Quantization
-
Efficient Quantization-Aware Distillation with Cross-Modal Alignment for Edge Vision-Language Models
2026-09-15 06:12:54Jinwoo Jeon, GyuYeop Do, Yubin Lim, Nam-Joon Kim, Hyun Gon Ryu, Hyuk-Jae Lee, Byung-Jun LeeAbstract
Large-scale vision-language models (VLM) such as CLIP enable strong open-vocabulary reasoning, yet deploying these capabilities on resource-constrained edge devices remains challenging. EdgeVL addresses this problem by distilling CLIP representations into lightweight multi-modal encoders and applying quantization-aware training (QAT) for efficient Open-Vocabulary Classification (OVC) on edge hardware. However, its two-stage optimization applies different objectives for distillation and QAT, and contrastive learning is performed within the quantized student space, which can result in inconsistent optimization and reduced training efficiency. Moreover, identical supervision across RGB and non-RGB modalities may lead to modality imbalance. We propose a unified framework for quantized semantic distillation tailored to edge deployment. By jointly optimizing distillation and quantization within a unified teacher-anchored framework, our method ensures consistent training under quantization, suppressing hard negatives and enlarging decision margins. Additionally, we design a lightweight cross-attention adapter that enhances non-RGB representations through RGB-guided semantic transfer, narrowing the modality gap. Extensive experiments demonstrate consistent improvements on non-RGB modalities while maintaining deployment efficiency.
Abstract (translated)
URL
https://arxiv.org/abs/2609.16689
PDF
-
Channel-Wise and Token-Aware Post-Training Quantization for Visual State Space Duality
2026-09-15 05:21:09Jonghyeon Lim, Changhoon YimAbstract
State space models (SSMs), particularly Mamba, have emerged as efficient alternatives to attention-based architectures and have been extended to vision through ViM, VMamba, and Visual State Space Duality (VSSD). Yet the low-bit post-training quantization (PTQ) behavior of VSSD remains insufficiently understood. A weight-activation split on VSSD-Tiny identifies activation quantization as the dominant low-bit bottleneck, while representative inputs to selected VSSD-backbone linear layers exhibit strong channel-wise magnitude variation and token-localized extremes. We propose the Channel-wise Token-balanced Output-Aware Clipping (CTOAC) method, which learns per-input-channel clipping bounds by minimizing a token-balanced reconstruction loss on the corresponding linear outputs. Only the selected linear layers and their input activations are quantized; other backbone operations retain their original precision. Across VSSD-Tiny, VSSD-Small, and VSSD-Base, the proposed CTOAC method retains ImageNet-1K accuracy and remains substantially more robust than the evaluated baselines at more aggressive precision settings. Applying the same quantization scope to VSSD backbones on COCO and ADE20K preserves strong object detection, instance segmentation, and semantic segmentation performance. An optimized RTX 4090 deployment configuration achieves up to 1.42x end-to-end speedup over FP32.
Abstract (translated)
URL
https://arxiv.org/abs/2609.16656
PDF
-
Large Language Models in the Loop: A Stability- and Network-Aware Survey in Networked Control, Cyber-Physical, and Multi-Agent Systems
2026-09-15 03:50:12Haiping Du, Linping ChanAbstract
Modern networked control systems (NCSs), cyber-physical systems (CPSs), and complex multi-agent network systems (CNSs) increasingly rely on large language models (LLMs) for high-level decision-making. However, the slow, stochastic nature of LLMs directly conflicts with the strict stability and safety guarantees required by these physical systems. This survey presents a unified analysis of how LLMs can be admitted into the control loop of NCS, CPS, and CNS without compromising closed-loop guarantees. We organize this around a core principle: the LLM operates as a slow supervisor adjusting high-level goals and constraints, while a fast, certified inner loop maintains physical stability. Under this framework, LLM integration maps directly to classical networked control challenges, where inference latency acts as delay, API failures as packet dropouts, tokenization as quantization, and hallucinations as bounded disturbances. We assess current developments across all these three domains, highlighting that rising model capabilities are frequently accompanied by a drop in formal safety assurances. Finally, we propose concrete future research directions, identifying the widespread lack of formal stability proofs as the field's central open problem.
Abstract (translated)
URL
https://arxiv.org/abs/2609.16599
PDF
-
Where Post-Training Quantization Breaks Text Embedders: A Measured Map Across Four Embedder Families
2026-09-14 21:57:47Hyojung HanAbstract
Weight-only post-training quantization is the cheapest way to shrink a retrieval embedder, and the received advice for applying it -- protect the embedding table, allocate bits by module sensitivity, prefer a ranking-aware objective over weight reconstruction -- was carried into LLM quantization largely intact. We test that advice on retrieval embedders directly, quantizing five checkpoints from four architecture families across a grid of bit widths and group sizes, and isolating the embedding, attention and feed-forward blocks at each width. Every heuristic fails to transfer as stated. The embedding table never emerges as the dominant isolated protection priority in any family, despite being the largest tensor in several of them. Module sensitivity does not survive as a transferable ordering: at INT4/g16 the spread between modules is too small to allocate against, at INT3 the ordering becomes family-dependent and joint damage stops being the sum of its parts, and at INT2 comparable reconstruction error accompanies retention ranging from 1.3 to 65.9 percent of full precision. A cheap reconstruction proxy is useful for screening uniform bit widths but substantially less reliable for choosing which tensors to protect; its apparent strength across the whole grid is a range-extension artifact. A distilled 109M student at INT3 holds 78.04 NDCG@10 in 68.4 MB and dominates the extreme-PTQ arm of its own 0.6B teacher, 297.9 MB at 64.46, on both size and quality -- but only inside the task it was distilled for. Sizes are byte counts of files that exist rather than arithmetic estimates, and the measurement repository carries the byte provenance for every one of them.
Abstract (translated)
URL
https://arxiv.org/abs/2609.16391
PDF
-
VC-Attention: Value Smoothing and Softmax Casting for Low-bit Attention
2026-09-14 16:17:35Xingyang Li, Dongyun Zou, Shining Zhang, Jiacheng Chen, Haocheng Xi, Lvmin Zhang, Jun-Yan Zhu, Song Han, Zhekai Zhang, Yujun Lin, Muyang LiAbstract
Diffusion Transformers deliver state-of-the-art video generation, but their long spatiotemporal sequences make attention the dominant deployment cost, and a deployable low-bit kernel must be accurate and fast. Accuracy is limited by outliers: a block's quantization scale is set by its largest entries, leaving typical entries confined to a narrow range of representable values. Prior work smooths queries and keys, but value outliers follow no fixed channel or spatiotemporal structure and remain the dominant source of output error. Speed is limited by softmax: low-bit Tensor Cores accelerate only the two matrix multiplications, so the high-precision exponential between them becomes the longest pipeline stage on datacenter GPUs. We propose VC-Attention, a training-free low-bit attention framework that addresses both by pairing Value smoothing with a fused probability Cast. V-Smooth reorders value tokens by lightweight online clustering, so the tokens in a hardware block quantize well together. It quantizes only the residual after subtracting the block mean, and restores that mean from the row sum the online softmax already maintains. ExpCast-FP8 maps log-domain scores directly to E4M3 probability codes with one fused multiply-add, eliminating the FP32 exponential and the format conversion. We implement VC-Attention for B200, B300, H200, RTX PRO 6000, and RTX 5090. Across Wan2.2, LongCat-Video, HunyuanVideo-1.5, and MiniMax-H3, VC-Attention improves fidelity over low-bit baselines, speeds up the attention kernel over BF16 FlashAttention-4 by 1.46-1.59x on datacenter Blackwell and Hopper and by 2.3-3.6x on workstation cards, and generates a clip 1.13-1.19x and 1.36-1.70x faster end to end.
Abstract (translated)
URL
https://arxiv.org/abs/2609.15810
PDF
-
On The Robustness-Resolution Tradeoff In Temporal Quantization Of Event Streams
2026-09-14 11:51:03Sayeed Shafayet Chowdhury, Ruhi SharminAbstract
Event pipelines often discretize asynchronous timestamps before learning. This step looks harmless, but its stability depends directly on temporal resolution. We study this dependence at the representation level. We first show that hard temporal binning is discontinuous: an arbitrarily small timestamp shift near a boundary can move unit event mass between bins. We then define a class of nonnegative, mass-preserving, resolution-faithful continuous encoders and prove that every encoder in this class has global L1 sensitivity at least 2/Delta, where Delta denotes bin width. Linear two-bin interpolation attains this limit. Local support and first-moment preservation also make it unique. Experiments on SHD, N-MNIST, and DVS128 Gesture support the analysis. Across uniform timestamp budgets, linear interpolation lowers mean representation drift by 47-72% while keeping clean accuracy nearly unchanged. On DVS Gesture, it produces zero prediction flips across all tested budgets and three seeds. On SHD, measured drift follows 1/Delta with R^2 = 0.992.
Abstract (translated)
URL
https://arxiv.org/abs/2609.22295
PDF
-
SpliTEE: Improving LLM Inference on Trusted Hardware with Differentially Private GPU Outsourcing
2026-09-14 04:55:17Shashie Dilhara Batan Arachchige, Robin Carpentier, Hassan Jameel Asghar, Dali KaafarAbstract
User prompts provided to large language models (LLMs) may contain sensitive or private information that can be misused by remotely deployed models, such as through inadvertent memorization during retraining. One way to protect user prompts is to execute the LLM inside a trusted execution environment (TEE), with the guarantee that the service provider has no access to computations performed within or information exchanged with the TEE. However, current TEEs are primarily CPU-based and significantly slower than GPUs optimized for LLM inference. To circumvent this, Tramer and Boneh (2019) proposed Slalom, which splits neural network inference between a TEE and an untrusted GPU and encrypts intermediate inputs sent to the GPU. We extend this split-inference architecture to LLM inference and instead protect intermediate inputs using differential privacy. We show that masking intermediate representations is necessary by showing that a prompt-reconstruction attack can recover prompts from these representations with nearly 80% accuracy. Our main contribution is a global sensitivity analysis of key LLM functions, which bounds the required scale of differentially private noise. Unlike encryption, differential privacy avoids quantization, allowing the LLM to remain in the floating-point domain. We also derive an upper bound on floating-point error from masking and noise cancellation in the TEE as a function of the privacy parameter epsilon. We implement our architecture using Intel TDX and evaluate it with two LLMs: Llama-3.2-3B and Qwen3-4B. Our split execution is nearly twice as fast as fully CPU-based inference inside TDX and 5-15 seconds faster than encryption-based Slalom while achieving higher accuracy. Finally, we demonstrate that prompt reconstruction, even with knowledge of the differential privacy mechanism, cannot recover more information than is contained in an unrelated prompt.
Abstract (translated)
URL
https://arxiv.org/abs/2609.15039
PDF
-
Tracing the Origins: Legacy Codec Identification in Neural Audio Transcoding
2026-09-14 02:08:15Wonje Heo, Shinee Youn, Yooshin Kim, Chuck Chae, Donghoon ShinAbstract
Residual Vector Quantization (RVQ)-based neural audio codecs (NACs) enable high-fidelity audio distribution at unprecedentedly low bitrates through discrete token-based representations. However, this shift disrupts traditional forensics, as non-linear neural transcoding obscures the underlying traces of legacy compression. This study defines the forensic gap and proposes a Transformer-based framework designed to leverage the hierarchical and temporal dependencies inherent in RVQ sequences. By modeling inter-layer causal relationships and dynamic forensic significance, our model effectively disentangles superimposed artifacts from legacy-to-neural transcoding. Experimental results achieve 97%+ accuracy for codec identification and robust joint identification performance across 32-128 kbps. These results demonstrate that traditional codec traces persist even after neural transcoding, supporting the feasibility and necessity of neural-codec-aware audio forensics.
Abstract (translated)
URL
https://arxiv.org/abs/2609.14916
PDF
-
GGUF-Metadata Prediction of Single-Sequence llama.cpp Throughput Across Three Systems
2026-09-14 00:31:14Xinyu Qiu, Chuhong Xu, Bo Su, Ziyao Chen, Ruiyang Xu, Shimeng DaiAbstract
We predict single-sequence model throughput from GGUF metadata using roofline-shaped predictors with quantization-specific scale factors fitted on reference models. The scored cohort comprises 318 phase-depth measurements from 53 host-file configurations on two Apple M4 Max systems and an NVIDIA RTX 5080. On host-specific held-out sets of four, five, and two configurations, an active-parameter decode model obtains 13.1%, 14.4%, and 36.1% mean absolute percentage error (MAPE), versus 49.4%, 55.3%, and 51.9% when charging total parameters. Leave-one-host-out coefficients fitted on the other two systems yield 11.6%, 16.8%, and 36.0% test MAPE. A low-bit model ladder changes ordering across runtime stacks. The P2 prefill baseline gives 18.7%, 22.2%, and 108.2% test MAPE. GGUF structure helps on all three systems, but fitted efficiencies are not universal.
Abstract (translated)
URL
https://arxiv.org/abs/2609.14864
PDF
-
TriCalRAG: A Three-Strategy, Retrieval-Augmented Benchmark for On-Premise LLM-Based Root Cause Analysis in AIOps
2026-09-13 19:52:46Rohit Patel, Susil Kumar Mohanty, Jeenal ChaudharyAbstract
Cloud-hosted large language models (LLMs) are increasingly used for root cause analysis (RCA) in AIOps pipelines, but they introduce data privacy risk, network latency, and per-query cost that scale poorly with production log volumes. We present TriCalRAG, a benchmark evaluating open-weight LLMs served locally via vLLM on a single high-memory workstation GPU (NVIDIA RTX PRO 6000, 96GB) against a classical LSTM-based log anomaly detector (DeepLog), across four real, publicly available log datasets (BGL, HDFS, Thunderbird, OpenStack). We evaluate two open-weight models (Qwen2.5-14B, Mistral-Small) under three prompting strategies: zero-shot, few-shot, and retrieval-augmented generation (RAG) over a labeled incident history, reporting accuracy, precision/recall, and F1 with bootstrap 95% confidence intervals across 3 random seeds, alongside throughput and VRAM footprint. Our results show that RAG not only improves mean F1 by 0.10-0.27 over zero-shot prompting but, more importantly, substantially stabilizes model calibration: zero-shot prompting drives both models toward near-degenerate behavior (predicting "anomaly" on up to 100% of incidents on some datasets), while RAG keeps predicted-positive rates close to the true class balance in the majority of configurations. Mistral-Small achieves higher macro-averaged F1 than Qwen2.5-14B (0.644 vs. 0.560) but exhibits calibration failures in more configurations (7 vs. 5 of 12), while running at roughly half the throughput - indicating the better model choice depends on whether a deployment prioritizes peak accuracy or predictable behavior across prompting conditions. Ablations show batching scales throughput 41 times on a single card and that 4-bit quantization reduces latency 20% with no measurable accuracy loss. We release our benchmark harness, dataset splits, and evaluation code to support reproducible on-premise AIOps research.
Abstract (translated)
URL
https://arxiv.org/abs/2609.14762
PDF
-
WaterKron and FlipFlop Hessian: Information-Theoretically Grounded Quantization with Kronecker-factored Hessians
Abstract
How should a Kronecker-factored Hessian approximation be chosen for post-training quantization? We address this question through WaterKron, which combines two-sided GPTQ with row- and column-dependent waterfilling scales and entropy coding. We derive its high-rate distortion with respect to the full Hessian using an explicit Kronecker-Hessian mismatch factor $\Phi$. This factor quantifies the asymptotic distortion penalty due to the Kronecker Hessian approximation and provides a criterion for selecting the factors optimally. Minimizing $\Phi$ leads to a Gaussian covariance-fitting problem with classical ``flip-flop'' updates. We thus give a rate-distortion justification for using the resulting FlipFlop Hessian in quantization. We evaluate it empirically, finding that the FlipFlop Hessian consistently improves KL divergence and perplexity over input-only, marginal, and Frobenius-based Hessian choices.
Abstract (translated)
URL
https://arxiv.org/abs/2609.14706
PDF
-
MorphoStyle: Motion Style Transfer with Morphology Control
2026-09-12 23:40:27Xin Feng, Eleonora D'Arnese, Mohan SridharanAbstract
Human motion may be viewed as a combination of action content, style, and body morphology. Existing motion style transfer methods transfer a reference style onto a content motion while assuming a canonical body, whereas shape-aware motion generators adapt motion to a target shape without explicit style control. This separation of motion style and shape (morphology) makes it difficult to generate stylized motions for non-canonical bodies; naively combining a style transfer module with a shape-aware generator often leaks action content from the style reference and disrupts shape-consistent kinematics. We present MorphoStyle, a framework for shape-aware motion style transfer that is built on a shape-conditioned Finite-Scalar-Quantization Variational Auto-Encoder (FSQ-VAE). The key contribution is to pose the desired style transfer as modular latent disentanglement comprising: (i) a contrastive style encoder that extracts content-decoupled style embeddings; (ii) a text-guided style-routing mechanism that locates style-relevant joints in a text-motion feature space; and (iii) a manifold preserving style modulator that injects discriminative style embeddings in content features as a temporally-gated low-rank offset. Extensive experiments on benchmark datasets demonstrate that MorphoStyle outperforms competing baselines in terms of both shape control and motion style transfer, while simultaneously providing quantitative shape control. For more details, please see project website: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.14189
PDF
-
When Faster VLA Deployment Changes Closed-Loop Behavior: Task Success-Latency Analysis of SmolVLA Across PyTorch and ONNX Variants
2026-09-12 20:56:31Rafiqul IslamAbstract
Vision-language-action (VLA) deployment can reduce inference latency while changing closed-loop task behavior. We evaluate HuggingFaceVLA/smolvla_libero on an RTX 2060 (6 GB) in LIBERO Spatial and Object (MuJoCo 3.3.2, LeRobot 0.6.1, seed 42), comparing PyTorch+AMP with ONNX Runtime CUDA Execution Provider (CUDA EP). The main evaluation uses 100 episodes/suite; a paired rollout uses 300 episodes/suite. PyTorch+AMP reaches 70.0%/88.0% Spatial/Object success at 1181 ms p99. Requested-FP16 and requested-INT8 ONNX reduce tether-inspect p99 to 601 ms and 532 ms, while Spatial success falls to 41.0% and 40.0% and Object remains at 89.0%. A graph audit shows those artifacts are byte-identical FP32 graphs, so the requested-INT8 row is not operator-level INT8 quantization. A static language-width ablation (16/24/32 tokens) yields Spatial success of 41.0%, 75.0%, and 71.0%; widths 24 and 32 recover much of the Spatial drop while Object success and uniform-bench latency stay approximately stable. Width-24 ONNX Spatial success is comparable to the PyTorch+AMP baseline at roughly half the latency (Wilson intervals overlap; two-proportion chi-squared p=0.53). Context width is an important contributor in this stack; it does not account for every PyTorch-vs-ONNX difference. Deployment evaluation should jointly report latency, artifact inspection, interface constraints, and closed-loop success. Code: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.14146
PDF
-
AGENTQ: Quantization-Conditioned Backdoor Attacks on LLM Agents
2026-09-12 17:17:59Xiaoqun Liu, Qiben YanAbstract
Quantization is one of the default deployment paths for open-weight LLM agents, but it is not behavior-preserving: an adversary can release a full-precision checkpoint that passes audits yet misbehaves once quantized, termed as quantization-conditioned attack (QCA). Prior QCA work targets free-text generation, where harm is mediated by a human reader. In contrast, the agentic setting poses a more severe risk: the triggered payload is a structured function that can be executed without human oversight. We present the first study of QCA against LLM agents. We find that directly adapting prior backdoor-injection methods can produce malicious behavior after quantization, but substantially degrades benign utility, rendering the resulting attacks impractical. To understand the true upper bound of the threat, we propose AGENTQ, an attack framework that combines layer-banded LoRA injection with partial-PGD repair over a multi-codebook quantization-equivalence class. AGENTQ preserves normal agentic capability while concentrating malicious behavior in the quantized model. Across three trigger-action pairs and three codebooks (NF4, FP4, INT8), AGENTQ reaches up to 100% post-quantization attack success rate with minimal loss of benign utility, underscoring the need to make quantization-aware safety evaluation a standard requirement before open-weight agents are deployed.
Abstract (translated)
URL
https://arxiv.org/abs/2609.14060
PDF
-
Hardware-Aware Learned Representation Compression for Distributed In-Sensor Vision
2026-09-12 13:45:08Chengwei Zhou, Abu Masum, Xuming Chen, Mehran Moghadam, Sreetama Sarkar, Arnab Sanyal, Md Abdullah-Al Kaiser, M. Hassan Najafi, Sercan Aygun, Gourav DattaAbstract
In-sensor computing reduces the cost of transmitting high-resolution image data by performing early-stage processing near the sensor. However, the logic chip integrated with a CMOS image sensor (CIS) is tightly constrained in compute and memory, limiting conventional deep neural network partitioning. We present OASIS, a distributed in-sensor vision framework that uses a lightweight encoder to generate compact, task-relevant representations before off-chip transmission. The encoder is trained end-to-end using task, entropy, and reconstruction objectives, while the decoder is used only during training. OASIS supports two complementary deployment paths. The first applies 4-bit quantization and Huffman coding while preserving the spatial structure required by classification and dense-prediction tasks. The second uses Sobol-based hyperdimensional computing (HDC) to transform the encoder latent into a fixed-dimensional binary hypervector for associative-memory classification. For the SwinViT-based VWW model, mapping a $3\times3\times8$ latent to a 64-dimensional hypervector provides an additional $1.77\times$ communication reduction with less than one percentage point of accuracy loss relative to the 128-dimensional configuration, yielding an overall $18{,}816\times$ reduction compared with raw 8-bit image transmission. We implement the digital near-sensor pipeline on an AMD Xilinx Zynq UltraScale+ FPGA and characterize it using direct board-level power measurements and Vivado post-implementation analysis, together with circuit-simulated CIS models and a 7-nm ASIC projection. Across visual wake-word classification, hand tracking, and eye tracking, OASIS reduces total system energy by approximately $2\times$-$4.5\times$ while maintaining competitive accuracy, demonstrating a practical hardware-algorithm co-design path for communication-efficient in-sensor vision.
Abstract (translated)
URL
https://arxiv.org/abs/2609.13947
PDF
-
One Spectrum, Two Resources: Data-Memory Scaling in Autoregressive Prediction
Abstract
How much learned memory is needed to benefit from more data? We show that the two resources are governed by one predictive-energy spectrum in a positive-entropy autoregressive retrieval source. Each coordinate contributes its query probability times the squared radius of its unknown logit. Writing $\mu$ for the resulting energy spectrum, we prove the minimax law $\mathfrak R^*_{\rm value}(n,B)\asymp_R \Phi_\mu(n^{-1})+\Phi_\mu(\tau_B), \Phi_\mu(t)=\int\min\{x,t\}\,\mu(\mathrm dx),$ for $n$ prediction blocks and a learned state with at most $2^B$ values. Data set the resolution $1/n$; memory sets the level $\tau_B$ reached by optimal bit allocation. The complete curve also recovers the positive spectrum. Energy-dimension pairing is essential: two causal sources with identical block-energy and block-dimension marginals have different data and memory exponents. A masked query-key attention head learns the route and values, realizing the law with explicit routing, format, and arithmetic errors. Further results give exponent-adaptive allocation, finite-precision realization, and compute-precision laws under two-sided arithmetic assumptions. Experiments recover the data-memory collapse and coupling exponents, explain the routing and allocation mechanisms, and examine weight-only quantization across six pretrained-model scales.
Abstract (translated)
URL
https://arxiv.org/abs/2609.13500
PDF
-
Kraken: LLM-based Speech-to-Speech Translation via Low-bitrate VQ and Dual-path Source Conditioning
2026-09-11 16:41:22Hayato Futami, Hassan Shahmohammadi, Tushar Dhyani, Alkis Koudounas, Rapha\"el Lafargue, Yosuke Kashiwagi, Quentin Jodelet, Emiru TsunooAbstract
Speech-to-speech translation (S2ST) has advanced significantly with speech LLMs, offering the potential for joint optimization and preserving non-linguistic information. However, these models struggle with predicting high-bitrate speech tokens in LLMs, and face the challenge of relying on S2ST training data with ideally aligned speaker identity and prosody. We propose using low-bitrate tokens based on single-layer vector quantization, trained to reconstruct self-supervised learning (SSL) features. We also employ a separate token-to-waveform decoder named Autowave-X, which is also conditioned on the source speech to improve non-linguistic transfer, thereby relaxing the training data constraints. With the integration of these techniques, we propose an S2ST model named Kraken, which augments a pre-trained LLM with speech feature inputs and the low-bitrate token outputs, followed by Autowave-X vocoder. We built the model upon Qwen3-8B and trained it using 150k hours of multilingual and multitask speech data. We demonstrated that our model exhibited better translation quality than SeamlessM4T-Large v2 and Qwen2.5-Omni, along with improved speaker and prosody transfer capabilities.
Abstract (translated)
URL
https://arxiv.org/abs/2609.13045
PDF
-
StepAudio 3 Gen Technical Report
2026-09-11 15:05:20Bin Lin, Bo Zhao, Boyang Wang, Boyang Zhang, Boyong Wu, Chao Yan, Chen Geng, Chen Wu, Cheng Yi, Chengli Feng, Chenglin Zhu, DanNi Wan, Daxin Jiang, Dongqing Pang, Fei Tian, Feng Tian, Future Li, Gang Yu, Guanglong Yang, Jia Peng, Jiahao Song, Jiamin Fan, Jiangjie Zhen, Jianzheng Gao, Jun Chen, Li Xie, Lifang Zhang, Lingli Ji, Liying Shi, Lun Cai, Min Xu, Na Wang, Peilin Li, Peng Yang, Pengfei Tan, Qingjian Lin, Ruijie Xiong, Runze Li, Shenghua Hu, Shi Qiu, Siqi Tu, Siyi Zhou, Tianjiao Deng, Wanying Lu, Weiming Niu, Wen Sun, WenWen Qu, Xiangyu Zhang, Xianwei Zhang, XiaoSu Su, Xing Chen, Xinyu Liu, Xuerui Yang, Yang Li, Yang Yang, Yechang Huang, Yibo Zhu, Yifan Zhang, Yiyang Xu, Yu Fu, Yu Luo, Yu Zhou, Yumang Wang, Yunzhou Ju, Yuxiang Yang, Zekai Liu, Zengwei Yao, Zhenwei Mou, Zheqi Dai, Zhiyue Wu, Zichao ZhouAbstract
We introduce StepAudio 3 Gen, a general-purpose audio generation model that supports zero-shot text-to-speech (TTS), voice design, vocal generation, sound effects, music, vibe speech, and mixtures of multiple audio types within a unified framework. At its core, StepAudio 3 Gen is a discrete autoregressive generator that models audio directly over residual vector quantization (RVQ) tokens, departing from the diffusion Transformer-based continuous generation paradigm prevalent in recent general audio models. Its StepAudio Tokenizer represents general audio at 12.5 Hz in a shared $16 \times 2048$ residual code space, jointly quantizing semantic and waveform-level acoustic features so that each code layer preserves both types of information. For generation, the backbone predicts the first codebook along the time axis using autoregressive modeling, while a lightweight causal Transformer completes the remaining fifteen codebooks along the codebook axis. Our study further identifies three key design principles: (1) interference-aware progressive pretraining for acquiring audio capabilities while preserving the textual abilities of the large language model, (2) RVQ Adaptor for effectively incorporating multi-codebook acoustic representations, and (3) discrete autoregressive modeling over a shared representation across general audio domains. With progressive pretraining, multi-task instruction training, and supervised fine-tuning, StepAudio 3 Gen achieves state-of-the-art performance on both TTS and voice design, while retaining strong generation capabilities across speech, vocals, sound effects, and music. Audio samples are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.12945
PDF
-
ESTS at WMT26: Routing-Informed Expert Pruning for Model Compression
2026-09-11 00:30:43Liu O. Martin, Lucas Bandarkar, Nanyun PengAbstract
We describe six submissions under the team name ESTS to the unconstrained WMT26 Model Compression Shared Task for English--Simplified Chinese and English--Egyptian Arabic. We submit three compression operating points per translation direction, all derived from GPT-OSS-20B. We use task-specific routing mass to rank experts and cross-lingual routing divergence to allocate retained capacity across layers, then physically remove low-importance experts. The resulting specialists are recovery-tuned on GPT-5.1-generated synthetic translation data and further compressed by applying MXFP4 quantization to the retained expert projection weights. We additionally implement a robust inference system for the instruction-conditioned WMT26 setting, including category inference, output validation, retries, segmented fallback, and source-owned JSON reconstruction. Across our six submissions, parameter counts range from 4.186B to 7.770B and packed artifact sizes from 4.55 to 6.33~GiB. Internal xCOMET-XL evaluation using GPT-5.1 pseudo-references provides an internal comparison across the submitted compression operating points.
Abstract (translated)
URL
https://arxiv.org/abs/2609.12310
PDF
-
Why Does Post-Training Quantization Work?
2026-09-10 15:32:32Yuxiang Chen, Michael Beyer, Jun Zhu, Jianfei ChenAbstract
Post-training quantization compresses large language models (LLMs) by storing their weights at reduced precision, and each quantized weight introduces an error into the hidden states. Naively, these errors should accumulate with depth and corrupt next-token prediction; randomly initialized models accumulate these discrepancies rapidly, whereas quantized pretrained models accumulate much less hidden-state error and largely maintain downstream task performance, even though they were never trained with quantization noise. This raises the question we address: why does post-training quantization work? Comparing full-precision and quantized forward passes, we identify two mechanisms that characterize pretrained quantization robustness. First, the error a layer newly introduces tends to oppose the error it inherits from the layer's input. The two cancel partially such that the discrepancy between full-precision and quantized passes grows slowly. This counteracting residual interaction develops during pretraining. Our quantitative analysis identifies it as a major factor slowing hidden-error growth. Second, LM-head geometry preferentially preserves the scores and probabilities of high-ranked tokens, which typically represent the model's most confident predictions. Together, these mechanisms explain why quantization error that passes through numerous layers can still produce only small output changes, and we verify the findings across models and quantization settings.
Abstract (translated)
URL
https://arxiv.org/abs/2609.11716
PDF
