superconducting qubit readout that embeds co-designed Neural Networks (NNs) into the Quantum Instrumentation Control Kit (QICK). Capitalizing on the custom firmware and software of the QICK platform, which is built on Xilinx RFSoC FPGAs, we aim to leverage machine learning (ML) to address critical challenges in qubit readout accuracy and scalability. The workflow utilizes the hls4ml package and employs quantization-aware training to translate ML models into hardware-efficient FPGA implementations via user-friendly Python APIs. We experimentally demonstrate the design, optimization, and integration of an ML algorithm for single transmon qubit readout, achieving 96% single-shot fidelity with a latency of 32ns and less than 16% FPGA look-up table resource utilization. Our results offer the community an accessible workflow to advance ML-driven readout and adaptive control in quantum information processing applications.
The success of deep learning has sparked significant interest in designing computer hardware optimized for the high computational demands of neural network inference. As further miniaturization of digital CMOS processors becomes increasingly challenging, alternative computing paradigms, such as analog computing, are gaining consideration. Particularly for compute-intensive tasks such as matrix multiplication, analog computing presents a promising alternative due to its potential for significantly higher energy efficiency compared to conventional digital technology. However, analog computations are inherently noisy, which makes it challenging to maintain high accuracy on deep neural networks. This work investigates the effectiveness of training neural networks with quantization to increase the robustness against noise. Experimental results across various network architectures show that quantization-aware training with constant scaling factors enhances robustness. We compare these methods with noisy training, which incorporates a noise injection during training that mimics the noise encountered during inference. While both two methods increase tolerance against noise, noisy training emerges as the superior approach for achieving robust neural network performance, especially in complex neural architectures.
Post-training quantization (PTQ) has emerged as a widely adopted technique for compressing and accelerating Large Language Models (LLMs). The major challenge in LLM quantization is that uneven and heavy-tailed data distributions can expand the quantization range, thereby reducing bit precision for most values. Recent methods attempt to eliminate outliers and balance inter-channel differences by employing linear transformations; however, they remain heuristic and are often overlook optimizing the data distribution across the entire quantization space.In this paper, we introduce Quantization Space Utilization Rate (QSUR), a novel metric that effectively assesses the quantizability of transformed data by measuring the space utilization of the data in the quantization space. We complement QSUR with mathematical derivations that examine the effects and limitations of various transformations, guiding our development of Orthogonal and Scaling Transformation-based Quantization (OSTQuant). OSQuant employs a learnable equivalent transformation, consisting of an orthogonal transformation and a scaling transformation, to optimize the distributions of weights and activations across the entire quantization space. Futhermore, we propose the KL-Top loss function, designed to mitigate noise during optimization while retaining richer semantic information within the limited calibration data imposed by PTQ. OSTQuant outperforms existing work on various LLMs and benchmarks. In the W4-only setting, it retains 99.5\% of the floating-point accuracy. In the more challenging W4A4KV4 configuration, OSTQuant reduces the performance gap by 32\% on the LLaMA-3-8B model compared to state-of-the-art methods. \href{https://github.com/BrotherHappy/OSTQuant}{https://github.com/BrotherHappy/OSTQuant}.