Quantization
-
ClunoGS: Compact Language 3D Gaussian Splatting via Unified Optimization
2026-10-21 16:43:53Jian Xu, Yi Li, Ying Huang, Siyi Wu, Bingzhe Li, Yuede Ji, Miao YinAbstract
Language 3D Gaussian Splatting enables open-vocabulary scene understanding by associating Gaussian primitives with language features. These semantic attributes increase model storage, while most existing 3DGS compression methods are designed mainly to preserve RGB rendering quality. One straightforward approach is to apply sparsification and quantization sequentially. However, pruning based only on rendering quality can remove Gaussians that have small RGB contributions but are important for semantic prediction. Once removed, their semantic information cannot be recovered by later quantization. We propose ClunoGS, a compact framework for Language 3DGS that combines semantics-aware Gaussian sparsification and vector quantization in a unified constrained objective. Semantic supervision is used during sparsification, and the objective jointly considers rendering quality, semantic accuracy, and compactness. We optimize the unified objective with alternating primal, sparsification, quantization, and dual updates. We evaluate ClunoGS on open-vocabulary localization and segmentation using the LERF and ScanNet-10 benchmarks. Under aggressive compression, our unified, semantics-aware formulation substantially alleviates the degradation of semantic representations caused by sequential compression and semantics-agnostic Gaussian sparsification. Ablations further show that jointly coordinating sparsification and quantization better preserves semantic accuracy than applying them independently.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30013
PDF
-
Bio-Inspired Auditory Processing: Coupling Parallel Resonate-and-Fire Dynamics with Ternary Recurrent Networks for Speech Recognition
2026-10-21 13:29:46Pavlos Poulos, Katelyn Sadorf, Themos Stafylakis, Costas A. AnastassiouAbstract
Spiking Neural Networks (SNNs) promise energy-efficient inference through sparse, event-driven computation, yet remain constrained on continuous, context-dependent tasks such as automatic speech recognition (ASR). A key bottleneck is the Leaky Integrate-and-Fire (LIF) neuron, whose membrane decay erases information across pauses and phonetic boundaries while it must both extract features and retain long-range context. Inspired by cortical separation of nonlinear filtering and somatic integration, we introduce the Spiking Ternary GRU (stGRU): a ternary-weight gated recurrent unit performs temporal filtering, while downstream Resonate-and-Fire (RF) neurons generate the binary spike stream driving the next block's event pathway. This makes binary event transmission a native training constraint rather than a separate deployment-time activation quantization step; together with ternary weights, learned projections avoid dense floating-point matrix MACs. Trained end-to-end on LibriSpeech-960, Small, Medium, and Large stGRUs reach 8.09%, 7.04%, and $6.44 \pm 0.17\%$ WER on test-clean without an external language model. To our knowledge, these are the first matrix-MAC-free spiking networks without a learned ANN sequence decoder to achieve single-digit WER on LibriSpeech. The architecture also achieves 10.88% test CER on Aishell-1 (9.28% on dev) when trained from scratch. Internally, trained networks develop a depth-graded hierarchy: shallow layers act as fast acoustic filters, while deep layers form slower semantic integrators whose latent representations segregate into part-of-speech manifolds and word-level spike events. Deep-layer firing rates follow lognormal distributions with a sparsity gradient, mirroring hierarchical processing observed in human cortex in vivo. Our biologically grounded architecture combines competitive ASR with favorable modeled compute and memory cost.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29720
PDF
-
Efficient Generative Image Compression with Quantized Latents Refinement
2026-10-21 13:03:09Hao Cao, Yinhuan Huang, Wenqi Guo, Zhijin Qin, Jungong HanAbstract
Generative image compression (GIC) delivers strong perceptual quality at low bitrates. However, GIC methods with strong perceptual quality often rely on pretrained diffusion models with high latency, while recent lightweight GIC methods reduce latency but deliver lower perceptual quality. We observe that, for lightweight GIC methods at low bitrates, coarse deterministic quantization can collapse nearby encoder latents corresponding to different fine details to the same point, making these details difficult to recover. Motivated by this, we propose GR-GIC, an efficient GIC that delivers competitive reconstruction performance with fast inference. For better performance, we introduce Guided Refinement (GR), which combines latent refinement guided by entropy context with adaptive perturbations for training regularization, jointly improving reconstruction at low bitrates. For fast inference, we introduce reorganized projected residual vector quantization (P-RVQ), which equivalently reorganizes residual updates into low-dimensional lookups to expose greater parallelism and reduce sequential computation. Experiments on Kodak show that for performance, compared with OneDC, GR-GIC achieves BD-rate savings of 23.19%, 37.71%, 22.30%, and 22.14% for LPIPS, DISTS, PSNR, and FID, respectively. For fast inference, the highly optimized implementation of GR-GIC encodes and decodes in 4.60 and 5.98 ms on an NVIDIA A100. Specifically, the reorganized P-RVQ implementation achieves a $3.14\times$ average speedup over the original implementation under the same FP32 setting. On an NVIDIA Jetson Orin NX, GR-GIC achieves latency comparable to lightweight GIC methods.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29691
PDF
-
Effects of Model Scale and Quantization on Social Bias and Abstention in Question Answering
2026-10-21 12:38:04Kuniko Paxton, Ifeoluwa Wuraola, Dhavalkumar ThakkerAbstract
The deployment of small and quantized language models in resource-constrained settings raises an important fairness question: when a compressed model produces fewer biased responses, does this reflect fairer behavior or greater abstention? We study this problem in Question Answering (QA) using the BBQ benchmark across multiple model sizes and quantization settings from Qwen3, Gemma-3, and SmolLM2. We also examine whether the observed trends remain stable across alternative prompt formulations and when the ``Unknown'' option is removed in a forced-choice setting. Our results show that larger models achieved higher QA accuracy, but bias showed no simple scaling trend and quantization effects varied across configurations. Notably, 20 of the 26 configurations with fewer biased responses also show increased over-abstention, indicating that lower observed bias can partly reflect greater abstention. These findings show that lower observed bias can partly reflect greater abstention, and that fairness evaluation should consider biased responses, abstention, prompt formulation, and answer-option design jointly.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29657
PDF
-
Joint Semantic-Channel Coding and Modulation for Generative Image Transmission
2026-10-21 12:25:37Hao Cao, Jingkai Ying, Yinhuan Huang, Zhijin Qin, Yuan ShenAbstract
Digital image transmission typically optimizes source coding, channel coding, and modulation separately, limiting how source representations adapt to channel impairments. We propose SemQAM, a joint semantic-channel coding and modulation scheme for generative image transmission converting source to modulated symbols, such as standard QAM symbols. Channel impairments create two challenges: they can corrupt the latent representation at the receiver, and this corruption can compound through autoregressive coding, degrading reconstruction quality. First, SemQAM represents latent features with vector quantization (VQ), maps each VQ index to a QAM symbol, and updates the codeword corresponding to the detected symbol so that channel impairments directly affect codebook learning. Moreover, SemQAM uses the conditional expectation of received codewords to approximate the receiver reconstruction and reduce error propagation during autoregressive coding. SemQAM supports multiple SNRs and CBRs in a unified model, while also supporting conventional source coding through SemQAM-C. Experiments show that (i) under the same LDPC and QAM settings, SemQAM-C outperforms OneDC in LPIPS and DISTS with substantially lower complexity, showing that the unified model retains competitive capability for source coding. (ii) Compared with this separated counterpart, SemQAM reduces channel uses by 18.57% at matched LPIPS and outperforms the compared JSCC methods, demonstrating the benefit of joint coding.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29643
PDF
-
TernScope: Model-Hardware Co-Design for Efficient Ternary Endoscopic Segmentation
2026-10-21 11:56:23Zhuoyu Wu, Wenhui Ou, Qiawei Zheng, Xuhong Qiu, Wenqi Fang, Rapha\"{e}l C. -W. PhanAbstract
Edge medical segmentation must reconcile small model size, high accuracy, and fast inference. TernScope addresses this tradeoff through model--hardware co-design, learning a compact segmentation network whose low-bit operators also simplify hardware execution. Split-channel large-kernel blocks preserve spatial context with limited computation, while branch fusion removes redundant inference paths. Ternary Progressive Quantization-aware Training (TPQ) then adapts the fused network from deep to shallow layers, through the decoder, and finally to the input and prediction layers. All 25 convolutions use ternary weights and 8-bit activations, enabling packed storage and multiplier-free computation. The resulting model uses 0.2483 M parameters and 0.0949 GMACs while retaining competitive segmentation accuracy on in-domain and external datasets. Under the same 45 nm technology, TernScope reduces compute-array area by 88.46% compared with the original Gemmini, a systolic-array accelerator with 8-bit $\times$ 8-bit multipliers. These results demonstrate a path from accurate lightweight medical segmentation to compact low-bit hardware.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29617
PDF
-
PACE: Periodic and Channel-Aligned Encoding of Four-Bit Recurrent States
2026-10-21 10:31:52Patrick Haller, Filipe Laitenberger, Alan AkbikAbstract
Linear-attention language models keep their context in a recurrent state of fixed size, where an inference system stores one copy of this state for every sequence it decodes at the same time. Storing the state at lower precision therefore lets more sequences share an accelerator. Existing state quantizers round the state after every generated token and report that four bits lose accuracy unless the bit width is raised or the model is fine-tuned.\par We introduce PACE, a post-hoc four-bit state quantizer. PACE rounds the state only once every few generated tokens, as chunked kernels already do once per chunk during prefill, and keeps the tokens generated in between in a small buffer. Rounding less often reduces the error that accumulates in the state. Additionally, we find that the large entries of the gated-delta state gather in columns, because some channels of the update written at each token remain large across many tokens. PACE therefore shares each quantization scale within one column. It also keeps the heads that retain rounding error longest at eight bits. On four hybrid models from 4B to 35B parameters, PACE with mixed four- and eight-bit heads is competitive with eight-bit DAMP on the three larger models at about two thirds of its memory, PACE with four-bit heads throughout scores higher than Q-Mamba at equal memory, and rounding every few tokens preserves the decoding speedup of the buffer.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29545
PDF
-
Right On Q: Recovering Reasoning under Quantization via Self-Generated Trajectories
2026-10-21 08:03:24Boyuan Zheng, Jing Xiong, Chenwei Wu, Hui Shen, Zesen Zhao, Hei Ting Una Chan, Xueshen Liu, Zhen Zhang, Liyue Shen, Zechun Liu, Z. Morley MaoAbstract
Quantization-aware training (QAT) aims to recover reasoning capabilities in low-bit large language models (LLMs). Existing recipes typically supervise the quantized student with the original full-precision model on fixed response sequences. However, the resulting low-bit quantized models still struggle on long-horizon reasoning tasks and often fail to terminate within a token budget that suffices for the full-precision model. We identify two challenges in closing this gap. First, fixed-response supervision leaves a mismatch between training prefixes and those encountered during student generation, and this trajectory divergence grows as responses lengthen. Second, on its own prefixes, the student underpredicts some next-token candidates and overpredicts others, and this distributional misalignment becomes more frequent at high-entropy prefixes. We present OnQ, a low-bit recovery pipeline that combines on-policy distillation and candidate-wise Dynamic KL. Starting from block-wise calibration, the quantized student generates responses and receives full-precision teacher supervision on its own prefixes. Dynamic KL weights forward and reverse KL by each candidate's teacher-student probability ratio to balance correction of under- and overprediction. Across Qwen3-1.7B, 4B, and 8B at 2- and 3-bit precision, OnQ achieves the highest nine-benchmark average, with the largest gains on long-horizon reasoning benchmarks. At 2 bits, it outperforms every baseline on every benchmark at every model size, improving the Qwen3-8B average by 10.26 points over the strongest baseline. It also produces shorter responses with fewer truncations and repetitive errors.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29378
PDF
-
Compressing Onboard Language Models for Space Robots under a Behavioral-Fidelity Requirement
2026-10-21 06:59:07Jia Deng, Zizhe Wang, Xin Lu, Yi Huang, Zehao JiaAbstract
Language models are being explored as the interface through which space robots interpret operator commands and select among verified skills. Running them onboard calls for low-bit compression, and flight use adds requirements that generic post-training quantization leaves implicit: a fixed memory budget and storage format, a static artifact under configuration control, and behavior close to the full-precision model validated on the ground. We treat compression as a ground-segment step and study QSelect, which composes one static low-bit model from compatible AWQ, GPTQ, and AutoRound checkpoints by assigning each linear module to one quantizer family and scoring every substitution at the final next-token distribution. The default fidelity mode needs no labels or tuned thresholds and accepts a substitution only when the forward KL divergence from the BF16 reference decreases; an optional target-adapted mode adds answer likelihood on a small labeled set. On DeepSeek-R1-Distill-Qwen-7B, Llama-3.1-8B, and Llama-3.2-3B at 3 and 4 bits, composed models reach lower mean held-out perplexity than every single-family model in all six settings, and the fidelity mode lowers perplexity and KL further. In a matched scaffold, final-output scoring gives lower held-out KL than representation-similarity scoring in all 12 paired runs. With labels from training or development splits of the evaluated task families, the target-adapted mode raises 3-bit pooled accuracy on the two Llama models by 4.3 and 4.5 points over the best single-family model. At 3 bits, unquantized embedding and output layers make up 43-46% of the packed artifact. All evaluations use general-domain benchmarks on a ground GPU; space-domain tasks and flight hardware are left to future work.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29302
PDF
-
PaceVLA: Adaptive Temporal Action Quantization in Vision-Language-Action Models
2026-10-21 05:55:50Hojin Jeon, Jimin Lee, Taekwan Kim, Nayoung Oh, Taeyoung Kim, Myungkyu Koo, Jinmo Koo, Joonwoo Ahn, Jinwoo ShinAbstract
Fast and reliable manipulation is crucial for practical robot deployment. Although Vision-Language-Action (VLA) models have recently shown remarkable progress as generalist robot policies, their execution often remains temporally inefficient. A key source of this inefficiency is that policies learn fine-grained temporal patterns from teleoperated demonstrations, requiring many low-level action steps. Temporal action quantization offers a simple way to accelerate execution by aggregating consecutive low-level actions into temporally coarser actions. However, applying action quantization without considering when fine-grained control is required introduces a speed-accuracy trade-off: accelerated actions may efficiently reduce redundant steps in some phases, while discarding critical motion details in precision-sensitive interactions. To address this, we propose PaceVLA, a unified framework for adaptive temporal execution in VLAs that integrates two coupled components: (1) mode-specific action decoders built on a shared action expert, supporting multiple temporal granularities and prediction horizons, and (2) a quantizability-guided mode selection scheme which decides whether coarse execution is appropriate, and a context-conditioned router that selects an execution mode at each inference step. Experiments across three multi-task simulation benchmarks and six real-world tasks show that PaceVLA improves execution efficiency while maintaining task success rate across multiple embodiments, including long-horizon and dexterous manipulation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.25961
PDF
-
Vector-Quantized World Model: Latent Feedback Motion Planning for Robot Manipulation
2026-10-20 20:08:37Xuezhou Xu, Chung-Pang Wang, Zhenzhen Li, Benjamin RiviereAbstract
Visual world models have shown strong potential for robot planning, yet long-horizon manipulation remains challenging due to compounding prediction errors, large search complexity, and latent plans that the low-level controller cannot reliably reach. In this work, we introduce Vector-Quantized World Model (VQ-WM), a latent feedback motion planning framework that integrates discrete long-horizon search with continuous visual feedback control: a discrete latent predictor models transitions between vector-quantized latent states, allowing the planner to look far ahead and find a shortest path to the goal through discrete search and a continuous learned controller sequentially reaches the planned states through visual feedback, correcting deviations to mitigate compounding errors. Whereas most vector quantization uses Euclidean distance, we use a novel estimator of the first-passage-time as a quasi-metric, which aligns the planning abstraction with the controller's capabilities. We evaluate on the physical robot manipulation of the Tower of Hanoi puzzle: whereas existing state-of-the-art foundation models (VLAs, WAMs) only succeed when the solution trace in their training dataset, our method achieves 100\% success rate with only short-play data, demonstrate its ability to compose long-horizon solutions beyond the short demonstrated trajectories and execute them reliably. Our solution also outperforms strongest world-models by 13.3--85.3 percentage points on simulation benchmarks spanning manipulation and locomotion, demonstrating consistent progress and new capabilities.
Abstract (translated)
URL
https://arxiv.org/abs/2610.25600
PDF
-
Beyond 4 bits: Data-Centric Post-training Quantization and Recovery
2026-10-20 17:10:04Rhea Sanjay Sukthanker, Riccardo Grazzi, James Hensman, Pashmina CameronAbstract
Post-training quantization (PTQ) compresses the weights of a trained model with limited compute and data. Down to 4 bits it is near-lossless, but at 2 bits scalar PTQ configurations fall to chance-level accuracy. We study three choices usually treated as implementation details that matter little at 4 bits but decide 2-bit quality: the quantization representation, the calibration data, and the recovery fine-tuning data, across six dense and mixture-of experts models (ranging from $3.8$ to $32$B parameters). Representation is the largest factor and vector codebooks restore usable models where scalar PTQ pipelines collapse. Most of the remaining loss in accuracy is on \emph{generative} tasks, and the amount of performance drop is model-dependent. Furthermore, at a short recovery budget, calibration and recovery data are coupled. The best recovery data depends on the calibration data, so the two must be chosen jointly. The resulting recipe involves a vector codebook representation, a small task-aligned calibration mixture, and an inexpensive codebook-only recovery stage. Including codebooks, it uses about $2.3$ effective bits per weight and retains $89$ to $97\%$ of full-precision question-answering (QA) accuracy where scalar PTQ is at chance, and, including recovery, uses $1.4$ to $30\times$ less compute than the strongest quantization-aware-training (QAT) baseline.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28537
PDF
-
Local Inference of LLMs on Long Contexts: A Data Systems Perspective
2026-10-20 13:56:12Mattias D\"urrmeier, Alberto LernerAbstract
Large language models can solve complex problems across a variety of subjects, such as mathematics and coding. For such complex analytical tasks, models with a long context perform better. Serving models with long context is currently mostly done on powerful systems with multiple GPUs, which are too expensive for most users. An alternative is local inference, or serving a model on a user's own machine, which removes per-token costs and rate limits, besides keeping sensitive data private and giving users full control over the models they run. The common understanding, however, is that local LLMs need compression (e.g., quantization) to run on a single GPU. In this paper, we set out to study the limitations that prevent serving long-context models for single users, besides the capacity issue. Rather than characterizing a single model on one GPU, we analyze how different models behave across several GPU generations. This lets us uncover not just individual bottlenecks but broader trends in where inefficiencies arise. Our analysis confirms that, as expected, self-attention dominates the runtime for long-context inference. More unexpectedly, our evaluation reveals that the current structure of self-attention leaves key GPU resources significantly underutilized, and the newest GPU is often the least well served. Many of these findings are fundamentally data management problems, from paging the KV cache to improving cache utilization and synchronization in the GPU, which is why we bring this study to the Data Systems community. To engage this community, we present our findings as a series of experimental observations and formulate recommendations for future work.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28223
PDF
-
ACE-Compression: An Ability-Centric Evaluation of Compressed Large Language Models
2026-10-20 09:26:20Jiahui Ning, Liang Chen, Yuchen Liang, Yi Zhang, Fenglei FanAbstract
Model compression makes large language models deployable, but today's evaluation tells an incomplete story. Standard practice measures retained task performance and stops there. This misses a critical fact: compression perturbs the distributed knowledge embedded across billions of weights, and task scores alone cannot reveal which underlying abilities survive and which quietly collapse. To solve this question, we introduce ACE-Compression, an ability-centric evaluation framework. Instead of asking only "does the compressed model still answer correctly?", we ask "which of its core abilities remain intact?" Our framework probes four deployment-facing dimensions: procedural reasoning and constraint execution, knowledge grounding and truthfulness, social cognition and affective understanding, and persona fidelity and stylistic consistency. Across more than 35 compressed checkpoints from six open-weight LLMs-covering quantization and pruning-we find that compression redistributes ability profiles. Coarse, local, and surface-level signals often persist, while fine-grained, long-tail, and context-dependent abilities become fragile. Quantization consistently outperforms pruning in retention, but the practical meaning of that gap depends entirely on which ability an application requires. Based on our findings, deployers can choose compression methods by the ability profile their application needs, not just by a leaderboard average. And algorithm designers gain diagnostic signals to build compression methods that preserve what matters. We believe that ACE-Compression turns compression from a black-box accuracy drop into an interpretable intervention-one we can measure, understand, and ultimately improve.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27913
PDF
-
E8-QAT: Efficient 2-Bit Quantization-Aware Training with Implicit Lattice Codebooks
2026-10-20 06:17:09Xinyu Lei, Hong Huang, Jiebin Zhang, Hongyu Li, Jinlong Song, Decheng Wu, Guanghua Yu, Jianchen Zhu, Sujian LiAbstract
Low-bit quantization reduces the memory footprint of large language models, but preserving accuracy with efficient training remains challenging. Scalar quantization encodes weights independently, while explicit-codebook vector quantization introduces costly search during training. We propose E8-QAT, a 2-bit quantization-aware training framework that uses the $E_8$ lattice as an implicit eight-dimensional codebook. It jointly quantizes weight groups through arithmetic projection without storing or searching an explicit codebook during the training forward pass. A 2-bit core encoding and sparse extension preserve the selected lattice points exactly, adding less than 0.1\% to core index storage. On Qwen3.6-35B-A3B, E8-QAT matches the performance of the full-precision model in terms of the average score across the four tasks. E8-QAT improves over the state-of-the-art baseline by 2.03\% on IFEval and by 1.51\% on LiveCodeBench, under the same data, schedule, and bit width. The 2-bit representation reduces peak memory by 80\% and supports a $4\times$ larger batch. At each representation's largest feasible batch, E8-QAT reaches $1.97\times$ the decoding throughput of BF16.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27699
PDF
-
Token-Set Choice Confounds POPE: A Systematic Audit of Yes/No Extraction in VLM Hallucination Evaluation
2026-10-20 04:43:52Kesav Kumar Jayakumar, Karthigeyan ThilakAbstract
We set out to reproduce a recent hallucination-mitigation result on the POPE benchmark. The LLaVA-1.5-7B baseline we measured, 82.21 F1 on the adversarial split, sat 1.21 to 8.67 points above the adversarial baselines we planned to compare against (81.00, 79.30, and 73.54). We looked at the evaluation code and found a readout choice that moves F1 on its own. A two-token logit readout decides yes versus no by comparing two vocabulary IDs, 3582 ("yes") and 1217 ("no"). Across 9,000 greedy-decode questions spanning all three splits, neither ID is generated even once. The model emits " Yes" (token 3869) and " No" (token 1939), because SentencePiece prepends a space byte after the prompt suffix "ASSISTANT:". Read with an eight-token lookup that covers the space-prefixed surface forms, LLaVA-1.5-7B scores F1 0.8221, 0.8498, and 0.8713 on the adversarial, popular, and random splits. That 6.13-point adversarial shift exceeds the adversarial gain reported by five of the six papers we surveyed. None of them states its readout, and our reference ran under 4-bit NF4 quantization, so we cannot say how much of any reported gain a matched evaluation would keep. That comparison stays unresolved until readout and precision are matched. On the fixed baseline we run nine inference-time corrections of our own (none beat it) and dissect why the adversarial split resists logit-level intervention. In a 3,000-question probe, wrong predictions draw more image attention than correct ones (Cohen's $d=-0.46$, $p=1.4\times10^{-24}$), the opposite of what grounding-based corrections assume. A four-readout audit across four further VLMs from three tokenizer families confirms the confound is protocol-level, not model-specific: porting the LLaVA-1.5 IDs verbatim drops F1 to 0.00 on the two disjoint-vocabulary models, whereas a tokenizer-derived readout holds between 0.80 and 0.92.
Abstract (translated)
URL
https://arxiv.org/abs/2610.25964
PDF
-
Steering intonational tunes in text-to-speech synthesis
2026-10-19 23:51:11Juan Pablo Vigneaux, Jennifer ColeAbstract
Text-to-speech systems built on residual vector quantization often generate audio in two stages: a transformer backbone emits the first level of each frame's token stack, and a smaller depth model generates the rest autoregressively. We ask which components of such a model are responsible for prosodic intonation, more precisely for generating the English nuclear tune (for instance, the falling pitch melody that distinguishes an assertion from a question's rising melody), and whether interventions on those components can be used to control intonation. Prompt-based control mostly fails; only a final question mark is a reliable lever. Interchange interventions in CSM and S2 locate the tune: the backbone's hidden state carries it to the waveform through the depth model, and the first level of the stack carries little or none of it. Steering the backbone state with a difference-of-means vector computed on new material generated with a final period or question mark moves the tune, but the push toward an assertion degrades the voice. Steering vectors computed instead from the backbone states associated with human imitations of theoretically-specified tunes, whose text carries no punctuation, and applied inside the generation loop, move most recordings to the other condition in both systems (even when fitted without the speaker being steered), at a small cost in predicted naturalness. These imitation-derived vectors are aligned with the one obtained from the two punctuated conditions far beyond a shuffled-label null, so the backbone state of a punctuated prompt already represents intonation. The steered recordings broadly become falling or rising, according to the intent, but the edge tone moves more coherently than the pitch accent, and the target tune cannot be selected precisely beyond this broad tendency.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27403
PDF
-
When PTQ Calibration Helps or Hurts: Dose-Driven Collapse and Validation-Gated Abstention in Quantized Text Embedders
2026-10-19 23:43:39Hyojung Han, Ryangkyung Kang, Youngeun Kim, Seung-Hun JeonAbstract
Learned post-training quantization (PTQ) calibration fits quantizer parameters on a small calibration set, and for a text embedder the natural supervision is retrieval relevance. We ask two questions that are usually conflated: how far to calibrate, and whether to calibrate at all. Across three embedders (e5-base-v2, BGE-M3, EmbeddingGemma-300m) at 4-bit weight quantization, 161 calibration runs, the outcome of a fixed procedure ranges from a gain to catastrophic degradation, and the driver is the dose of the update, measured as the functional displacement D_func of the quantized model from its uncalibrated start. On EmbeddingGemma the same learning-rate grid gives +1.47 to -71.53 points of in-domain NDCG@10 change; a label-free cosine objective collapses at the same dose (-69.52 vs -71.63), so the collapse is a property of learnable calibration rather than of retrieval supervision. Objective sensitivity is heterogeneous: on e5, ranking calibration is substantially more robust than cosine at matched displacement (+14.16 +/- 4.34 pp, 95% t-CI [+3.37, +24.96], three paired slates), whereas on BGE-M3 the ordering is directionally opposite but unresolved ([-5.18, +2.38]). Dose is the main axis but not a sufficient statistic: equal D_func reached along different paths differs by up to 1.61 pp. We therefore treat calibration as an optional intervention. A pre-registered rule that accepts a checkpoint only when a per-query paired confidence interval on a 25% calibration-validation split clears the uncalibrated baseline was run on 6 independent slates and five arms. It never selected a checkpoint below its slate's baseline (0 harmful selections in 30 decisions), retained 8 of 11 beneficial updates, and paid a mean oracle regret of 1.52 pp with one abstention that forfeited 11.13 pp, a safety-opportunity trade-off, not a free improvement. We report what remains open, including label efficiency.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27397
PDF
-
PerQ: Inverse Generative Modeling for Neural Image Compression via Quantization Error Compensation
2026-10-19 22:57:59Tianhao Peng, Ho Man Kwan, Fan Zhang, Shan Liu, Dave BullAbstract
Neural image codecs face a fundamental trade-off between pixel fidelity and perceptual quality. Distortion-optimized codecs preserve pixel-wise fidelity but produce blurred reconstructions at low bitrates, whereas generative codecs leverage learned priors to improve perceptual quality but tend to exhibit saturating PSNR at higher bitrates and require separate models for different bitrate targets. We propose PerQ, which reframes perceptual reconstruction as inverse generative modeling with bounded support. Specifically, we define the quantization cell as the set of latents that yield the same transmitted symbols and derive a transport bound showing that reconstructions from this cell remain within a rate-determined radius of the codec output. This motivates performing generative reconstruction within the codec-induced quantization cell, rather than sampling from an unconstrained image prior. PerQ implements this principle by attaching a rate-conditioned flow-matching compensator to a frozen, multi-rate distortion-optimized backbone, enabling perceptual enhancement while preserving the backbone's coding performance. Experiments on Kodak and CLIC2020 show that PerQ achieves perceptual performance comparable to generative codecs such as MS-ILLM in LPIPS and FID, while maintaining competitive PSNR relative to distortion-optimized codecs such as ELIC. A single model checkpoint supports a wide bitrate range with fast decoding, and the compensator provides consistent gains across three different backbones. The implementation is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27374
PDF
-
Channel-Level Mask Guidance for Slimmable Image Compression
2026-10-19 17:19:04Chen Chen, Chenggang Wu, Yongbiao Chen, Weidong Qiu, Zhengwei Qi, Haibing GuanAbstract
Learned image compression (LIC) has achieved promising rate-distortion performance, but most methods lack scalability for diverse platforms with varying computational and storage constraints. Existing slimmable LIC methods achieve flexible complexity via consecutive-channel activation but suffer from two limitations: forced consecutive-channel selection degrades narrow-model performance at low bitrates, and the lack of stable variable-rate control limits adaptability to varying bandwidth. We propose a channel-level mask-guided slimmable image compression method. A learnable channel-wise mask with sparsity-promoting regularization enables non-contiguous top-w channel selection, letting narrow models use the most informative channels. After training, mask merging and channel reordering convert the model into a standard slimmable network via Mask-to-Slimmable (M2S) conversion, enabling efficient inference without architectural changes. For variable-rate control, we combine lambda- and width-conditioned modulation with quantization-step scaling and a progressive training pipeline stabilized by a freeze-then-fine-tune strategy. Experiments on Kodak across E2E, Hyperprior, and Checkerboard backbones show superior rate-distortion performance across widths and bitrates, with large gains for narrow models at low bitrates. At w=48, it achieves up to 19.22%, 24.11%, and 22.76% BD-rate savings over the same-width Slimmable baseline on E2E, Hyperprior, and CKBD, respectively. A single VBR model replaces 40 independent models, reducing storage by 14.8x to 15.9x. Code: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27037
PDF
