Quantization
-
RECAP: Calibrating Low-Bit Reasoning Models on the History They Read
2026-10-11 01:10:33Chi Ruan, Yujie Chen, Zixiang Zhao, Jingcai Guo, Haotong QinAbstract
Low-bit quantization reduces the cost of serving reasoning models but can substantially degrade long chains of thought. Each step reads cached keys and values spanning thousands of tokens, allowing quantization errors to accumulate along the trajectory. Standard calibration operates on short segments that omit the earlier context needed by later reasoning steps. Even when this context is retained, cache reconstruction error alone does not capture how quantization errors affect subsequent attention. We present RECAP, a post-training quantization method that calibrates reasoning models on the history they read. Dynamic-prefix calibration separates long historical context from short-block reconstruction. It regenerates prefix keys and values with the current quantizer and rotates the active block to supervise every position along the trajectory. Prefix read geometry further supervises historical cache errors according to their effects on attention outputs, weighting key and value errors asymmetrically. With both components applied only during calibration, RECAP outperforms the baseline across multiple reasoning models and quantization settings. Under W4A4KV4, RECAP improves accuracy over the baseline by 13.3 percentage points on MATH500 with DeepSeek-R1-Distill-Qwen-1.5B and 15.8 points on AIME-120 with Qwen3-4B. These results highlight the importance of preserving historical dependencies during calibration for low-bit reasoning.
Abstract (translated)
URL
https://arxiv.org/abs/2610.15259
PDF
-
HE-NAC: High-Efficiency Neural Audio Coding via Hierarchical and Residual Multi-Band Quantization
2026-10-11 00:33:45Darius P\'etermann, Wootaek Lim, Seungkwon Beack, Minje KimAbstract
High-fidelity audio coding requires faithful reconstruction of perceptually important high-frequency components, which are often lost due to the aggressive temporal decimation and quantization of neural codecs. State-of-the-art neural audio coding methods rely on adversarial learning to hallucinate these missing components, yielding perceptually plausible but imprecise reconstructions; a limitation for high-fidelity music applications. As a remedy, we propose high-efficiency neural audio coding (HE-NAC), which provides a multi-band extension of the popular residual vector quantization (RVQ) scheme. HE-NAC introduces an effective training mechanism that harmonizes a bitrate-free blind upsampling process, residual learning to compensate it with minimal bitrate, and a multi-head decoder architecture for band-specific reconstruction. As a result, HE-NAC achieves a balanced trade-off, reliably reconstructing the core-band ($\leq$ 3 kHz) component while preserving overall sound fidelity. Subjective listening tests and objective evaluations demonstrate that HE-NAC outperforms both the DAC baseline and MPEG HE-AAC at equal or lower bitrates.
Abstract (translated)
URL
https://arxiv.org/abs/2610.15243
PDF
-
VLCQ: Post-training quantization for deep neural networks using variable length coding
2026-10-10 18:02:45Reem Abdel-Salam, Ahmed H. Abdel-Gawad, Amr G. WassalAbstract
Quantization plays a crucial role in efficiently deploying deep learning models on resource-constrained devices. Post-training quantization does not require access to the original dataset or retraining the full model. Current methods that achieve high performance close to baseline results typically require INT8 fixed-point integers. However, achieving higher model compression with lower bit-widths often leads to significant performance degradation. In this paper, we propose VLCQ, a post-training quantization method that relaxes the fixed-point encoding constraint, enabling more flexible exploration of quantization techniques through variable-length encoding. This approach achieves competitive or improved performance while reducing bit-widths, without requiring access to training data or model fine-tuning. Extensive experiments on image classification, segmentation, and object detection demonstrate that VLCQ improves model compression and accuracy compared with state-of-the-art post-training quantization approaches. For per-channel quantization, VLCQ surpasses FP32 accuracy and Piece-Wise Linear Quantization (PWLQ) on most evaluated models, achieving up to 6x compression compared with FP32 and up to 1.7x compared with PWLQ. When prioritizing compression with limited accuracy loss, VLCQ achieves up to 12.25x compression compared with FP32 with less than 4% performance degradation. For per-tensor quantization, VLCQ is competitive with Data-Free Quantization (DFQ), while providing greater flexibility to achieve lower bit rates across different tasks and models.
Abstract (translated)
URL
https://arxiv.org/abs/2610.15092
PDF
-
SAGE: Pushing 2-Bit Vector-Quantized LLMs toward Real-World Usability
2026-10-10 09:05:14Xianglong Yan, Zenan Wu, Kewei Zhang, Kaicheng Yang, Yulun ZhangAbstract
Extreme low-bit weight-only quantization is an effective approach to reducing the deployment costs of large language models. Below 3 bits, scalar quantization often suffers severe accuracy loss, while vector quantization (VQ) better preserves accuracy through more expressive joint encoding. However, existing VQ methods still exhibit substantial degradation on real-world tasks, and two design challenges remain underexplored: (1) protecting sensitive weight channels within jointly encoded vectors; and (2) allocating a fixed bit budget among representation components, such as direction and radius, to suit different weight matrices. To address these challenges, we propose SAGE: Split Adaptation and Grouping by Energy for weight-only vector quantization. First, Energy-Stratified Grouping (ESG) couples channel grouping with activation-weighted fitting, grouping channels from different energy tiers within each subvector to prioritize sensitive channels. Second, Adaptive Bit Split (ABS) balances direction-codebook capacity and radius precision per matrix, selecting the split with the lowest activation-weighted reconstruction error under a fixed budget. Across benchmarks covering knowledge, mathematics, code generation, instruction following, and tool use, SAGE achieves state-of-the-art performance in the 2-bit quantization setting. On Qwen3-30B-A3B, it retains 95.3% of the BF16 model's average benchmark score. With our custom-designed CUDA kernel, SAGE delivers up to a 2.55-fold non-streaming end-to-end speedup over FP16. The code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.14745
PDF
-
KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache Quantization
2026-10-10 01:24:03Sihwan Park, Sung-Yub Kim, Doohyuk Jang, Eunho YangAbstract
Asynchronous rollout improves the efficiency of reinforcement learning with verifiable rewards (RLVR) by reducing under-utilization during long-tail generation. Sustained concurrency also concentrates context processing into fewer decode steps, increasing the KV workload per step. We characterize this \emph{KV concentration} and find that KV cache quantization delivers substantially greater acceleration than weight-only and weight-activation quantization in our matched-workload comparisons. However, INT4 KV rollout can degrade learning even with token-level truncated importance sampling (TIS). We propose \emph{KV4-RL}, which combines calibration-free INT4 KV quantization with selective INT8 protection of attention sinks and recent context. Protecting a bounded number of positions reduces sampler mismatch while retaining most of the speedup from INT4. Across 400 iterations of Qwen3-8B-Base GRPO training, KV4-RL achieves $1.80\times$ cumulative rollout speedup and $1.27\times$ cumulative full-iteration speedup over BF16 rollout, reaching $1.97\times$ rollout speedup over the final 50 iterations. Under a common BF16 evaluator, its final six-benchmark mean accuracy is within 0.5 percentage points of BF16-rollout training for both Qwen3-4B-Base and Qwen3-8B-Base.
Abstract (translated)
URL
https://arxiv.org/abs/2610.14448
PDF
-
ReplaySSM: Cache SSM Inputs, Not State
2026-10-09 15:59:40Ze-Wei Liou, Harris Nover, Amey Naik, Andrii Skliar, Igor Shovkun, Ido Shahaf, Tri DaoAbstract
State-space models (SSMs) maintain a fixed-size recurrent state, but reading and writing it at every decoding step can bottleneck hybrid models. This eager state management causes large state I/O with low arithmetic intensity, leaves speculative decoding no cheap rollback, and feeds state quantization error into subsequent recurrent updates. We introduce ReplaySSM, which unrolls the recurrence to represent the state as a checkpoint and a buffer of recent SSM inputs. This separates output computation from checkpoint updates. By updating the checkpoint periodically, ReplaySSM avoids state writes on most steps and roughly halves the dominant state memory traffic. Keeping recent inputs also makes speculative rollback a buffer-pointer operation and enables output-only verification without materializing intermediate states. The same separation allows a low-precision checkpoint to serve output computation on most steps while a higher-precision checkpoint serves recurrent updates, so rounding error in the low-precision copy never enters later updates. We implement ReplaySSM in vLLM for Mamba-2 and Gated DeltaNet. At serving batch size 256, it accelerates standard decoding by $1.20$--$1.47\times$ over a tuned vLLM baseline across four hybrid models, and delivers about $1.8\times$ the throughput of vLLM's speculative decoding on the two 120B-class models. Separate state precisions also reduce output KL divergence and downstream accuracy losses compared with a shared low-precision checkpoint.
Abstract (translated)
URL
https://arxiv.org/abs/2610.13778
PDF
-
DualQuant: Dual-scale Calibration-free Weight-and-Activation Quantization for LLMs
2026-10-09 15:51:38Mahsa Salmani, Sayeh Sharify, Hesham MostafaAbstract
Low-bit inference on accelerators with native 4-bit support requires quantizing both weights and activations to realize the full throughput gains these formats provide. While post-training weight quantization has been extensively studied, activation quantization remains considerably more challenging and typically requires calibration data. In this work, we introduce DUALQUANT, a dual-scale, calibration-free post-training quantization framework that directly minimizes the weight-reconstruction Mean Squared Error (MSE) and, as an immediate by-product, yields per-channel activation rescalings prior to activation quantization. Our key observation is that the scaling factors obtained by minimizing weight-reconstruction error provide effective activation rescalings, enabling competitive activation quantization without calibration data. We formulate DUALQUANT as a non-differentiable weight-reconstruction problem and develop an efficient solver based on variable splitting and block coordinate descent, where each subproblem admits a closed-form update. DUALQUANT provides a unified algorithm supporting a wide range of formats, including the full MX family and NVFP4. To our knowledge, DUALQUANT is the first dual-scaling method to formally support MX and NV block formats. Across Llama and Qwen models evaluated on WikiText-2, C4, and zero-shot downstream tasks, DUALQUANT matches or exceeds the accuracy of calibration-based SmoothQuant across most models and formats, and substantially outperforms calibration-free baselines, with large gains for MX and NV block formats. We validate these gains with a real packed W4A4 inference path for Llama-3.1-8B on an RTX 4090. The resulting implementation closely matches the accuracy of fake quantization while delivering up to 2.61x memory reduction and up to 2.46x throughput speedup over BF16.
Abstract (translated)
URL
https://arxiv.org/abs/2610.13767
PDF
-
CNN-Transformer with Sigmoid Label for High-Resolution Sub-Degree DOA Estimation
2026-10-09 11:34:23Lifeng Wang, Xintong Li, Aifei LiuAbstract
This paper addresses the grid quantization error of deep-learning (DL)-based direction-of-arrival (DOA) estimation. Classification- based DOA methods discretize the continuous angular space into a grid, so that with a peak-based read-out, the accuracy and resolution are limited by the grid step and by the width of the label peak. To address this issue, we propose a sigmoid label that encodes the sources' DOAs as the sum of angle-shifted monotonic sigmoid functions. Differentiating the output of a CNN-Transformer network trained with the sigmoid label produces sharp spectral peaks at the source DOAs. Combined with a moderately refined grid step of 0.25 degrees, the CNN-Transformer with the proposed sigmoid label attains sub-degree accuracy and a 1-degree resolution, breaking the limit of the grid quantization error. Under the unified CNN-Transformer backbone, we compare the proposed Sigmoid label with the Gaussian label, the 0-1 hard label, and the binary cross-entropy (BCE) baseline. Numerical results demonstrate that the Sigmoid label attains the highest accuracy and superior resolution compared to other labels.
Abstract (translated)
URL
https://arxiv.org/abs/2610.13433
PDF
-
Beyond In-Distribution Preservation: Recovering Generalization in Quantized VLAs via Vulnerability-Oriented Tuning
2026-10-09 10:33:09Shen Ruan, Wenchang Gao, Jin Wang, Siao Liu, Zhaxizhuoma, Dongchun Ren, Xin ZhengAbstract
Post-training quantization has been shown to preserve VLA performance under standard evaluation conditions, but whether it preserves the full-precision model's robustness and generalization remains underexplored. In this study, we systematically study the robustness and generalization of post-quantized VLA policies under environmental disturbances. Empirical results show that quantized policies can become fragile to subtle environmental variations despite retaining comparable in-distribution performance. We further observe that action discrepancies are concentrated in a small subset of rollout states, while teacher guidance has opposite effects depending on discrepancy: it improves generalization at high-discrepancy states but can degrade it at low-discrepancy states. These findings reveal that effective post-quantization recovery requires selectively intervening on vulnerable states rather than globally distilling the student. We therefore propose Policy-Induced Vulnerability-Oriented Tuning (PIVOT-Q), a vulnerability-aware On-Policy Distillation (OPD) framework that selectively corrects vulnerable states encountered during quantized-student rollouts using the frozen full-precision policy as a teacher. PIVOT-Q identifies vulnerable states using discounted accumulated discrepancies over a short horizon, applies phase-balanced sparse supervision, and uses a Behavioral Anchor to prevent unnecessary changes. Experiments under seven LIBERO-Plus environmental variations demonstrate consistent recovery across multiple VLA backbones and quantization methods. Notably, PIVOT-Q consistently outperforms full-state distillation across all settings while using only 7.4% of its state-level distillation budget. Our code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.05745
PDF
-
ECGWmark: Deep Watermarking for Electrocardiogram Signals
2026-10-09 10:32:03Kuan-Yu Chen, Kuan-Chen Wang, Yu Tsao, Jian-Jiun DingAbstract
Secure transmission of electrocardiogram (ECG) signals remains a major challenge in telecardiology. While traditional mathematical watermarking can embed patient metadata directly into ECG signals, it is highly fragile against temporal desynchronization such as circular phase shifts. In contrast, existing deep watermarking approaches for temporal signals, such as audio, typically rely on lossy quantization, which compromises the high-frequency micro-morphological details critical for clinical diagnosis. To bridge this gap, we propose ECGWmark, a fully convolutional and quantization-free deep watermarking framework for medical time-series. ECGWmark operates in a continuous, full-resolution latent space and utilizes a localized frame-level mechanism to improve the robustness to severe desynchronization. Crucially, we introduce a differentiable Physiological Loss guided by a clinical classifier to preserve diagnostic semantics. Experiments on the benchmark MIT-BIH database show that ECGWmark maintains channel-bit extraction accuracy above 90% under the evaluated circular temporal shifts. Furthermore, under a rigorous inter-patient diagnostic evaluation, Class-V sensitivity changes from 71.97% to 76.50%, whereas Macro-F1 changes from 35.25% to 34.71% and Class-S sensitivity from 8.89% to 1.39%; these results indicate mixed class-level effects.
Abstract (translated)
URL
https://arxiv.org/abs/2610.13357
PDF
-
QCLR: A Plug-in Robustification Module via Quantization for Contrastive Learning in Self-Supervised Skeleton-Based Human Action Recognition
2026-10-09 09:30:15Nanzhu Chen, Feixiang He, Yunfeng Diao, Duygu Sarikaya, He WangAbstract
Skeleton-based human action recognition has gained increasing attention due to the stability of skeleton data across diverse environments and advances in motion capture technologies. Following the success of supervised learning approaches, self-supervised learning approaches, particularly contrastive learning ones, have achieved impressive performance. However, they remain vulnerable to adversarial attacks. In this work, we propose a plug-in robustification module via quantization for contrastive learning in self-supervised skeleton-based human action recognition (QCLR), which enhances the robustness of representations against adversarial attacks while preserving their representational capability. Specifically, we employ a dual-level quantization scheme across the encoding pipeline, which breaks the gradient propagation and absorbs small adversarial perturbations within the quantization bins. To mitigate the impact of inevitable quantization error, we further design an attention-guided multi-faceted quantization strategy to maintain the representation quality. Experiments on several benchmark datasets demonstrate that integrating QCLR into a progressive lineage of mainstream contrastive learning models substantially improves robustness against both gradient-based white-box and gradient-free black-box adversarial attacks, while preserving highly competitive classification accuracy.
Abstract (translated)
URL
https://arxiv.org/abs/2610.13281
PDF
-
RAMP: Recurrent Adaptive Mixed Precision for Efficient Prefill in Recurrent-Depth Language Models
2026-10-09 05:57:33Ke Wan, Chen ChenAbstract
Recurrent-depth language models improve inference-time reasoning by repeatedly applying a shared network to refine internal representations, but the additional recurrent computation also increases inference cost. This overhead is particularly pronounced during long-context prefill, where the full input sequence is repeatedly processed across recurrent depth. Weight-and-activation quantization offers a natural way to accelerate this computation through lower-precision arithmetic, yet repeatedly applying quantized computation introduces numerical perturbations that accumulate along the recurrent trajectory. We systematically analyze how such errors propagate across recurrent depth and show that quantization errors introduced earlier are attenuated by more subsequent recurrent steps and therefore have a smaller effect on the final representation. Matched-budget experiments strongly support this prediction, showing that placing quantized computation earlier in recurrent depth consistently yields the best fidelity. Based on this insight, we introduce RAMP (Recurrent Adaptive Mixed Precision), which performs early recurrent refinement using low-precision weight-and-activation computation and restores high precision for terminal refinement. Across context lengths up to 4K, RAMP preserves near-full-precision prefill quality while achieving up to 1.16x end-to-end first-prefill speedup on an NVIDIA H100, corresponding to a 13.9% latency reduction.
Abstract (translated)
URL
https://arxiv.org/abs/2610.13041
PDF
-
VFold: Symmetry-Aware Cross-Layer Value Cache Compression
2026-10-08 17:14:28Neha Verma, Sungwon Kim, Kenton Murray, Kevin DuhAbstract
While caching key-value (KV) states accelerates Large Language Model (LLM) decoding, this cache can dominate memory usage at long context lengths. One solution is to compress this memory by exploiting inter-layer cache similarities. However, most existing techniques necessitate architectural changes to LLMs and incur substantial overhead. In this work, we propose a symmetry-aware value cache merging strategy that reduces cache memory while avoiding both harmful performance degradation and architectural overhead during decoding. Furthermore, we show that this approach can be exploited alongside existing cache compression techniques, composing with high-ratio quantization or key cache pruning to reach compression ratios that neither method reaches alone, with minimal additional cost. Ultimately, our findings reveal a major source of underutilized capacity in the value cache, offering a simple yet highly effective direction for scaling context windows under memory constraints.
Abstract (translated)
URL
https://arxiv.org/abs/2610.12338
PDF
-
DVD: Dynamic Vector Decoding for Efficient MLLM-based Perception
2026-10-08 16:32:28Jinghua Hou, Zhe Liu, Hengshuang ZhaoAbstract
Multimodal large language models have made remarkable progress in bridging vision and language, facilitating various perception tasks essential for human-machine interaction, robotics, and autonomous driving. However, existing MLLM-based perception methods predominantly rely on text-based coordinate representation, which suffers from excessive token overhead, or fixed-range quantization, which suffers from range and precision constraints, especially for 3D domains with unbounded spatial range and high localization accuracy requirements. To address these challenges, we propose a dynamic vector decoding method named DVD, which unifies the representation of 2D and 3D perception tasks. Specifically, we first transform diverse perceptual representation (i.e., 2D bounding boxes, 2D masks, and 3D bounding boxes) into 1D vector sequences, which are then mapped to compact discrete tokens in the high-dimensional space. Then, a lightweight de-tokenizer enables seamless integration with MLLMs by decoding output tokens back to original 2D and 3D perceptual representations. Extensive experiments on 2D and 3D perception benchmarks including RefCOCO series, SUN-RGBD, KITTI, Hypersim, nuScenes demonstrate that DVD achieves superior performance in 2D and 3D tasks and reduces significantly the token overhead and inference latency. DVD provides an efficient and general framework for integrating perception capabilities into MLLMs, overcoming the inherent limitations of existing methods.
Abstract (translated)
URL
https://arxiv.org/abs/2610.12266
PDF
-
Evaluating Local Language Model Agents for Reproducible Data Engineering: An Empirical Software Engineering Study of Mobility Workflows
2026-10-08 08:26:46Jorge Garc\'ia-Carrasco, Javier Sanchis, Alejandro Reina-Reina, Alejandro Mat\'e, Juan TrujilloAbstract
Context: Large language model (LLM) agents are increasingly used as software and data-engineering assistants, yet evidence about locally deployable open-weight agents remains limited. Existing evaluations often emphasize textual responses or isolated code generation rather than the validity of complete engineering artifacts. Objectives: We evaluate whether local LLM agents can produce correct and reproducible data-engineering artifacts, quantify the effect of a closed-loop workspace condition, and examine trade-offs in model scale, architecture, quantization, runtime, tool use, and failure. Methods: We introduce a benchmark of fifteen mobility-workflow tasks covering data discovery, connectors, transport-feed processing, semantic enrichment, feature engineering, validation, visualization, and reporting. Deterministic checkers assess generated scripts, tables, structured files, figures, and reports. Ten local configurations are evaluated in one-shot and closed-loop conditions, with five repetitions per model, mode, and task, yielding 1,500 scored attempts on a consumer-grade GPU. Results: Among models larger than two billion parameters, the workspace condition increases pass rates by 26.7-52.0 percentage points over one-shot generation. The strongest configuration reaches 85.3% artifact-level success, and a quantized 9-billion-parameter model reaches 69.3% with an approximately 6.5 GB memory footprint. Gains are largest when intermediate artifacts expose errors the agent can inspect and repair. Conclusion: Local open-weight agents can support a meaningful subset of software-intensive data-engineering work, but reliability depends on model capability, task verifiability, and deterministic validation. The benchmark provides a reproducible method for evaluating complete agent configurations before adoption in engineering workflows.
Abstract (translated)
URL
https://arxiv.org/abs/2610.11482
PDF
-
FlyMark: Training-Free Invisible Watermarking of 3D Gaussian Splatting via a Fruit Fly Connectome
2026-10-08 06:56:53Ziyuan Luo, Haoliang Li, Renjie WanAbstract
A trained 3D Gaussian Splatting (3DGS) scene ships as a portable parameter array that can be copied, pruned, requantized, or repackaged outside its training pipeline, so ownership evidence is most useful when it lives in the released parameters and remains checkable long after the embedding tooling is gone. Existing 3DGS watermarks typically tie embedding or extraction to scene optimization, a learned decoder, or rendered views, so the evidence survives only as long as a second trained artifact does. FlyMark instead writes a keyed message into the parameters a 3DGS file already stores. Its carrier directions are derived from the photoreceptors of a published connectome, a citable versioned artifact that fixes the geometry exhaustively and leaves nothing to tune per scene. A virtual observer reads cone-wise apparent luminance along a scene-normalized orbit from stored centers, colors, and opacities; a keyed dithered quantization-index-modulation code replicates each message bit across these observations; and one sparse bounded least-squares solve realizes the targets through achromatic shifts of existing degree-zero colors under a hard per-channel linear-RGB bound. All geometry and higher-order appearance parameters are preserved bit-identically, and extraction needs only cone queries, rounding, and majority voting. Under a model-domain threat model on synthetic and real scenes, FlyMark attains high clean bit accuracy and visual fidelity while cleanly separating matched from wrong keys.
Abstract (translated)
URL
https://arxiv.org/abs/2610.11364
PDF
-
Deflating the Hessian: Rank-4 W4A4 Quantization for Multimodal Diffusion Transformers
2026-10-08 06:22:19Shiwen Wang, Pengxiang Zhao, Xiaoming YuanAbstract
In diffusion transformers, low-rank branches can mitigate 4-bit weight--activation (W4A4) post-training quantization (PTQ) loss by decomposing each weight into a low-bit residual and a high-precision low-rank component. Existing low-rank PTQ approaches, however, either optimize low-rank compensation and residual quantization separately, often requiring higher ranks, or rely on second-order weight updates without explicitly modeling activation quantization error, which becomes particularly pronounced under 4-bit quantization. To address these limitations, we present \method{}, a unified framework modeling low-rank-assisted W4A4 PTQ as a coupled calibration problem and deriving optimization-based solvers from the joint objective. Eliminating the output-side low-rank factor yields a \emph{deflated Hessian} that discounts residual errors already captured by the low-rank component, while an activation-noise surrogate is incorporated to suppress activation quantization error. Across five diffusion backbones, rank-4 \method{} consistently outperforms rank-4 SVDQuant in PSNR and LPIPS. It further surpasses rank-32 SVDQuant on SANA-1.6B, FLUX.1-schnell, and FLUX.1-dev with an $8\times$ smaller rank and up to $6.25\times$ faster quantization. Furthermore, on the Qwen3-8B LLM, rank-4 \method{} improves MMLU accuracy from 61.50\% to 68.17\% over rank-32 SVDQuant. Overall, \method{} achieves better W4A4 performance with substantially lower rank and quantization cost.
Abstract (translated)
URL
https://arxiv.org/abs/2610.11315
PDF
-
Read What Matters: Query-Adaptive Quantization for KV Caches
2026-10-08 04:47:41Siddharth Bhandari, Lucas Gretta, Krishna Balasubramanian, Shiva KasiviswanathanAbstract
KV-cache entries are stored before their future queries are known, but each decoding query needs precision in different places. We study this mismatch using separate budgets for retained bits and bits fetched per query. ReadKV stores each key and value in a progressive code whose prefixes support different reconstruction precisions. For each query, it allocates key-channel prefixes using the query, computes attention from the reconstructed keys, and then allocates value-token prefixes using that attention. Stored entries remain unchanged. Each stage optimizes a calibrated distortion objective under a fixed budget; we prove exact allocation under diminishing refinement gains and relate these objectives to attention-output error. We also exhibit a finite-dimensional attention family where query-dependent access strictly outperforms every query-independent reader at the same read budget, even with unrestricted competing encoders and decoders. Across six base models, reading four bits on average from an eight-bit cache increases C4 perplexity by at most 0.66%, using about one quarter of the logical reads and half the retained capacity of a 16-bit cache. It is consistently more accurate than storing and fully reading four bits at the same payload-read budget. Retaining more bits than each query fetches is aimed at long-context decoding, where the cache bytes moved per step, rather than the weights, dominate cost. Long-context question answering and retrieval on two instruction-tuned models provide additional quality evidence. On the tested 8K-token, batch-one, single-layer workload on an NVIDIA A10G, a restricted eight-bit ReadKV reader with a two-bit mean payload-read budget has 39% lower latency than the tested TurboQuant codec.
Abstract (translated)
URL
https://arxiv.org/abs/2610.11245
PDF
-
When Lower Reconstruction Loss Hurts: Distributionally Robust Refinement for Low-Bit LLM Quantization
2026-10-08 04:22:29Yanlong Zhao, Xiaoyuan Cheng, Huihang Liu, Baihua He, Xinyu Zhang, Harrison Bo Hua Zhu, Wenlong Chen, Li Zeng, Zhuo SunAbstract
Weight-only post-training quantization (PTQ) relies heavily on reconstruction loss minimization to preserve model quality at low precision. We show that the weights favored by minimizing this loss need not yield better model performance on new tasks. In fact, we find that lower reconstruction loss can even degrade model performance on the same calibration data. Our analysis further shows that weights with lower reconstruction loss on calibration data can have higher loss than other weights when the distribution of input activations changes. Motivated by these observations and our analysis, we propose Distributionally Robust Quantization (DRQ), a post-hoc refinement process that minimizes worst-case reconstruction loss over a constrained set of input activation distributions. DRQ refines the integer codes representing quantized weights within the existing quantization grid, keeping quantization parameters and inference operators unchanged. Extensive experiments show that DRQ improves models quantized by six representative PTQ methods, including AWQ, GPTQ, and ParoQuant, and delivers gains across both dense and mixture-of-experts large language models. These results establish DRQ as a general post-hoc refinement framework for weight-only PTQ, achieving better downstream performance without adding inference overhead.
Abstract (translated)
URL
https://arxiv.org/abs/2610.11226
PDF
-
Bridging KV-Cache Quantization and Linear Attention: From Theory to Pretrained Weight Migration
2026-10-08 04:12:49Kaicheng Xiao, Liran Dong, Haotian Li, Guoliang XingAbstract
KV-cache quantization and linear attention are two representative approaches to tackling the storage and computational costs of Transformers. KV-cache quantization compresses individual KV entries into discrete codes but retains all entries, whereas linear attention recurrently aggregates multiple historical KV contributions into a fixed-size continuous state but can introduce interference. This contrast raises the question of whether per-KV compression and multi-KV aggregation can be bridged within a single mechanism for efficient attention. We identify RAM-Net as such a bridge through soft assignments over a discrete address space. These assignments determine recurrent updates to the continuous slot state associated with each address. Under a restricted RAM-Net construction, we prove that soft address assignments extend hard quantized matching to a separable read-write overlap that locally approximates full-attention similarity and supports recurrent aggregation. These connections further enable Transformer-to-RAM-Net weight migration through a new path based on a soft-quantized intermediate construction. Across nine pretrained Transformer models from 0.3B to 7B parameters, RAM-Net recovers an average of 87.1% of the teachers' accuracy gains over random guessing across six commonsense and knowledge tasks using only a 500M-token budget per model.
Abstract (translated)
URL
https://arxiv.org/abs/2610.11214
PDF
