Despite extensive efforts to align Large Language Models (LLMs) with human values and safety rules, jailbreak attacks that exploit certain vulnerabilities continuously emerge, highlighting the need to strengthen existing LLMs with additional safety properties to defend against these attacks. However, tuning large models has become increasingly resource intensive and may have difficulty ensuring consistent performance. We introduce Speculative Safety-Aware Decoding (SSD), a lightweight decoding-time approach that equips LLMs with the desired safety property while accelerating inference. We assume that there exists a small language model that possesses this desired property. SSD integrates speculative sampling during decoding and leverages the match ratio between the small and composite models to quantify jailbreak risks. This enables SSD to dynamically switch between decoding schemes to prioritize utility or safety, to handle the challenge of different model capacities. The output token is then sampled from a new distribution that combines the distributions of the original and the small models. Experimental results show that SSD successfully equips the large model with the desired safety property, and also allows the model to remain helpful to benign queries. Furthermore, SSD accelerates the inference time, thanks to the speculative sampling design.

Offline reinforcement learning (RL) has achieved significant advances in domains such as robotic control, autonomous driving, and medical decision-making. Most existing methods primarily focus on training policies that maximize cumulative returns from a given dataset. However, many real-world applications require precise control over policy performance levels, rather than simply pursuing the best possible return. Reinforcement learning via supervised learning (RvS) frames offline RL as a sequence modeling task, enabling the extraction of diverse policies by conditioning on different desired returns. Yet, existing RvS-based transformers, such as Decision Transformer (DT), struggle to reliably align the actual achieved returns with specified target returns, especially when interpolating within underrepresented returns or extrapolating beyond the dataset. To address this limitation, we propose Doctor, a novel approach that Double Checks the Transformer with target alignment for Offline RL. Doctor integrates the strengths of supervised learning (SL) and temporal difference (TD) learning by jointly optimizing the action prediction and value estimation. During inference, Doctor introduces a double-check mechanism: actions are first sampled around the desired target returns and then validated with value functions. This ensures more accurate alignment between predicted actions and desired target returns. We evaluate Doctor on the D4RL and EpiCare benchmarks, demonstrating aligned control yields stronger performance and tunable expertise, showing its effectiveness in a wide range of tasks.

We present Jet-Nemotron, a new family of hybrid-architecture language models, which matches or exceeds the accuracy of leading full-attention models while significantly improving generation throughput. Jet-Nemotron is developed using Post Neural Architecture Search (PostNAS), a novel neural architecture exploration pipeline that enables efficient model design. Unlike prior approaches, PostNAS begins with a pre-trained full-attention model and freezes its MLP weights, allowing efficient exploration of attention block designs. The pipeline includes four key components: (1) learning optimal full-attention layer placement and elimination, (2) linear attention block selection, (3) designing new attention blocks, and (4) performing hardware-aware hyperparameter search. Our Jet-Nemotron-2B model achieves comparable or superior accuracy to Qwen3, Qwen2.5, Gemma3, and Llama3.2 across a comprehensive suite of benchmarks while delivering up to 53.6x generation throughput speedup and 6.1x prefilling speedup. It also achieves higher accuracy on MMLU and MMLU-Pro than recent advanced MoE full-attention models, such as DeepSeek-V3-Small and Moonlight, despite their larger scale with 15B total and 2.2B activated parameters.

The manual creation of the "Impression" section in radiology reports is a primary driver of radiologist burnout. To address this challenge, we propose a coarse-to-fine framework that leverages open-source large language models (LLMs) to automatically generate and personalize impressions from clinical findings. The system first produces a draft impression and then refines it using machine learning and reinforcement learning from human feedback (RLHF) to align with individual radiologists' styles while ensuring factual accuracy. We fine-tune LLaMA and Mistral models on a large dataset of reports from the University of Chicago Medicine. Our approach is designed to significantly reduce administrative workload and improve reporting efficiency while maintaining high standards of clinical precision.

LLMs (Large language models) have revolutionized NLP (Natural Language Processing), yet their pedagogical value for low-resource languages remains unclear. We present GRILE (Grammar Romanian Inference and Language Explanations) , the first open benchmark of 1,151 multiple-choice questions harvested from Romanian high-stakes exams (National Evaluation, Baccalaureate, university admissions). GRILE enables us to probe two complementary abilities of seven state-of-the-art multilingual and Romanian-specific LLMs: (i) selecting the correct answer, and (ii) producing linguistically accurate explanations. While Gemini 2.5 Pro reaches 83% accuracy, most open-weight models stay below 65%, and 48% of their explanations contain factual or pedagogical flaws according to expert review. A detailed error analysis pinpoints systematic weaknesses in morphology and in applying the latest DOOM3 orthographic norms. All data, code and a public web demo are released to catalyze future research. Our findings expose open challenges for trustworthy educational NLP in low-resource settings and establish GRILE as a new test-bed for controllable explanation generation and evaluation.

With the rapid advancement of image generation techniques, robust forgery detection has become increasingly imperative to ensure the trustworthiness of digital media. Recent research indicates that the learned semantic concepts of pre-trained models are critical for identifying fake images. However, the misalignment between the forgery and semantic concept spaces hinders the model's forgery detection performance. To address this problem, we propose a novel Semantic Discrepancy-aware Detector (SDD) that leverages reconstruction learning to align the two spaces at a fine-grained visual level. By exploiting the conceptual knowledge embedded in the pre-trained vision language model, we specifically design a semantic token sampling module to mitigate the space shifts caused by features irrelevant to both forgery traces and semantic concepts. A concept-level forgery discrepancy learning module, built upon a visual reconstruction paradigm, is proposed to strengthen the interaction between visual semantic concepts and forgery traces, effectively capturing discrepancies under the concepts' guidance. Finally, the low-level forgery feature enhancemer integrates the learned concept level forgery discrepancies to minimize redundant forgery information. Experiments conducted on two standard image forgery datasets demonstrate the efficacy of the proposed SDD, which achieves superior results compared to existing methods. The code is available at https://github.com/wzy1111111/SSD.

With the success of pre-trained language models (PLMs) in various application fields beyond natural language processing, language models have raised emerging attention in the field of time series forecasting (TSF) and have shown great prospects. However, current PLM-based TSF methods still fail to achieve satisfactory prediction accuracy matching the strong sequential modeling power of language models. To address this issue, we propose Cross-Model and Cross-Modality Learning with PLMs for time series forecasting (CC-Time). We explore the potential of PLMs for time series forecasting from two aspects: 1) what time series features could be modeled by PLMs, and 2) whether relying solely on PLMs is sufficient for building time series models. In the first aspect, CC-Time incorporates cross-modality learning to model temporal dependency and channel correlations in the language model from both time series sequences and their corresponding text descriptions. In the second aspect, CC-Time further proposes the cross-model fusion block to adaptively integrate knowledge from the PLMs and time series model to form a more comprehensive modeling of time series patterns. Extensive experiments on nine real-world datasets demonstrate that CC-Time achieves state-of-the-art prediction accuracy in both full-data training and few-shot learning situations. With the rapid growth of the Internet of Things, vast amounts of time series data are being generated, driving increasing interest in time series forecasting (TSF) Kaastra & Boyd (1996); Faloutsos et al. (2018). Current TSF methods primarily design specific modules to exploit the inherent knowledge of the time series data, and achieve good prediction accuracy Liu et al. (2024c); Nie et al. (2023), which we call time-series-specific models in this paper. Recently, pre-trained language models (PLMs) have demonstrated remarkable success across diverse fields Wang et al. (2024); Wu et al. (2024), prompting exploration in TSF Zhou et al. (2023); Jin et al. (2024a). Some approaches attempt to leverage the representation capacity and sequential modeling capability of PLMs to capture time series patterns for TSF, which we call PLM-based models Liu et al. (2024d). Although these methods show good prospects, they have not yet achieved satisfactory prediction accuracy, leaving an under-explored problem of how to effectively activate the potential of PLMs for TSF.

Quantization method plays a crucial role in improving model efficiency and reducing deployment costs, enabling the widespread application of deep learning models on resource-constrained devices. However, the quantization process inevitably introduces accuracy degradation. In this paper, we propose Mixture of Quantization Experts( abbr. MoQE), a quantization inference framework based on the Mixture-of-Experts (MoE) architecture, aiming to jointly improve the performance of quantization models. MoQE combines multiple quantization variants of one full-precision model as specialized "quantization experts" and dynamically routes input data to the most suitable expert based on its characteristics. MoQE alleviates the performance degradation commonly seen in single quantization models through specialization quantization expert models. We design lightweight, structure-aware router models tailored for both CV and NLP tasks. Experimental evaluations on ResNet, LLaMA, and Qwen model families across benchmark datasets including ImageNet, WikiText, C4, and OpenWebText demonstrate that MoQE achieves performance comparable to SOT A quantization model, without incurring significant increases in inference latency. Quantization method plays a pivotal role in the field of machine learning, particularly in enhancing model efficiency and reducing resource consumption. As deep learning models grow increasingly complex, their demand for computational resources escalates, constraining deployment on resource-limited devices and increasing operational costs. Furthermore, quantization method streamlines the model optimization pipeline, enabling developers to achieve efficient deployment within shorter timeframes and accelerating time-to-market for AI-driven products. Consequently, quantization method serves not only as a critical enabler for improving the accessibility and practicality of machine learning models but also as a key facilitator in the broader dissemination of artificial intelligence technologies. However, their practical deployment faces several critical challenges.

Autoregressive Language Models instantiate a factorized likelihood over token sequences, yet their strictly sequential decoding process imposes an intrinsic lower bound on inference latency. This bottleneck has emerged as a central obstacle to the scalable deployment of large-scale generative models. Existing acceleration techniques partially mitigate token-level latency by relying on auxiliary draft models or introducing an additional training phase, but fail to address the dominant memory and communication costs. We present READER, a provably lossless speculative decoding framework that bypasses the training of the auxiliary draft model. READER formalizes speculative decoding as a stochastic tree construction problem and exploits the empirical redundancy structure of natural language to generate high-probability candidate continuations. Our method revisits the problem of constructing draft trees, establishing substantial statistical improvements over stochastic draft-tree methods and providing a complexity-theoretic analysis that characterizes the optimality frontier of speculative decoding under bounded computation and memory resources. Beyond the single-sequence regime traditionally considered in prior work, we introduce a memory-optimal key-value cache-serving strategy that guarantees amortized sublinear overhead in the batch dimension, allowing READER to scale to realistic inference workloads. Comprehensive experiments demonstrate up to 6.13x wall-clock speedup on single-prompt inference and up to 5.92x on batched inference, consistently surpassing prior speculative decoding baselines, while preserving exact output equivalence, with even more pronounced gains in retrieval-augmented generation pipelines. Our results close a key gap between theoretical parallelism limits and practical LLM inference, suggesting a new standard for efficient deployment.

The rise of Omni-modal Large Language Models (OLLMs), which integrate visual and auditory processing with text, necessitates robust safety evaluations to mitigate harmful outputs. However, no dedicated benchmarks currently exist for OLLMs, and existing benchmarks fail to assess safety under joint audio-visual inputs or cross-modal consistency. To fill this gap, we introduce Omni-SafetyBench, the first comprehensive parallel benchmark for OLLM safety evaluation, featuring 24 modality variations with 972 samples each, including audio-visual harm cases. Considering OLLMs' comprehension challenges with complex omni-modal inputs and the need for cross-modal consistency evaluation, we propose tailored metrics: a Safety-score based on Conditional Attack Success Rate (C-ASR) and Refusal Rate (C-RR) to account for comprehension failures, and a Cross-Modal Safety Consistency score (CMSC-score) to measure consistency across modalities. Evaluating 6 open-source and 4 closed-source OLLMs reveals critical vulnerabilities: (1) only 3 models achieving over 0.6 in both average Safety-score and CMSC-score; (2) safety defenses weaken with complex inputs, especially audio-visual joints; (3) severe weaknesses persist, with some models scoring as low as 0.14 on specific modalities. Using Omni-SafetyBench, we evaluated existing safety alignment algorithms and identified key challenges in OLLM safety alignment: (1) Inference-time methods are inherently less effective as they cannot alter the model's underlying understanding of safety; (2) Post-training methods struggle with out-of-distribution issues due to the vast modality combinations in OLLMs; and, safety tasks involving audio-visual inputs are more complex, making even in-distribution training data less effective. Our proposed benchmark, metrics and the findings highlight urgent needs for enhanced OLLM safety. Omni-modal large language models (OLLMs) have advanced rapidly in understanding and generating content from integrated visual, audio, and text inputs. This enables them to handle complex tasks, such as describing audio-visual scenes or following voice instructions with visual context. Despite these advancements, ensuring their safety remains a critical concern that prevents these models from causing harm or acting in unethical, incorrect, or biased ways (Yi et al., 2024). Developing corresponding benchmarks serves as the cornerstone for reasonable assessment of safety evaluation. Numerous safety benchmarks have been established for text-only LLMs and vision-language models (Zhang et al., 2023; Liu et al., 2024b), and recent work has extended to specialized Omni-SafetyBench provides a large-scale dataset with diverse modal combinations and highlights cross-modal safety consistency as a key evaluation factor. However, for OLLMs capable of processing audio-visual joint inputs, there is currently a lack of benchmarks specifically designed to evaluate their safety.