This work introduces Semantically Masked VQ-GAN (SQ-GAN), a novel approach integrating generative models to optimize image compression for semantic/task-oriented communications. SQ-GAN employs off-the-shelf semantic semantic segmentation and a new specifically developed semantic-conditioned adaptive mask module (SAMM) to selectively encode semantically significant features of the images. SQ-GAN outperforms state-of-the-art image compression schemes such as JPEG2000 and BPG across multiple metrics, including perceptual quality and semantic segmentation accuracy on the post-decoding reconstructed image, at extreme low compression rates expressed in bits per pixel.
Many hardware proposals have aimed to accelerate inference in AI workloads. Less attention has been paid to hardware acceleration of training, despite the enormous societal impact of rapid training of AI models. Physics-based computers, such as thermodynamic computers, offer an efficient means to solve key primitives in AI training algorithms. Optimizers that normally would be computationally out-of-reach (e.g., due to expensive matrix inversions) on digital hardware could be unlocked with physics-based hardware. In this work, we propose a scalable algorithm for employing thermodynamic computers to accelerate a popular second-order optimizer called Kronecker-factored approximate curvature (K-FAC). Our asymptotic complexity analysis predicts increasing advantage with our algorithm as $n$, the number of neurons per layer, increases. Numerical experiments show that even under significant quantization noise, the benefits of second-order optimization can be preserved. Finally, we predict substantial speedups for large-scale vision and graph problems based on realistic hardware characteristics.
In this paper, we propose a method to perform empirical analysis of the loss landscape of machine learning (ML) models. The method is applied to two ML models for scientific sensing, which necessitates quantization to be deployed and are subject to noise and perturbations due to experimental conditions. Our method allows assessing the robustness of ML models to such effects as a function of quantization precision and under different regularization techniques -- two crucial concerns that remained underexplored so far. By investigating the interplay between performance, efficiency, and robustness by means of loss landscape analysis, we both established a strong correlation between gently-shaped landscapes and robustness to input and weight perturbations and observed other intriguing and non-obvious phenomena. Our method allows a systematic exploration of such trade-offs a priori, i.e., without training and testing multiple models, leading to more efficient development workflows. This work also highlights the importance of incorporating robustness into the Pareto optimization of ML models, enabling more reliable and adaptive scientific sensing systems.
Fine-tuning large language models (LLMs) is increasingly costly as models scale to hundreds of billions of parameters, and even parameter-efficient fine-tuning (PEFT) methods like LoRA remain resource-intensive. We introduce LowRA, the first framework to enable LoRA fine-tuning below 2 bits per parameter with minimal performance loss. LowRA optimizes fine-grained quantization - mapping, threshold selection, and precision assignment - while leveraging efficient CUDA kernels for scalable deployment. Extensive evaluations across 4 LLMs and 4 datasets show that LowRA achieves a superior performance-precision trade-off above 2 bits and remains accurate down to 1.15 bits, reducing memory usage by up to 50%. Our results highlight the potential of ultra-low-bit LoRA fine-tuning for resource-constrained environments.
Compute-in-memory (CIM) is an efficient method for implementing deep neural networks (DNNs) but suffers from substantial overhead from analog-to-digital converters (ADCs), especially as ADC precision increases. Low-precision ADCs can reduce this overhead but introduce partial-sum quantization errors degrading accuracy. Additionally, low-bit weight constraints, imposed by cell limitations and the need for multiple cells for higher-bit weights, present further challenges. While fine-grained partial-sum quantization has been studied to lower ADC resolution effectively, weight granularity, which limits overall partial-sum quantized accuracy, remains underexplored. This work addresses these challenges by aligning weight and partial-sum quantization granularities at the column-wise level. Our method improves accuracy while maintaining dequantization overhead, simplifies training by removing two-stage processes, and ensures robustness to memory cell variations via independent column-wise scale factors. We also propose an open-source CIM-oriented convolution framework to handle fine-grained weights and partial-sums efficiently, incorporating a novel tiling method and group convolution. Experimental results on ResNet-20 (CIFAR-10, CIFAR-100) and ResNet-18 (ImageNet) show accuracy improvements of 0.99%, 2.69%, and 1.01%, respectively, compared to the best-performing related works. Additionally, variation analysis reveals the robustness of our method against memory cell variations. These findings highlight the effectiveness of our quantization scheme in enhancing accuracy and robustness while maintaining hardware efficiency in CIM-based DNN implementations. Our code is available at https://github.com/jiyoonkm/ColumnQuant.
The implementation of Hyperdimensional Computing (HDC) on In-Memory Computing (IMC) architectures faces significant challenges due to the mismatch between highdimensional vectors and IMC array sizes, leading to inefficient memory utilization and increased computation cycles. This paper presents MEMHD, a Memory-Efficient Multi-centroid HDC framework designed to address these challenges. MEMHD introduces a clustering-based initialization method and quantization aware iterative learning for multi-centroid associative memory. Through these approaches and its overall architecture, MEMHD achieves a significant reduction in memory requirements while maintaining or improving classification accuracy. Our approach achieves full utilization of IMC arrays and enables one-shot (or few-shot) associative search. Experimental results demonstrate that MEMHD outperforms state-of-the-art binary HDC models, achieving up to 13.69% higher accuracy with the same memory usage, or 13.25x more memory efficiency at the same accuracy level. Moreover, MEMHD reduces computation cycles by up to 80x and array usage by up to 71x compared to baseline IMC mapping methods when mapped to 128x128 IMC arrays, while significantly improving energy and computation cycle efficiency.
This study investigates the impact of gradient compression on distributed training performance, focusing on sparsification and quantization techniques, including top-k, DGC, and QSGD. In baseline experiments, random-k compression results in severe performance degradation, highlighting its inefficacy. In contrast, using top-k and DGC at 50 times compression yields performance improvements, reducing perplexity by up to 0.06 compared to baseline. Experiments across 1, 2, and 4 workers demonstrate that conservative sparsification can have a regularizing effect, especially for smaller models, while compression ratios above 5000 times impair performance, particularly for DGC. Communication times are reduced across all compression methods, with top-k and DGC decreasing communication to negligible levels at high compression ratios. However, increased computation times offset this efficiency for top-k due to sorting demands, making it less scalable than DGC or QSGD. In convergence tests, sparsification techniques show accelerated convergence, requiring fewer epochs than the baseline, which has implications for computational savings. Although precision trade-offs emerge, floating point errors are mitigated by compression. This study's findings underscore the need to tune hyperparameters specifically for each compression technique to achieve optimal model performance, especially in distributed training systems.
Conditional expectation \mathbb{E}(Y \mid X) often fails to capture the complexity of multimodal conditional distributions \mathcal{L}(Y \mid X). To address this, we propose using n-point conditional quantizations--functional mappings of X that are learnable via gradient descent--to approximate \mathcal{L}(Y \mid X). This approach adapts Competitive Learning Vector Quantization (CLVQ), tailored for conditional distributions. It goes beyond single-valued predictions by providing multiple representative points that better reflect multimodal structures. It enables the approximation of the true conditional law in the Wasserstein distance. The resulting framework is theoretically grounded and useful for uncertainty quantification and multimodal data generation tasks. For example, in computer vision inpainting tasks, multiple plausible reconstructions may exist for the same partially observed input image X. We demonstrate the effectiveness of our approach through experiments on synthetic and real-world datasets.
In practical applications, lattice quantizers leverage discrete lattice points to approximate arbitrary points in the lattice. An effective lattice quantizer significantly enhances both the accuracy and efficiency of these approximations. In the context of high-dimensional lattice quantization, previous work proposed utilizing low-dimensional optimal lattice quantizers and addressed the challenge of determining the optimal length ratio in orthogonal splicing. Notably, it was demonstrated that fixed length ratios and orthogonality yield suboptimal results when combining low-dimensional lattices. Building on this foundation, another approach employed gradient descent to identify optimal lattices, which inspired us to explore the use of neural networks to discover matrices that outperform those obtained from orthogonal splicing methods. We propose two novel approaches to tackle this problem: the Household Algorithm and the Matrix Exp Algorithm. Our results indicate that both the Household Algorithm and the Matrix Exp Algorithm achieve improvements in lattice quantizers across dimensions 13, 15, 17 to 19, 21, and 22. Moreover, the Matrix Exp Algorithm demonstrates superior efficacy in high-dimensional settings.
Large language models have been increasing in size due to their success in a wide range of applications. This calls for a pressing need to reduce memory usage to make them more accessible. Post-training quantization is a popular technique which uses fewer bits (e.g., 4--8 bits) to represent the model without retraining it. However, it remains a challenging task to perform quantization in an ultra-low-bit setup (e.g., 2 bits). In this paper, we propose InvarExplore, a unified framework that systematically explores different model invariance at the same time, allowing us to take advantage of the synergy between each type of invariance. Importantly, InvarExplore features a discrete search algorithm that enables us to explore permutation invariance, which is under-studied as it cannot be optimized with gradient-based methods. Results show that InvarExplore is compatible with existing state-of-the-art methods, achieving an add-on performance improvement over strong competing methods.
Large transformer models are known to produce high-norm tokens. In vision transformers (ViTs), such tokens have been mathematically modeled through the singular vectors of the linear approximations of layers. However, in large language models (LLMs), the underlying causes of high-norm tokens remain largely unexplored, and their different properties from those of ViTs require a new analysis framework. In this paper, we provide both theoretical insights and empirical validation across a range of recent models, leading to the following observations: i) The layer-wise singular direction predicts the abrupt explosion of token norms in LLMs. ii) The negative eigenvalues of a layer explain its sudden decay. iii) The computational pathways leading to high-norm tokens differ between initial and noninitial tokens. iv) High-norm tokens are triggered by the right leading singular vector of the matrix approximating the corresponding modules. We showcase two practical applications of these findings: the improvement of quantization schemes and the design of LLM signatures. Our findings not only advance the understanding of singular defects in LLMs but also open new avenues for their application. We expect that this work will stimulate further research into the internal mechanisms of LLMs and will therefore publicly release our code.
We develop diffusion models for simulating lattice gauge theories, where stochastic quantization is explicitly incorporated as a physical condition for sampling. We demonstrate the applicability of this novel sampler to U(1) gauge theory in two spacetime dimensions and find that a model trained at a small inverse coupling constant can be extrapolated to larger inverse coupling regions without encountering the topological freezing problem. Additionally, the trained model can be employed to sample configurations on different lattice sizes without requiring further training. The exactness of the generated samples is ensured by incorporating Metropolis-adjusted Langevin dynamics into the generation process. Furthermore, we demonstrate that this approach enables more efficient sampling of topological quantities compared to traditional algorithms such as Hybrid Monte Carlo and Langevin simulations.
Image compression under ultra-low bitrates remains challenging for both conventional learned image compression (LIC) and generative vector-quantized (VQ) modeling. Conventional LIC suffers from severe artifacts due to heavy quantization, while generative VQ modeling gives poor fidelity due to the mismatch between learned generative priors and specific inputs. In this work, we propose Hybrid-Diffusion Image Compression (HDCompression), a dual-stream framework that utilizes both generative VQ-modeling and diffusion models, as well as conventional LIC, to achieve both high fidelity and high perceptual quality. Different from previous hybrid methods that directly use pre-trained LIC models to generate low-quality fidelity-preserving information from heavily quantized latent, we use diffusion models to extract high-quality complimentary fidelity information from the ground-truth input, which can enhance the system performance in several aspects: improving indices map prediction, enhancing the fidelity-preserving output of the LIC stream, and refining conditioned image reconstruction with VQ-latent correction. In addition, our diffusion model is based on a dense representative vector (DRV), which is lightweight with very simple sampling schedulers. Extensive experiments demonstrate that our HDCompression outperforms the previous conventional LIC, generative VQ-modeling, and hybrid frameworks in both quantitative metrics and qualitative visualization, providing balanced robust compression performance at ultra-low bitrates.
Recently, large language model (LLM) based text-to-speech (TTS) systems have gradually become the mainstream in the industry due to their high naturalness and powerful zero-shot voice cloning capabilities.Here, we introduce the IndexTTS system, which is mainly based on the XTTS and Tortoise model. We add some novel improvements. Specifically, in Chinese scenarios, we adopt a hybrid modeling method that combines characters and pinyin, making the pronunciations of polyphonic characters and long-tail characters controllable. We also performed a comparative analysis of the Vector Quantization (VQ) with Finite-Scalar Quantization (FSQ) for codebook utilization of acoustic speech tokens. To further enhance the effect and stability of voice cloning, we introduce a conformer-based speech conditional encoder and replace the speechcode decoder with BigVGAN2. Compared with XTTS, it has achieved significant improvements in naturalness, content consistency, and zero-shot voice cloning. As for the popular TTS systems in the open-source, such as Fish-Speech, CosyVoice2, FireRedTTS and F5-TTS, IndexTTS has a relatively simple training process, more controllable usage, and faster inference speed. Moreover, its performance surpasses that of these systems. Our demos are available at https://index-tts.github.io.
Graph Neural Networks (GNNs) are vital for learning from graph-structured data, enabling applications in network analysis, recommendation systems, and speech analytics. Deploying them on edge devices like client PCs and laptops enhances real-time processing, privacy, and cloud independence. GNNs aid Retrieval-Augmented Generation (RAG) for Large Language Models (LLMs) and enable event-based vision tasks. However, irregular memory access, sparsity, and dynamic structures cause high latency and energy overhead on resource-constrained devices. While modern edge processors integrate CPUs, GPUs, and NPUs, NPUs designed for data-parallel tasks struggle with irregular GNN computations. We introduce GraNNite, the first hardware-aware framework optimizing GNN execution on commercial-off-the-shelf (COTS) SOTA DNN accelerators via a structured three-step methodology: (1) enabling NPU execution, (2) optimizing performance, and (3) trading accuracy for efficiency gains. Step 1 employs GraphSplit for workload distribution and StaGr for static aggregation, while GrAd and NodePad handle dynamic graphs. Step 2 boosts performance using EffOp for control-heavy tasks and GraSp for sparsity exploitation. Graph Convolution optimizations PreG, SymG, and CacheG reduce redundancy and memory transfers. Step 3 balances quality versus efficiency, where QuantGr applies INT8 quantization, and GrAx1, GrAx2, and GrAx3 accelerate attention, broadcast-add, and SAGE-max aggregation. On Intel Core Ultra AI PCs, GraNNite achieves 2.6X to 7.6X speedups over default NPU mappings and up to 8.6X energy gains over CPUs and GPUs, delivering 10.8X and 6.7X higher performance than CPUs and GPUs, respectively, across GNN models.
We introduce Quantized Language-Image Pretraining (QLIP), a visual tokenization method that combines state-of-the-art reconstruction quality with state-of-the-art zero-shot image understanding. QLIP trains a binary-spherical-quantization-based autoencoder with reconstruction and language-image alignment objectives. We are the first to show that the two objectives do not need to be at odds. We balance the two loss terms dynamically during training and show that a two-stage training pipeline effectively mixes the large-batch requirements of image-language pre-training with the memory bottleneck imposed by the reconstruction objective. We validate the effectiveness of QLIP for multimodal understanding and text-conditioned image generation with a single model. Specifically, QLIP serves as a drop-in replacement for the visual encoder for LLaVA and the image tokenizer for LlamaGen with comparable or even better performance. Finally, we demonstrate that QLIP enables a unified mixed-modality auto-regressive model for understanding and generation.
Post-training quantization (PTQ) has emerged as a promising solution for reducing the storage and computational cost of vision transformers (ViTs). Recent advances primarily target at crafting quantizers to deal with peculiar activations characterized by ViTs. However, most existing methods underestimate the information loss incurred by weight quantization, resulting in significant performance deterioration, particularly in low-bit cases. Furthermore, a common practice in quantizing post-Softmax activations of ViTs is to employ logarithmic transformations, which unfortunately prioritize less informative values around zero. This approach introduces additional redundancies, ultimately leading to suboptimal quantization efficacy. To handle these, this paper proposes an innovative PTQ method tailored for ViTs, termed AIQViT (Architecture-Informed Post-training Quantization for ViTs). First, we design an architecture-informed low rank compensation mechanism, wherein learnable low-rank weights are introduced to compensate for the degradation caused by weight quantization. Second, we design a dynamic focusing quantizer to accommodate the unbalanced distribution of post-Softmax activations, which dynamically selects the most valuable interval for higher quantization resolution. Extensive experiments on five vision tasks, including image classification, object detection, instance segmentation, point cloud classification, and point cloud part segmentation, demonstrate the superiority of AIQViT over state-of-the-art PTQ methods.
This paper is focused on statistical learning from data that come as probability measures. In this setting, popular approaches consist in embedding such data into a Hilbert space with either Linearized Optimal Transport or Kernel Mean Embedding. However, the cost of computing such embeddings prohibits their direct use in large-scale settings. We study two methods based on measure quantization for approximating input probability measures with discrete measures of small-support size. The first one is based on optimal quantization of each input measure, while the second one relies on mean-measure quantization. We study the consistency of such approximations, and its implication for scalable embeddings of probability measures into a Hilbert space at a low computational cost. We finally illustrate our findings with various numerical experiments.
KV cache quantization can improve Large Language Models (LLMs) inference throughput and latency in long contexts and large batch-size scenarios while preserving LLMs effectiveness. However, current methods have three unsolved issues: overlooking layer-wise sensitivity to KV cache quantization, high overhead of online fine-grained decision-making, and low flexibility to different LLMs and constraints. Therefore, we thoroughly analyze the inherent correlation of layer-wise transformer attention patterns to KV cache quantization errors and study why key cache is more important than value cache for quantization error reduction. We further propose a simple yet effective framework KVTuner to adaptively search for the optimal hardware-friendly layer-wise KV quantization precision pairs for coarse-grained KV cache with multi-objective optimization and directly utilize the offline searched configurations during online inference. To reduce the computational cost of offline calibration, we utilize the intra-layer KV precision pair pruning and inter-layer clustering to reduce the search space. Experimental results show that we can achieve nearly lossless 3.25-bit mixed precision KV cache quantization for LLMs like Llama-3.1-8B-Instruct and 4.0-bit for sensitive models like Qwen2.5-7B-Instruct on mathematical reasoning tasks. The maximum inference throughput can be improved by 38.3% compared with KV8 quantization over various context lengths.
Diffusion transformers (DiTs) combine transformer architectures with diffusion models. However, their computational complexity imposes significant limitations on real-time applications and sustainability of AI systems. In this study, we aim to enhance the computational efficiency through model quantization, which represents the weights and activation values with lower precision. Multi-region quantization (MRQ) is introduced to address the asymmetric distribution of network values in DiT blocks by allocating two scaling parameters to sub-regions. Additionally, time-grouping quantization (TGQ) is proposed to reduce quantization error caused by temporal variation in activations. The experimental results show that the proposed algorithm achieves performance comparable to the original full-precision model with only a 0.29 increase in FID at W8A8. Furthermore, it outperforms other baselines at W6A6, thereby confirming its suitability for low-bit quantization. These results highlight the potential of our method to enable efficient real-time generative models.