Quantization and fine-tuning are crucial for deploying large language models (LLMs) on resource-constrained edge devices. However, fine-tuning quantized models presents significant challenges, primarily stemming from: First, the mismatch in data types between the low-precision quantized weights (e.g., 4-bit) and the high-precision adaptation weights (e.g., 16-bit). This mismatch limits the computational efficiency advantage offered by quantized weights during inference. Second, potential accuracy degradation when merging these high-precision adaptation weights into the low-precision quantized weights, as the adaptation weights often necessitate approximation or truncation. Third, as far as we know, no existing methods support the lossless merging of adaptation while adjusting all quantized weights. To address these challenges, we introduce lossless ternary adaptation for quantization-aware fine-tuning (LoTA-QAF). This is a novel fine-tuning method specifically designed for quantized LLMs, enabling the lossless merging of ternary adaptation weights into quantized weights and the adjustment of all quantized weights. LoTA-QAF operates through a combination of: i) A custom-designed ternary adaptation (TA) that aligns ternary weights with the quantization grid and uses these ternary weights to adjust quantized weights. ii) A TA-based mechanism that enables the lossless merging of adaptation weights. iii) Ternary signed gradient descent (t-SignSGD) for updating the TA weights. We apply LoTA-QAF to Llama-3.1/3.3 and Qwen-2.5 model families and validate its effectiveness on several downstream tasks. On the MMLU benchmark, our method effectively recovers performance for quantized models, surpassing 16-bit LoRA by up to 5.14\%. For task-specific fine-tuning, 16-bit LoRA achieves superior results, but LoTA-QAF still outperforms other methods.
Efficient deployment of large audio-language models for speech translation remains challenging due to their significant computational requirements. In this paper, we address this challenge through our system submissions to the "Model Compression" track at the International Conference on Spoken Language Translation (IWSLT 2025). We experiment with a combination of approaches including iterative layer pruning based on layer importance evaluation, low-rank adaptation with 4-bit quantization (QLoRA), and knowledge distillation. In our experiments, we use Qwen2-Audio-7B-Instruct for speech translation into German and Chinese. Our pruned (student) models achieve up to a 50% reduction in both model parameters and storage footprint, while retaining 97-100% of the translation quality of the in-domain (teacher) models.
The Key-Value (KV) cache in generative large language models (LLMs) introduces substantial memory overhead. Existing works mitigate this burden by offloading or compressing the KV cache. However, loading the entire cache incurs significant latency due to PCIe bandwidth bottlenecks in CPU-GPU communication, while aggressive compression causes notable performance degradation. We identify that certain layers in the LLM need to maintain global information and are unsuitable for selective loading. In contrast, other layers primarily focus on a few tokens with dominant activations that potentially incur substantial quantization error. This observation leads to a key insight that loading dominant tokens and quantizing all tokens can complement each other. Building on this insight, we propose a hybrid compression method, TailorKV, which seamlessly integrates quantization and offloading. TailorKV develops an inference framework along with a hardware-friendly implementation that leverages these complementary characteristics. Extensive long-context evaluations exhibit that TailorKV achieves nearly lossless performance under aggressive compression settings, outperforming the state-of-the-art. Particularly, the Llama-3.1-8B with 128k context can be served within a single RTX 3090 GPU, reaching 82 ms per token during decoding.
Diffusion Transformers (DiTs) have emerged as the state-of-the-art architecture for video generation, yet their computational and memory demands hinder practical deployment. While post-training quantization (PTQ) presents a promising approach to accelerate Video DiT models, existing methods suffer from two critical limitations: (1) dependence on lengthy, computation-heavy calibration procedures, and (2) considerable performance deterioration after quantization. To address these challenges, we propose DVD-Quant, a novel Data-free quantization framework for Video DiTs. Our approach integrates three key innovations: (1) Progressive Bounded Quantization (PBQ) and (2) Auto-scaling Rotated Quantization (ARQ) for calibration data-free quantization error reduction, as well as (3) $\delta$-Guided Bit Switching ($\delta$-GBS) for adaptive bit-width allocation. Extensive experiments across multiple video generation benchmarks demonstrate that DVD-Quant achieves an approximately 2$\times$ speedup over full-precision baselines on HunyuanVideo while maintaining visual fidelity. Notably, DVD-Quant is the first to enable W4A4 PTQ for Video DiTs without compromising video quality. Code and models will be available at https://github.com/lhxcs/DVD-Quant.
Artificial intelligence and machine learning models deployed on edge devices, e.g., for quality control in Additive Manufacturing (AM), are frequently small in size. Such models usually have to deliver highly accurate results within a short time frame. Methods that are commonly employed in literature start out with larger trained models and try to reduce their memory and latency footprint by structural pruning, knowledge distillation, or quantization. It is, however, also possible to leverage hardware-aware Neural Architecture Search (NAS), an approach that seeks to systematically explore the architecture space to find optimized configurations. In this study, a hardware-aware NAS workflow is introduced that couples an edge device located in Belgium with a powerful High-Performance Computing system in Germany, to train possible architecture candidates as fast as possible while performing real-time latency measurements on the target hardware. The approach is verified on a use case in the AM domain, based on the open RAISE-LPBF dataset, achieving ~8.8 times faster inference speed while simultaneously enhancing model quality by a factor of ~1.35, compared to a human-designed baseline.
Large language models (LLMs) now support context windows exceeding 128K tokens, but this comes with significant memory requirements and high inference latency. Quantization can mitigate these costs, but may degrade performance. In this work, we present the first systematic evaluation of quantized LLMs on tasks with long-inputs (>64K tokens) and long-form outputs. Our evaluation spans 9.7K test examples, five quantization methods (FP8, GPTQ-int8, AWQ-int4, GPTQ-int4, BNB-nf4), and five models (Llama-3.1 8B and 70B; Qwen-2.5 7B, 32B, and 72B). We find that, on average, 8-bit quantization preserves accuracy (~0.8% drop), whereas 4-bit methods lead to substantial losses, especially for tasks involving long context inputs (drops of up to 59%). This degradation tends to worsen when the input is in a language other than English. Crucially, the effects of quantization depend heavily on the quantization method, model, and task. For instance, while Qwen-2.5 72B remains robust under BNB-nf4, Llama-3.1 70B experiences a 32% performance drop on the same task. These findings highlight the importance of a careful, task-specific evaluation before deploying quantized LLMs, particularly in long-context scenarios and with languages other than English.
Recently, significant progress has been made in developing reasoning-capable Large Language Models (LLMs) through long Chain-of-Thought (CoT) techniques. However, this long-CoT reasoning process imposes substantial memory overhead due to the large Key-Value (KV) Cache memory overhead. Post-training KV Cache quantization has emerged as a promising compression technique and has been extensively studied in short-context scenarios. However, directly applying existing methods to long-CoT LLMs causes significant performance degradation due to the following two reasons: (1) Large cumulative error: Existing methods fail to adequately leverage available memory, and they directly quantize the KV Cache during each decoding step, leading to large cumulative quantization error. (2) Short-context calibration: Due to Rotary Positional Embedding (RoPE), the use of short-context data during calibration fails to account for the distribution of less frequent channels in the Key Cache, resulting in performance loss. We propose Progressive Mixed-Precision KV Cache Quantization (PM-KVQ) for long-CoT LLMs to address the above issues in two folds: (1) To reduce cumulative error, we design a progressive quantization strategy to gradually lower the bit-width of KV Cache in each block. Then, we propose block-wise memory allocation to assign a higher bit-width to more sensitive transformer blocks. (2) To increase the calibration length without additional overhead, we propose a new calibration strategy with positional interpolation that leverages short calibration data with positional interpolation to approximate the data distribution of long-context data. Extensive experiments on 7B-70B long-CoT LLMs show that PM-KVQ improves reasoning benchmark performance by up to 8% over SOTA baselines under the same memory budget. Our code is available at https://github.com/thu-nics/PM-KVQ.
The emergence of multi-codebook neutral audio codecs such as Residual Vector Quantization (RVQ) and Group Vector Quantization (GVQ) has significantly advanced Large-Language-Model (LLM) based Text-to-Speech (TTS) systems. These codecs are crucial in separating semantic and acoustic information while efficiently harnessing semantic priors. However, since semantic and acoustic information cannot be fully aligned, a significant drawback of these methods when applied to LLM-based TTS is that large language models may have limited access to comprehensive audio information. To address this limitation, we propose DistilCodec and UniTTS, which collectively offer the following advantages: 1) This method can distill a multi-codebook audio codec into a single-codebook audio codec with 32,768 codes while achieving a near 100\% utilization. 2) As DistilCodec does not employ a semantic alignment scheme, a large amount of high-quality unlabeled audio (such as audiobooks with sound effects, songs, etc.) can be incorporated during training, further expanding data diversity and broadening its applicability. 3) Leveraging the comprehensive audio information modeling of DistilCodec, we integrated three key tasks into UniTTS's pre-training framework: audio modality autoregression, text modality autoregression, and speech-text cross-modal autoregression. This allows UniTTS to accept interleaved text and speech/audio prompts while substantially preserving LLM's text capabilities. 4) UniTTS employs a three-stage training process: Pre-Training, Supervised Fine-Tuning (SFT), and Alignment. Source code and model checkpoints are publicly available at https://github.com/IDEA-Emdoor-Lab/UniTTS and https://github.com/IDEA-Emdoor-Lab/DistilCodec.
Visual autoregressive (VAR) modeling has marked a paradigm shift in image generation from next-token prediction to next-scale prediction. VAR predicts a set of tokens at each step from coarse to fine scale, leading to better image quality and faster inference speed compared to existing diffusion models. However, the large parameter size and computation cost hinder its deployment on edge devices. To reduce the memory and computation cost, we propose FPQVAR, an efficient post-training floating-point (FP) quantization framework for VAR featuring algorithm and hardware co-design. At the algorithm level, we first identify the challenges of quantizing VAR. To address them, we propose Dual Format Quantization for the highly imbalanced input activation. We further propose Group-wise Hadamard Transformation and GHT-Aware Learnable Transformation to address the time-varying outlier channels. At the hardware level, we design the first low-bit FP quantizer and multiplier with lookup tables on FPGA and propose the first FPGA-based VAR accelerator featuring low-bit FP computation and an elaborate two-level pipeline. Extensive experiments show that compared to the state-of-the-art quantization method, our proposed FPQVAR significantly improves Fr\'echet Inception Distance (FID) from 10.83 to 3.58, Inception Score (IS) from 175.9 to 241.5 under 4-bit quantization. FPQVAR also significantly improves the performance of 6-bit quantized VAR, bringing it on par with the FP16 model. Our accelerator on AMD-Xilinx VCK190 FPGA achieves a throughput of 1.1 image/s, which is 3.1x higher than the integer-based accelerator. It also demonstrates 3.6x and 2.8x higher energy efficiency compared to the integer-based accelerator and GPU baseline, respectively.
Large language models (LLMs) are considered valuable Intellectual Properties (IP) for legitimate owners due to the enormous computational cost of training. It is crucial to protect the IP of LLMs from malicious stealing or unauthorized deployment. Despite existing efforts in watermarking and fingerprinting LLMs, these methods either impact the text generation process or are limited in white-box access to the suspect model, making them impractical. Hence, we propose DuFFin, a novel $\textbf{Du}$al-Level $\textbf{Fin}$gerprinting $\textbf{F}$ramework for black-box setting ownership verification. DuFFin extracts the trigger pattern and the knowledge-level fingerprints to identify the source of a suspect model. We conduct experiments on a variety of models collected from the open-source website, including four popular base models as protected LLMs and their fine-tuning, quantization, and safety alignment versions, which are released by large companies, start-ups, and individual users. Results show that our method can accurately verify the copyright of the base protected LLM on their model variants, achieving the IP-ROC metric greater than 0.95. Our code is available at https://github.com/yuliangyan0807/llm-fingerprint.
Large language models (LLMs) have made exciting achievements across various domains, yet their deployment on resource-constrained personal devices remains hindered by the prohibitive computational and memory demands of task-specific fine-tuning. While quantization offers a pathway to efficiency, existing methods struggle to balance performance and overhead, either incurring high computational/memory costs or failing to address activation outliers, a critical bottleneck in quantized fine-tuning. To address these challenges, we propose the Outlier Spatial Stability Hypothesis (OSSH): During fine-tuning, certain activation outlier channels retain stable spatial positions across training iterations. Building on OSSH, we propose Quaff, a Quantized parameter-efficient fine-tuning framework for LLMs, optimizing low-precision activation representations through targeted momentum scaling. Quaff dynamically suppresses outliers exclusively in invariant channels using lightweight operations, eliminating full-precision weight storage and global rescaling while reducing quantization errors. Extensive experiments across ten benchmarks validate OSSH and demonstrate Quaff's efficacy. Specifically, on the GPQA reasoning benchmark, Quaff achieves a 1.73x latency reduction and 30% memory savings over full-precision fine-tuning while improving accuracy by 0.6% on the Phi-3 model, reconciling the triple trade-off between efficiency, performance, and deployability. By enabling consumer-grade GPU fine-tuning (e.g., RTX 2080 Super) without sacrificing model utility, Quaff democratizes personalized LLM deployment. The code is available at https://github.com/Little0o0/Quaff.git.
Spoken Language Understanding (SLU) systems must balance performance and efficiency, particularly in resource-constrained environments. Existing methods apply distillation and quantization separately, leading to suboptimal compression as distillation ignores quantization constraints. We propose QUADS, a unified framework that optimizes both through multi-stage training with a pre-tuned model, enhancing adaptability to low-bit regimes while maintaining accuracy. QUADS achieves 71.13\% accuracy on SLURP and 99.20\% on FSC, with only minor degradations of up to 5.56\% compared to state-of-the-art models. Additionally, it reduces computational complexity by 60--73$\times$ (GMACs) and model size by 83--700$\times$, demonstrating strong robustness under extreme quantization. These results establish QUADS as a highly efficient solution for real-world, resource-constrained SLU applications.
Convolutional Neural Networks (CNNs) have gained high popularity as a tool for computer vision tasks and for that reason are used in various applications. There are many different concepts, like single shot detectors, that have been published for detecting objects in images or video streams. However, CNNs suffer from disadvantages regarding the deployment on embedded platforms such as re-configurable hardware like Field Programmable Gate Arrays (FPGAs). Due to the high computational intensity, memory requirements and arithmetic conditions, a variety of strategies for running CNNs on FPGAs have been developed. The following methods showcase our best practice approaches for a TinyYOLOv3 detector network on a XILINX Artix-7 FPGA using techniques like fusion of batch normalization, filter pruning and post training network quantization.
Deep neural networks have achieved state-of-the-art results in a wide range of applications, from natural language processing and computer vision to speech recognition. However, as tasks become increasingly complex, model sizes continue to grow, posing challenges in latency and memory efficiency. To meet these constraints, post-training quantization has emerged as a promising solution. In this paper, we propose a novel hardware-efficient quantization and inference scheme that exploits hardware advantages with minimal accuracy degradation. Specifically, we introduce a W4A8 scheme, where weights are quantized and stored using 4-bit integer precision, and inference computations are performed using 8-bit floating-point arithmetic, demonstrating significant speedups and improved memory utilization compared to 16-bit operations, applicable on various modern accelerators. To mitigate accuracy loss, we develop a novel quantization algorithm, dubbed Dual Precision Quantization (DPQ), that leverages the unique structure of our scheme without introducing additional inference overhead. Experimental results demonstrate improved performance (i.e., increased throughput) while maintaining tolerable accuracy degradation relative to the full-precision model.
Using Resistive Random Access Memory (RRAM) crossbars in Computing-in-Memory (CIM) architectures offers a promising solution to overcome the von Neumann bottleneck. Due to non-idealities like cell variability, RRAM crossbars are often operated in binary mode, utilizing only two states: Low Resistive State (LRS) and High Resistive State (HRS). Binary Neural Networks (BNNs) and Ternary Neural Networks (TNNs) are well-suited for this hardware due to their efficient mapping. Existing software projects for RRAM-based CIM typically focus on only one aspect: compilation, simulation, or Design Space Exploration (DSE). Moreover, they often rely on classical 8 bit quantization. To address these limitations, we introduce CIM-Explorer, a modular toolkit for optimizing BNN and TNN inference on RRAM crossbars. CIM-Explorer includes an end-to-end compiler stack, multiple mapping options, and simulators, enabling a DSE flow for accuracy estimation across different crossbar parameters and mappings. CIM-Explorer can accompany the entire design process, from early accuracy estimation for specific crossbar parameters, to selecting an appropriate mapping, and compiling BNNs and TNNs for a finalized crossbar chip. In DSE case studies, we demonstrate the expected accuracy for various mappings and crossbar parameters. CIM-Explorer can be found on GitHub.
We introduce SLED, an alternative approach to speech language modeling by encoding speech waveforms into sequences of continuous latent representations and modeling them autoregressively using an energy distance objective. The energy distance offers an analytical measure of the distributional gap by contrasting simulated and target samples, enabling efficient training to capture the underlying continuous autoregressive distribution. By bypassing reliance on residual vector quantization, SLED avoids discretization errors and eliminates the need for the complicated hierarchical architectures common in existing speech language models. It simplifies the overall modeling pipeline while preserving the richness of speech information and maintaining inference efficiency. Empirical results demonstrate that SLED achieves strong performance in both zero-shot and streaming speech synthesis, showing its potential for broader applications in general-purpose speech language models.
Multi-frame video enhancement tasks aim to improve the spatial and temporal resolution and quality of video sequences by leveraging temporal information from multiple frames, which are widely used in streaming video processing, surveillance, and generation. Although numerous Transformer-based enhancement methods have achieved impressive performance, their computational and memory demands hinder deployment on edge devices. Quantization offers a practical solution by reducing the bit-width of weights and activations to improve efficiency. However, directly applying existing quantization methods to video enhancement tasks often leads to significant performance degradation and loss of fine details. This stems from two limitations: (a) inability to allocate varying representational capacity across frames, which results in suboptimal dynamic range adaptation; (b) over-reliance on full-precision teachers, which limits the learning of low-bit student models. To tackle these challenges, we propose a novel quantization method for video enhancement: Progressive Multi-Frame Quantization for Video Enhancement (PMQ-VE). This framework features a coarse-to-fine two-stage process: Backtracking-based Multi-Frame Quantization (BMFQ) and Progressive Multi-Teacher Distillation (PMTD). BMFQ utilizes a percentile-based initialization and iterative search with pruning and backtracking for robust clipping bounds. PMTD employs a progressive distillation strategy with both full-precision and multiple high-bit (INT) teachers to enhance low-bit models' capacity and quality. Extensive experiments demonstrate that our method outperforms existing approaches, achieving state-of-the-art performance across multiple tasks and benchmarks.The code will be made publicly available at: https://github.com/xiaoBIGfeng/PMQ-VE.
As the size of large language models grows exponentially, GPU memory has become a bottleneck for adapting these models to downstream tasks. In this paper, we aim to push the limits of memory-efficient training by minimizing memory usage on model weights, gradients, and optimizer states, within a unified framework. Our idea is to eliminate both gradients and optimizer states using zeroth-order optimization, which approximates gradients by perturbing weights during forward passes to identify gradient directions. To minimize memory usage on weights, we employ model quantization, e.g., converting from bfloat16 to int4. However, directly applying zeroth-order optimization to quantized weights is infeasible due to the precision gap between discrete weights and continuous gradients, which would otherwise require de-quantization and re-quantization. To overcome this challenge, we propose Quantized Zeroth-order Optimization (QZO), a novel approach that perturbs the continuous quantization scale for gradient estimation and uses a directional derivative clipping method to stabilize training. QZO is orthogonal to both scalar-based and codebook-based post-training quantization methods. Compared to full-parameter fine-tuning in bfloat16, QZO can reduce the total memory cost by more than 18$\times$ for 4-bit LLMs, and enables fine-tuning Llama-2-13B and Stable Diffusion 3.5 Large within a single 24GB GPU.
The impressive capabilities of Large Language Models (LLMs) come at the cost of substantial computational resources during deployment. While KV Cache can significantly reduce recomputation during inference, it also introduces additional memory overhead. KV Cache quantization presents a promising solution, striking a good balance between memory usage and accuracy. Previous research has shown that the Keys are distributed by channel, while the Values are distributed by token. Consequently, the common practice is to apply channel-wise quantization to the Keys and token-wise quantization to the Values. However, our further investigation reveals that a small subset of unusual tokens exhibit unique characteristics that deviate from this pattern, which can substantially impact quantization accuracy. To address this, we develop a simple yet effective method to identify these tokens accurately during the decoding process and exclude them from quantization as outlier tokens, significantly improving overall accuracy. Extensive experiments show that our method achieves significant accuracy improvements under 2-bit quantization and can deliver a 6.4 times reduction in memory usage and a 2.3 times increase in throughput.
Binary quantization approaches, which replace weight matrices with binary matrices and substitute costly multiplications with cheaper additions, offer a computationally efficient approach to address the increasing computational and storage requirements of Large Language Models (LLMs). However, the severe quantization constraint ($\pm1$) can lead to significant accuracy degradation. In this paper, we propose Double Binary Factorization (DBF), a novel method that factorizes dense weight matrices into products of two binary (sign) matrices, each accompanied by scaling vectors. DBF preserves the efficiency advantages of binary representations while achieving compression rates that are competitive with or superior to state-of-the-art methods. Specifically, in a 1-bit per weight range, DBF is better than existing binarization approaches. In a 2-bit per weight range, DBF is competitive with the best quantization methods like QuIP\# and QTIP. Unlike most existing compression techniques, which offer limited compression level choices, DBF allows fine-grained control over compression ratios by adjusting the factorization's intermediate dimension. Based on this advantage, we further introduce an algorithm for estimating non-uniform layer-wise compression ratios for DBF, based on previously developed channel pruning criteria. Code available at: https://github.com/usamec/double_binary