Quantization
-
Distributional Matching for Vector Quantization: A Unified Theoretical and Empirical Framework
2026-07-17 13:18:23Xianghong Fang, Litao Guo, Hengchao Chen, Yuxuan Zhang, XiaofanXia, Dingjie Song, Yexin Liu, Hao Wang, Harry Yang, Qiang Sun, Yuan YuanAbstract
The effectiveness of modern visual representation learning and autoregressive models critically depends on vector quantization (VQ), which discretizes continuous feature representations using a learnable codebook. Despite its widespread use, existing VQ methods often suffer from training instability and codebook collapse, arising from gradient mismatch induced by the straight-through estimator and the under-utilization of code vectors. In this work, we show that both issues can be traced to a fundamental mismatch between the distributions of feature vectors and code vectors, leading to inefficient representation and information loss. Building on this observation, we propose a distributional matching framework for vector quantization. We introduce principled criteria for desirable VQ behavior and demonstrate through theoretical analysis and empirical evaluation that aligning feature and code vector distributions provides a unifying mechanism for mitigating training instability and codebook collapse. We instantiate this framework using a Wasserstein-based objective with an efficient closed-form under a mild Gaussian approximation, and further show that a nonparametric alternative based on maximum mean discrepancy yields comparable performance. Extensive experiments on visual tokenization benchmarks support the effectiveness and robustness of the proposed approach.
Abstract (translated)
URL
https://arxiv.org/abs/2607.15933
PDF
-
An MLIR-Based Compilation Method for Large Language Models
2026-07-17 11:24:45Pengchao Hu, Zhibin Xin, Yifan Chen, Yangyang Zhou, Liang WangAbstract
Large Language Models (LLMs) have become the dominant workload on modern AI accelerators, yet deploying them on specialized hardware still faces two core challenges: how to import a trained model into a compiler-friendly intermediate representation, and how to efficiently schedule the autoregressive inference loop under limited on-chip memory. This paper presents an MLIR (Multi-Level Intermediate Representation) based compilation method for large language models, illustrated using two dialects of operators, TopOp and TpuOp. TopOp serves as a high-level graph dialect that is independent of both the source framework and the target chip, and is responsible for expressing model semantics; TpuOp serves as the target hardware dialect, carrying chip-related decisions such as quantization, layer groups, and memory layout. A model is first represented as TopOp, then lowered layer by layer to TpuOp, and finally a deployable binary is generated. In addition, each Transformer layer is split into three stages for static compilation: prefill, prefill_kv (prefill with historical key-value cache), and decode, so as to accommodate the different computational characteristics of prompt-parallel processing and per-token generation. The method has been implemented in the TPU-MLIR compiler{this https URL} and the LLM-TPU deployment project\footnote{this https URL}, supporting a variety of generative models including the Qwen, Llama, InternVL, and MiniCPM-V series, as well as multiple quantization and deployment forms such as GPTQ, AWQ, and AutoRound.
Abstract (translated)
URL
https://arxiv.org/abs/2607.15865
PDF
-
AuEmoChat: Authentic Emotion Understanding and Rendering for Conversational Speech Synthesis
2026-07-17 08:47:58Zhenqi Jia, Yuan Zhao, Aruukhan, Rui Liu, Haizhou LiAbstract
Conversational Speech Synthesis (CSS) aims to synthesize speech with human-like emotional expression and contextual consistency in user-agent interactions. Existing CSS methods struggle to render authentic human emotions due to limited predefined emotion label spaces (e.g., seven emotion categories), while redundant multimodal tokens in multi-turn dialogue history interfere with context understanding. To address these issues, we propose AuEmoChat, a CSS framework for authentic emotion understanding and rendering. First, we develop AuEmoCodec, which learns a discrete authentic emotion token space from large-scale emotional speech via finite scalar quantization, enabling a more authentic emotion representation than limited basic emotion categories. We further propose AuEmoToMe, an authentic-emotion-guided token merging algorithm that merges redundant tokens in multimodal dialogue history while preserving emotion-relevant context. We integrate it into an autoregressive text-speech model to predict the target authentic emotion token and speech tokens. Finally, we propose Authentic Emotion Flow Matching, which renders speech by jointly conditioning on merged dialogue context, target authentic emotion, and acoustic priors. Extensive experiments on the NCSSD-EmCap dataset demonstrate that AuEmoChat outperforms state-of-the-art CSS baselines and generates more expressive and authentic emotional speech.
Abstract (translated)
URL
https://arxiv.org/abs/2607.15755
PDF
-
Looped Latent Attention: Cross-Loop KV Compression for Looped Transformers
2026-07-16 20:58:16James O' Neill, Fergal ReidAbstract
Looped, weight-tied Transformers reduce parameters by reusing a block, but decoding still stores a separate K/V cache for every recurrence step. We show that this loop-indexed cache is highly structured. For a fixed token, layer and head, K/V vectors trace a short low-rank trajectory across loops, while the head and layer axes remain much flatter. We introduce Looped Latent Attention (LLA), a post-training cache codec that stores compact K and V latents and reconstructs loop-specific K/V vectors only when attention reads them. The default per-head codec compresses recurrence, while LLA-2D also folds heads into one latent for the extreme-compression regime. The codec is initialized from SVD of teacher activations and refined with KL and attention-output distillation. At matched cache budget, per-head LLA outperforms head-axis MLA, cross-layer sharing, KV quantization and final-loop reuse, showing that the recurrent cache is low-rank but not safely collapsible to a single state. The same axis advantage holds on Ouro-2.6B-Thinking and transfers to Huginn-3.5B, where an SVD codec remains near-lossless to 32x in decoder-independent evaluation. The cache reduction is exact. On one H200, the latent-store path increases measured Ouro-1.4B batch capacity at 4k context from 32 to 768 sequences at 21.3x compression. For long math rollouts, on-policy refinement on student-generated prefixes raises MATH-500 at 4x from 0.43 to 0.66 and reduces no-answer generations.
Abstract (translated)
URL
https://arxiv.org/abs/2607.15456
PDF
-
qZACH-ViT: Quantization-Aware Intrinsic Explanations with Recursive Attribution-Stabilized Optimization
2026-07-16 19:51:48Athanasios AngelakisAbstract
Compact medical-image classifiers need efficiency and interpretable evidence, yet these goals are often addressed separately. We introduce qZACH-ViT, a quantization-aware extension of the zero-token (CLS-token-free), position-free ZACH-ViT backbone with recursive intrinsic patch-level class evidence. We also introduce Recursive Attribution-Stabilized Optimization (RASO), which norm-matches classification and attribution gradients and removes attribution components that conflict with classification. We evaluate four controlled conditions on seven MedMNIST datasets using 50 training images per class and ten fixed seeds, completing 280 runs. All 210 qZACH-ViT checkpoints are converted to executable mixed-precision ONNX INT8 graphs containing 16 signed INT8 MatMulInteger projections with INT32 accumulation. Deployed mixed-precision INT8 qZACH-ViT with Adam improves the FP32 ZACH-ViT baseline mean on all seven datasets, with a mean paired gain of 0.0313 in the dataset-specific primary metric; qZACH-ViT with RASO yields a mean gain of 0.0368. Across 964,920 source-to-INT8 test comparisons, prediction agreement is 99.9751\%, with a mean absolute primary-metric change of 0.000133 and a maximum of 0.004386. Across 3,600 matched intrinsic maps, mean cosine similarity is 0.999955, mean rank correlation is 0.9944, and mean top-10\% overlap is 0.9692. ONNX artifacts are 70.0\% smaller than source checkpoints and provide $1.41\times$ and $2.39\times$ end-to-end CPU speedups with one and four threads. RASO significantly reduces sufficiency error and improves input-noise stability over Adam with the same attribution loss, but does not dominate every predictive or explainable artificial intelligence (XAI) metric. These results establish qZACH-ViT as a deployable compact intrinsically explainable model and RASO as a targeted stability-oriented optimization procedure.
Abstract (translated)
URL
https://arxiv.org/abs/2607.15421
PDF
-
Lazy Arithmetic using Systolic Arrays for Closing the Verification Gap on Embedded Systems
2026-07-16 15:36:51Taisa Kushner (Galois Inc), Ryan McCleeary (Galois Inc), Martin Brain (City St George University of London)Abstract
Complex algorithms such as deep neural networks are increasingly being deployed on embedded, resource constrained platforms. However, existing hardware and software schemes for implementing these models on the edge fall short, particularly for safety-critical applications such as medical devices. First, hardware such as GPUs, NPUs and TPUs are designed for throughput rather than correctness of computation of security, and are as such susceptible to fault injection attacks. Second, software schemes designed for porting algorithms onto edge devices -- such as quantization schemes -- are either static and sound (non-optimal power consumption), or dynamic yet unsound (non-optimal for safety-critical applications). To address both these needs we propose a both wholly new approach to real-time, dynamic and sound quantization, as well as the hardware to support it. First we developed a sound, real-time adaptive-precision quantization approach utilizing left-to-right arithmetic to pass the most significant bits (MSB) first, and dynamically adjust precision online while performing sensitivity analysis to quantify and manage the risk of decision-boundary crossings. Next, we propose a novel hardware approach utilizing systolic arrays to perform left-to-right arithmetic to generate the MSB first. Together this provides a wholly novel scheme for enabling not only resource-efficient neural networks and artificial intelligence at the edge, but broadly sound and resource-efficient high-precision mathematics on hardware that ensures resilience to bit flip attacks on the most critical bits. This is presented herein as work-in-progress, with software implementations completed and hardware in-progress.
Abstract (translated)
URL
https://arxiv.org/abs/2607.15328
PDF
-
TAMF-VTON: Texture-Aware Mask-Free Virtual Try-On via High-Fidelity Image Synthesis
2026-07-16 10:26:27Jie Wang, Qian He, Gaofeng He, Xiaogang Jin, Huamin WangAbstract
Recent diffusion-based virtual try-on (VTON) methods remain limited by their reliance on segmentation masks, insufficient preservation of fine-grained textures, and limited support for arbitrary multi-garment compositions. Consequently, existing approaches still face significant challenges in real-world e-commerce deployment. We present TAMF-VTON, a texture-aware, mask-free framework that enables high-fidelity image synthesis under practical unconstrained conditions. Our method requires no human parsing or inpainting masks at inference time and supports diverse garment styles, categories, and quantities, enabling the simultaneous transfer of multiple items while preserving body structure and intricate texture details. This is achieved through a unified generative pipeline with three key components: (1) a lightweight Mixture-of-Experts (MoE) adaptation scheme that enables efficient fine-tuning without compromising the base model's general editing capabilities; (2) a frequency-domain supervision mechanism that explicitly optimizes high-frequency spectral consistency to preserve high-fidelity textures; and (3) a robust data curation pipeline employing an adaptive inpainting strategy to simulate the inverse VTON process for high-quality training pair generation. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods in both quantitative metrics and perceptual quality. Optimized for efficiency, the model achieves inference in under 15 seconds per image on an NVIDIA RTX 4090 with INT4 quantization. By combining mask-free operation, flexible multi-garment composition, faithful texture preservation, and efficient inference on consumer hardware, TAMF-VTON demonstrates a commercially viable solution for scalable deployment in real-world digital fashion scenarios. The project is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14807
PDF
-
Non-vacuous Generalization Bounds for Reinforcement Learning with Verifiable Rewards
2026-07-16 02:42:24Yuxuan Zhu, Rohan Alur, Daniel KangAbstract
While reinforcement learning with verifiable rewards (RLVR) is widely used to improve the reasoning capabilities of large language models (LLMs), the generalizability of the resulting models remains poorly understood. In this work, we establish the first non-vacuous generalization bounds for parameter-efficient RLVR fine-tuning at the billion-parameter scale. Our approach adapts PAC-Bayes compression bounds to this setting, and addresses the inherent stochasticity of token generation by applying the Gumbel-max reparameterization trick. To operationalize these bounds, we propose the Progressive RLVR framework, which integrates RLVR with on-policy distillation, TinyLoRA, and model quantization. Progressive RLVR empirically retains 84-97% performance of standard LoRA fine-tuning while producing models that are 14,796x more compressible. We show that this framework yields non-vacuous generalization bounds in four domains: mathematical problem-solving, programming, general-knowledge reasoning, and Text-to-SQL. Our bounds exceed the accuracy of the base model by 9-51% and lie within 6-11% of the accuracy of the fine-tuned models.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14506
PDF
-
VideoRAE: Taming Video Foundation Models for Generative Modeling via Representation Autoencoders
2026-07-15 17:59:23Zhihao Xie, Junfeng Wu, Xinting Hu, Junchao Huang, Li JiangAbstract
Video generative models commonly rely on latent spaces learned by 3D Variational Autoencoders (3D-VAEs). However, conventional 3D-VAEs are mainly optimized for pixel-level reconstruction, which can limit the semantic and spatio-temporal structure captured by their latents. Meanwhile, Video Foundation Models (VFMs) such as V-JEPA 2 and VideoMAEv2 show strong video understanding capabilities, yet whether their frozen representations can be transformed into compact, reconstruction-capable, and generation-friendly video latents remains largely unexplored. We answer this question with VideoRAE, a representation autoencoder that leverages multi-scale hierarchical features from a frozen video foundation encoder and compresses them with a lightweight 1D self-attention projector. VideoRAE supports both continuous latents for Diffusion Transformers and discrete tokens for autoregressive models via multi-codebook high-dimensional quantization. During decoding, a local-and-global representation alignment objective with the frozen VFM teacher improves semantic preservation and enables training without KL regularization. Experiments show that VideoRAE achieves strong reconstruction in both continuous and discrete regimes. On UCF-101, it obtains state-of-the-art class-to-video gFVDs of 40 and 93 with AR and DiT generators, respectively, while converging approximately 5x faster than competing autoencoder baselines. In a controlled 2B-scale text-to-video study, replacing LTX-VAE with VideoRAE leads to faster convergence under comparable settings. These results validate frozen VFM representations as versatile and generation-friendly video latents. The model and code will be released on this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2607.14088
PDF
-
ExTernD: Expanded-Rank Ternary Decomposition Ternary LLM PTQ with Accuracy Approaching Any Quantization Level
Abstract
We introduce ExTernD (Expanded-rank Ternary Decomposition), a post-training factorization of each LLM weight matrix $A \in \mathbb{R}^{m \times n}$ into $A \approx B \mathrm{diag}(D) C$ with ternary factors $B \in \{-1,0,+1\}^{m \times k}$, $C \in \{-1,0,+1\}^{k \times n}$ and a real scale vector $D \in \mathbb{R}^k$. The inner rank $k = \mu \min(m,n)$ is deliberately expanded beyond full rank ($\mu > 1$), so that components past full rank correct the quantization error of earlier ones. We prove the residual decreases monotonically in $k$ and can be driven below any $\varepsilon > 0$: ExTernD approaches bf16 accuracy arbitrarily closely, which no ternary scheme with a fixed plane count can do. Memory and compute scale continuously with $\mu$, and factor sparsity continuously with a threshold $\tau$, so an accuracy target is hit exactly rather than rounded to the next bit-width. ExTernD matches Q4_K's per-matrix accuracy at 5.2-5.5 effective bpw (5.1-5.5 with importance weighting) on Gemma-4-E2B and Qwen3.5-4B, and a full Qwen3.5-4B conversion at $\mu = 3$ reaches 10.10 wikitext-2 perplexity against 9.78 for bf16 (+3.2%), placing it near the Q4_K/Q5_K accuracy band at ~5.7 effective bpw.
Abstract (translated)
URL
https://arxiv.org/abs/2607.13511
PDF
-
Silent Alarm: A J-Space Protocol for Comparing Danger Recognition Across Models and Quantization Levels
2026-07-14 14:08:18Roman Prosvirnin, Victor Minchenkov, Alexey Soldatov, Vladimir BashunAbstract
Jailbreak-robustness research typically evaluates safety through generated responses using an LLM-as-judge approach. Such evaluations, however, are sensitive to the benchmark's grading procedure and capture only observed behavior on a given set of attacks, without directly revealing the hidden fragility of the underlying safety mechanisms. This work proposes JADR (Jacobian Assessment of Danger Recognition), a protocol that measures a model's internal representation through Jacobian space (J-space, a recently proposed workspace of verbalizable concepts) before the first response token is generated. For every prompt and layer we record the top-k J-space tokens; these are grouped into six behavioral scenario axes and compared between a danger sample based on StrongREJECT and a safe control drawn from XSTest and OKTest. The method does not call on an external judge model: the computation runs entirely locally, on the activations of the model under evaluation, which lets us compare both different models against each other and modifications of a single model -- quantization and fine-tuning in particular -- on the same terms. The final comparison rests on the proposed SafetyAUC metric, complemented with bootstrap confidence intervals. The protocol is applied to six models (Qwen3-1.7B, Qwen3-4B, Qwen3-8B, Qwen3-Uncensored-4B, Qwen3-SafeRL-4B, Gemma 2 9B) across three weight-representation regimes -- BF16, INT8, and INT4 -- and checked against an independent behavioral evaluation with the StrongREJECT grader. The metric separates models with a strong versus a weak internal safety mechanism with statistical significance and captures substantively different effects across quantization regimes.
Abstract (translated)
URL
https://arxiv.org/abs/2607.12792
PDF
-
AVQ-Attention: Adaptive Vector-Quantized Attention
2026-07-14 14:02:22Winfried van den dool, Patrick Forr\'e, Amir Habibian, Yuki M. Asano, Max WellingAbstract
The $\mathcal{O}(N^2)$ complexity of attention over $N$ tokens remains a computational bottleneck in transformer models. Vector-Quantized (VQ) attention reduces this to $\mathcal{O}(MN)$ by representing keys with $M$ codewords, but applies uniform codebook capacity regardless of where attention mass concentrates: high-attention regions of key space may be coarsely approximated while low-attention regions waste representational capacity. We propose Adaptive Vector-Quantized (AVQ) Attention, which adaptively allocates codebook capacity based on attention importance. Starting from a small set of codewords, our method identifies the most important codes during the forward pass and refines them with pre-learned child codewords, achieving fine-grained quantization where it matters most while maintaining coarse quantization elsewhere. We develop an implementation using custom Triton kernels that enables the full adaptive refinement process, including importance scoring, child codeword insertion, and parent contribution replacement, to be carried out within the tiled computation paradigm of Flash Attention with minimal overhead. Our approach maintains $\mathcal{O}(MN)$ complexity while achieving improved accuracy-efficiency trade-offs compared to fixed-codebook VQ-attention.
Abstract (translated)
URL
https://arxiv.org/abs/2607.12789
PDF
-
A JoLT for the KV Cache: Near-Lossless KV Cache Compression via Joint Tucker and JL-Residual Allocation for LLMs
2026-07-14 09:23:20Rahul Krishnan, Volker SchulzAbstract
The key-value (KV) cache has become the dominant memory cost of transformer inference. It grows with batch size, context length, and depth, and at long context it, rather than the model weights, sets the ceiling on throughput. Two families of methods reduce it. Low-rank methods factor two-dimensional slices of the cache, either per-head matrices or cross-layer feature blocks, and quantization methods lower the bit-width of every entry. Neither family exploits the fact that the cache at a layer is naturally a third-order tensor whose three axes, the heads, the tokens, and the features, carry very different amounts of redundancy. We take this tensor view directly. Our method, JoLT, applies a partial Tucker decomposition that compresses only the token and feature axes while leaving the head and layer axes intact, and then restores the energy that truncation discards with a Johnson-Lindenstrauss (JL) rotated low-bit residual. A single Lagrangian dual allocates the Tucker ranks and the residual bit-widths together, per layer group and separately for keys and values, under one byte budget. The result is a near-lossless 2-3x compression: perplexity, GSM8K accuracy, and RULER needle-in-a-haystack retrieval all stay at or within statistical noise of the uncompressed baseline on both a grouped-query-attention model (Mistral-7B-v0.3) and a multi-head-attention model (LLaMA-2-13B). At 2x, JoLT reconstructs the cache to relative Frobenius error 0.009 (K) and 0.006 (V) on both architectures, roughly an order of magnitude below cross-layer SVD and 4-bit quantization. A randomized-SVD variant, FlashJoLT, delivers a 5-13x compression-time speedup at matched quality.
Abstract (translated)
URL
https://arxiv.org/abs/2607.12550
PDF
-
Cost-Governed RAG: Unified Per-Tenant Cost Attribution Across Retrieval and Generation in Multi-Tenant LLM Systems
Abstract
Enterprise Retrieval-Augmented Generation (RAG) deployments face a critical governance gap: while LLM generation cost is metered per token, the retrieval layer - vector memory, similarity compute, and embedding API calls - remains an unattributed shared cost, enabling invisible cross-subsidization among tenants. We present Cost-Governed RAG, an architecture that integrates a codebook-oblivious vector index (TurboVec) with a multi-tenant LLM governance gateway, creating a unified observability stack where embedding, retrieval, and generation costs are jointly attributable per tenant. The architecture exploits TurboVec's deterministic, closed-form memory formula to enable near-exact per-tenant retrieval cost calculation - a property unavailable in graph-based indexes with non-linear memory overhead. Deployed on Snowpark Container Services within a cloud data platform's governance boundary, the system achieves 99.96% end-to-end cost attribution accuracy across 100 simulated tenants (10M vectors, log-normal size distribution) with telemetry overhead below 0.04% of query latency. The architecture reduces retrieval infrastructure cost by 3.1-9.0x compared to managed vector database services under the pricing assumptions detailed in Section IV. We formalize a three-layer cost model and demonstrate that codebook-oblivious quantization enables deterministic per-tenant cost attribution while also removing the shared-codebook leakage surface present in trained quantizers - the latter observation being exploratory and subject to the limitations described in Section VII.
Abstract (translated)
URL
https://arxiv.org/abs/2607.12188
PDF
-
Efficient Tuning Before Low-Bit Post-Training Quantization for Stochastic Gradient Descent-optimized Models
2026-07-13 10:24:36Peng Xia, Junbiao Pang, Muhammad Ayub SabirAbstract
Post-training quantization (PTQ) compresses deep neural networks for deployment under limited memory and computational budgets. However, low-bit (i.e., 2-bit or 4-bit) PTQ often suffers from substantial performance degradation. Most existing PTQ methods operate on an unconstrained full-precision (FP) model and primarily address quantization errors through post-hoc reconstruction. We argue that low-bit PTQ accuracy is limited not only by post-quantization error minimization, but also by the quantization-error tolerance of a FP model itself. In this paper, we propose Efficient Tuning Before Quantization (ETBQ), a pre-conditioning tuning stage for Stochastic Gradient Descent (SGD)-optimized models before PTQ. During tuning, the FP model is optimized under perturbations sampled from the error distributions of weight and activation quantization, guiding the model toward a loss-landscape region that is less sensitive to the subsequent PTQ. Unlike QAT, ETBQ does not train a fake-quantized deployment model, which is computationally and memory intensive. Instead, ETBQ outputs a FP model that can be used by any PTQ backend. Experiments on CIFAR-100, Tiny-ImageNet, ImageNet, and Cityscapes provide consistent evidence that ETBQ improves low-bit PTQ across diverse tasks. Under W2A4 settings, e.g., ETBQ improves over naive PTQ by 2.14\% top-1 accuracy on Tiny-ImageNet and by 5.80\% mIoU on Cityscapes. Code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2607.11359
PDF
-
Calibrated e-CUSUM Decoding for Quantized Reasoning Models: Why Token Log-Probability Is the Wrong Observable for Decoding Monitors
2026-07-13 09:34:36El Hassane Ettifouri (Novelis Research, Paris, France), Ayoub Belfatmi (Novelis Research, Paris, France), Mahaman Sanoussi Yahaya Alassan (Novelis Research, Paris, France), Walid Dahhane (Novelis Research, Paris, France)Abstract
Low-bit quantization makes small reasoning models inexpensive to deploy but can degrade their chains of thought. This motivates decoder-side monitors that intervene when generation becomes unreliable. We show that a natural candidate, the centered token log-probability increment $\log p(w_t)+H_t$, is the wrong observable for this purpose. Under the model's own sampling law it is a mean-zero martingale by construction, so it measures sampling self-consistency rather than trajectory health and is nearly silent during confident repetition, where both $\log p(w_t)$ and entropy are close to zero. We introduce a training-free decoding controller that combines (i) a degeneration-aware alarm score fusing token uncertainty with explicit verbatim repetition and (ii) a calibrated e-process-inspired sequential detector. The raw product process is Ville-valid under a conditional-mean null, while the deployed CUSUM-floored statistic is treated as an empirical change detector because the score is history-dependent and autocorrelated. On GSM8K with DeepSeek-R1-Distill-Qwen-1.5B in FP16 and INT4, calibration turns a monitor that fires on 93--95% of generations into a selective detector of failing traces ($\phi \approx 0.3$, precision $\approx 0.6$ against a 0.38 base rate). In this pilot, the controller reduces measured verbatim-degeneration signals and yields a positive but statistically inconclusive INT4 accuracy change from 63% to 69% (paired McNemar $p=0.18$, $n=100$), at a 28% token-budget cost. We also find that non-termination, rather than looping, is the dominant failure mode on GSM8K. The main contribution is methodological: an explanation of why centered token log-probability is inadequate for decoder monitoring and a calibrated, cautiously evaluated replacement.
Abstract (translated)
URL
https://arxiv.org/abs/2607.11317
PDF
-
Q-BridgeNet: A Quantization Network for Cross-Lingual Sign Language Translation
2026-07-13 08:09:03Liqian Feng, Lintao Wang, Xiaochen Liu, Anusha Withana, Ken-Tye Yong, Dehui Kong, Zhiyong Wang, Kun HuAbstract
Most sign language translation (SLT) methods focus on isolated native sign-spoken pairs (e.g., American Sign Language - English). Extending language-specific SLT models to multilingual translation would improve accessibility by enabling communication across diverse sign and spoken language communities. However, existing multilingual SLT approaches still struggle to learn a unified model that minimizes cross-lingual conflicts while capturing shared cross-lingual semantics and preserving language-specific variations across different sign languages. Therefore, we propose Q-BridgeNet, a unified framework for multilingual SLT that jointly mitigates cross-lingual conflicts across both the sign language and spoken language sides. On the sign language side, Q-BridgeNet learns discrete Q-units via adaptive segmentation and residual vector quantization: a shared base codebook provides language-agnostic semantic primitives, while language-specific residual codebooks refine heterogeneous signing semantics. On the spoken language side, a multilingual LLM is fine-tuned to operate in the Q-unit space, leveraging cross-lingual priors to enable a unified SLT model. Experiments on PHOENIX14T, How2Sign, and CSL-Daily show that Q-BridgeNet effectively mitigates cross-lingual conflicts, achieving state-of-the-art performance on native sign-spoken pairs while also demonstrating strong generalization to non-native pairs. Our source code is publicly available at: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2607.11215
PDF
-
Weight-Adjusted Gradients Reveal Parameter Importance and Failure Modes in LLMs
2026-07-12 15:21:19Shrestha Datta, Hongfu Liu, Anshuman ChhabraAbstract
Understanding which parameters are influential in Large Language Models (LLMs) is central to improving their efficiency, reliability, and interpretability. We introduce Weight-Adjusted Gradients (WAG), a simple yet effective approach for estimating parameter importance that explicitly captures the interaction between model weights and first-order gradient information and identifies parameters that disproportionately influence model behavior, such as those responsible for collapse phenomena in LLMs. Across a range of models and settings, we show that WAG surfaces a tiny but critical subset of parameters whose modification leads to dramatic degradation in performance, a failure mode that existing importance metrics overlook. These findings reveal a previously underexplored interplay between weights and gradients, suggesting that parameter importance cannot be fully understood through either signal alone. The surprising effectiveness of WAG points to fundamental structural properties of trained networks and motivates new open questions about the role of zeroth-order and first-order information in deep learning. We demonstrate the practical utility of WAG across multiple applications, including expert allocation in mixture-of-expert architectures, parameter-specific unlearning, mixed-precision quantization, and layer selection for knowledge editing. Our results position WAG as a unified approach for analyzing, debugging, and controlling LLMs, and opens new directions for principled model-level interpretation.
Abstract (translated)
URL
https://arxiv.org/abs/2607.10803
PDF
-
Dual-Process Atomic Skill Learning: Decoupling Semantic Reasoning and Real-Time Control
2026-07-12 07:38:53Jun Chen, Erdent Bao, Wenlong Dong, Jierui Liu, Qi Cai, Hao Wan, Shaopeng Li, Weijun Qin, Jing Liang, Huiping ZhuangAbstract
Language-conditioned Imitation Learning (IL) is essential for enabling robots to perform complex tasks following natural language instructions. However, generalizing to multi-step compositional tasks remains a significant challenge. While hierarchical approaches attempt to address this by decomposing tasks into atomic skills, existing methods often suffer from training instability and codebook collapse due to the tight coupling between high-level skill reasoning and low-level action generation in joint training paradigms. Inspired by the Dual-Process Theory of cognition, we propose Dual-Process Atomic Skill Learning (DASL), a novel asynchronous hierarchical imitation learning framework that decouples slow semantic reasoning from fast, real-time motion control. DASL comprises a Slow-Frequency Policy that predicts interpretable, discrete skills via Vector Quantization, and a High-Frequency Policy that leverages a latent diffusion model and a Decision Transformer to generate precise actions conditioned on these latent skills. By asynchronously coordinating these modules and utilizing diffusion to structure the latent space, our framework mitigates the skill codebook interference problem common in joint training paradigms. Evaluations across simulation benchmarks and experiment demonstrate that DASL significantly outperforms state-of-the-art baselines, excelling in skill acquisition and compositional generalization to unseen instructions. GitHub page: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2607.10625
PDF
-
Differentiable Proxy Learning for Adaptive Quantization Control in H.264 Video Coding
2026-07-11 21:15:42Qihan Xu, Ivan V. Baji\'cAbstract
H.264 has been the most widely used video coding format for the past two decades due to its relative simplicity, efficiency, and wide availability of software and hardware implementations. However, optimizing codec parameters such as the quantization parameter (QP) for specific objectives (e.g., perceptual quality or machine vision tasks) is challenging due to the non-differentiable nature of standard video codecs. While differentiable proxies have recently been used to enable gradient-based optimization around standard codecs, their fidelity to the target codec is rarely explicitly characterized. In this paper, we propose a differentiable proxy learning method for H.264 intra codec to enable adaptive quantization control. Built upon a variable-rate learned compression model, the proposed proxy is made differentiable with respect to codec QP through a soft-indexing mechanism. It is then trained to approximate the rate-distortion behavior of H.264 under two quantization settings: global-QP, which uses one QP per image, and spatial-QP, which assigns QPs at the macroblock level. Using the frozen trained proxy, we develop a proxy-based adaptive quantization (AQ) framework for both perceptual optimization and machine vision tasks. Experimental results demonstrate that the proposed proxies closely approximate the rate-distortion behavior of H.264 intra codec. The resulting proxy-based AQ framework consistently improves rate-task trade-offs over fixed-QP H.264 baselines, achieving BD-rate reduction of up to 17.12% for semantic segmentation and 15.30% for MS-SSIM.
Abstract (translated)
URL
https://arxiv.org/abs/2607.10478
PDF
