LogitSpec: Accelerating Retrieval-based Speculative Decoding via Next Next Token Speculation

Speculative decoding (SD), where a small draft model is employed to propose draft tokens in advance and then the target model validates them in parallel, has emerged as a promising technique for LLM inference acceleration. Many endeavors to improve SD are to eliminate the need for a draft model and generate draft tokens in a retrieval-based manner in order to further alleviate the drafting overhead and significantly reduce the difficulty in deployment and applications. However, retrieval-based SD relies on a matching paradigm to retrieval the most relevant reference as the draft tokens, where these methods often fail to find matched and accurate draft tokens. To address this challenge, we propose LogitSpec to effectively expand the retrieval range and find the most relevant reference as drafts. Our LogitSpec is motivated by the observation that the logit of the last token can not only predict the next token, but also speculate the next next token. Specifically, LogitSpec generates draft tokens in two steps: (1) utilizing the last logit to speculate the next next token; (2) retrieving relevant reference for both the next token and the next next token. LogitSpec is training-free and plug-and-play, which can be easily integrated into existing LLM inference frameworks. Extensive experiments on a wide range of text generation benchmarks demonstrate that LogitSpec can achieve up to 2.61 $\times$ speedup and 3.28 mean accepted tokens per decoding step. Our code is available at https://github.com/smart-lty/LogitSpec.

CronusVLA: Transferring Latent Motion Across Time for Multi-Frame Prediction in Manipulation

Recent vision-language-action (VLA) models built on pretrained vision-language models (VLMs) have demonstrated strong generalization across manipulation tasks. However, they remain constrained by a single-frame observation paradigm and cannot fully benefit from the motion information offered by aggregated multi-frame historical observations, as the large vision-language backbone introduces substantial computational cost and inference latency. We propose CronusVLA, a unified framework that extends single-frame VLA models to the multi-frame paradigm through an efficient post-training stage. CronusVLA comprises three key components: (1) single-frame pretraining on large-scale embodied datasets with autoregressive action tokens prediction, which establishes an embodied vision-language foundation; (2) multi-frame encoding, adapting the prediction of vision-language backbones from discrete action tokens to motion features during post-training, and aggregating motion features from historical frames into a feature chunking; (3) cross-frame decoding, which maps the feature chunking to accurate actions via a shared decoder with cross-attention. By reducing redundant token computation and caching past motion features, CronusVLA achieves efficient inference. As an application of motion features, we further propose an action adaptation mechanism based on feature-action retrieval to improve model performance during finetuning. CronusVLA achieves state-of-the-art performance on SimplerEnv with 70.9% success rate, and 12.7% improvement over OpenVLA on LIBERO. Real-world Franka experiments also show the strong performance and robustness.

DRIFT: Dynamic Rule-Based Defense with Injection Isolation for Securing LLM Agents

Large Language Models (LLMs) are increasingly central to agentic systems due to their strong reasoning and planning capabilities. By interacting with external environments through predefined tools, these agents can carry out complex user tasks. Nonetheless, this interaction also introduces the risk of prompt injection attacks, where malicious inputs from external sources can mislead the agent's behavior, potentially resulting in economic loss, privacy leakage, or system compromise. System-level defenses have recently shown promise by enforcing static or predefined policies, but they still face two key challenges: the ability to dynamically update security rules and the need for memory stream isolation. To address these challenges, we propose DRIFT, a Dynamic Rule-based Isolation Framework for Trustworthy agentic systems, which enforces both control- and data-level constraints. A Secure Planner first constructs a minimal function trajectory and a JSON-schema-style parameter checklist for each function node based on the user query. A Dynamic Validator then monitors deviations from the original plan, assessing whether changes comply with privilege limitations and the user's intent. Finally, an Injection Isolator detects and masks any instructions that may conflict with the user query from the memory stream to mitigate long-term risks. We empirically validate the effectiveness of DRIFT on the AgentDojo benchmark, demonstrating its strong security performance while maintaining high utility across diverse models -- showcasing both its robustness and adaptability.

GENMANIP: LLM-driven Simulation for Generalizable Instruction-Following Manipulation

Robotic manipulation in real-world settings remains challenging, especially regarding robust generalization. Existing simulation platforms lack sufficient support for exploring how policies adapt to varied instructions and scenarios. Thus, they lag behind the growing interest in instruction-following foundation models like LLMs, whose adaptability is crucial yet remains underexplored in fair comparisons. To bridge this gap, we introduce GenManip, a realistic tabletop simulation platform tailored for policy generalization studies. It features an automatic pipeline via LLM-driven task-oriented scene graph to synthesize large-scale, diverse tasks using 10K annotated 3D object assets. To systematically assess generalization, we present GenManip-Bench, a benchmark of 200 scenarios refined via human-in-the-loop corrections. We evaluate two policy types: (1) modular manipulation systems integrating foundation models for perception, reasoning, and planning, and (2) end-to-end policies trained through scalable data collection. Results show that while data scaling benefits end-to-end methods, modular systems enhanced with foundation models generalize more effectively across diverse scenarios. We anticipate this platform to facilitate critical insights for advancing policy generalization in realistic conditions. Project Page: https://genmanip.axi404.top/.

Give Me FP32 or Give Me Death? Challenges and Solutions for Reproducible Reasoning

Large Language Models (LLMs) are now integral across various domains and have demonstrated impressive performance. Progress, however, rests on the premise that benchmark scores are both accurate and reproducible. We demonstrate that the reproducibility of LLM performance is fragile: changing system configuration such as evaluation batch size, GPU count, and GPU version can introduce significant difference in the generated responses. This issue is especially pronounced in reasoning models, where minor rounding differences in early tokens can cascade into divergent chains of thought, ultimately affecting accuracy. For instance, under bfloat16 precision with greedy decoding, a reasoning model like DeepSeek-R1-Distill-Qwen-7B can exhibit up to 9% variation in accuracy and 9,000 tokens difference in response length due to differences in GPU count, type, and evaluation batch size. We trace the root cause of this variability to the non-associative nature of floating-point arithmetic under limited numerical precision. This work presents the first systematic investigation into how numerical precision affects reproducibility in LLM inference. Through carefully controlled experiments across various hardware, software, and precision settings, we quantify when and how model outputs diverge. Our analysis reveals that floating-point precision -- while critical for reproducibility -- is often neglected in evaluation practices. Inspired by this, we develop a lightweight inference pipeline, dubbed LayerCast, that stores weights in 16-bit precision but performs all computations in FP32, balancing memory efficiency with numerical stability. Code is available at https://github.com/nanomaoli/llm_reproducibility.

MIRA: Medical Time Series Foundation Model for Real-World Health Data

A unified foundation model for medical time series -- pretrained on open access and ethics board-approved medical corpora -- offers the potential to reduce annotation burdens, minimize model customization, and enable robust transfer across clinical institutions, modalities, and tasks, particularly in data-scarce or privacy-constrained environments. However, existing generalist time series foundation models struggle to handle medical time series data due to their inherent challenges, including irregular intervals, heterogeneous sampling rates, and frequent missing values. To address these challenges, we introduce MIRA, a unified foundation model specifically designed for medical time series forecasting. MIRA incorporates a Continuous-Time Rotary Positional Encoding that enables fine-grained modeling of variable time intervals, a frequency-specific mixture-of-experts layer that routes computation across latent frequency regimes to further promote temporal specialization, and a Continuous Dynamics Extrapolation Block based on Neural ODE that models the continuous trajectory of latent states, enabling accurate forecasting at arbitrary target timestamps. Pretrained on a large-scale and diverse medical corpus comprising over 454 billion time points collect from publicly available datasets, MIRA achieves reductions in forecasting errors by an average of 10% and 7% in out-of-distribution and in-distribution scenarios, respectively, when compared to other zero-shot and fine-tuned baselines. We also introduce a comprehensive benchmark spanning multiple downstream clinical tasks, establishing a foundation for future research in medical time series modeling.

FuXi-Air: Urban Air Quality Forecasting Based on Emission-Meteorology-Pollutant multimodal Machine Learning

Air pollution has emerged as a major public health challenge in megacities. Numerical simulations and single-site machine learning approaches have been widely applied in air quality forecasting tasks. However, these methods face multiple limitations, including high computational costs, low operational efficiency, and limited integration with observational data. With the rapid advancement of artificial intelligence, there is an urgent need to develop a low-cost, efficient air quality forecasting model for smart urban management. An air quality forecasting model, named FuXi-Air, has been constructed in this study based on multimodal data fusion to support high-precision air quality forecasting and operated in typical megacities. The model integrates meteorological forecasts, emission inventories, and pollutant monitoring data under the guidance of air pollution mechanism. By combining an autoregressive prediction framework with a frame interpolation strategy, the model successfully completes 72-hour forecasts for six major air pollutants at an hourly resolution across multiple monitoring sites within 25-30 seconds. In terms of both computational efficiency and forecasting accuracy, it outperforms the mainstream numerical air quality models in operational forecasting work. Ablation experiments concerning key influencing factors show that although meteorological data contribute more to model accuracy than emission inventories do, the integration of multimodal data significantly improves forecasting precision and ensures that reliable predictions are obtained under differing pollution mechanisms across megacities. This study provides both a technical reference and a practical example for applying multimodal data-driven models to air quality forecasting and offers new insights into building hybrid forecasting systems to support air pollution risk warning in smart city management.

Serendipitous Recommendation with Multimodal LLM

Conventional recommendation systems succeed in identifying relevant content but often fail to provide users with surprising or novel items. Multimodal Large Language Models (MLLMs) possess the world knowledge and multimodal understanding needed for serendipity, but their integration into billion-item-scale platforms presents significant challenges. In this paper, we propose a novel hierarchical framework where fine-tuned MLLMs provide high-level guidance to conventional recommendation models, steering them towards more serendipitous suggestions. This approach leverages MLLM strengths in understanding multimodal content and user interests while retaining the efficiency of traditional models for item-level recommendation. This mitigates the complexity of applying MLLMs directly to vast action spaces. We also demonstrate a chain-of-thought strategy enabling MLLMs to discover novel user interests by first understanding video content and then identifying relevant yet unexplored interest clusters. Through live experiments within a commercial short-form video platform serving billions of users, we show that our MLLM-powered approach significantly improves both recommendation serendipity and user satisfaction.

AMoPO: Adaptive Multi-objective Preference Optimization without Reward Models and Reference Models

Existing multi-objective preference alignment methods for large language models (LLMs) face limitations: (1) the inability to effectively balance various preference dimensions, and (2) reliance on auxiliary reward/reference models introduces computational complexity. To address these challenges, we propose Adaptive Multi-objective Preference Optimization (AMoPO), a novel framework that achieves dynamic balance across preference dimensions. By introducing the multi-objective optimization paradigm to use the dimension-aware generation metrics as implicit rewards, AMoPO aligns LLMs with diverse preferences without additional reward models or reference models. We introduce an adaptive weight assignment mechanism that models the generation space as a Gaussian distribution, allowing dynamic prioritization of preference dimensions. Empirical results demonstrate that AMoPO outperforms state-of-the-art baselines by 28.5%, and the experiments on 7B, 14B, and 32B models reveal the scaling ability of AMoPO. Moreover, additional analysis of multiple dimensions verifies its adaptability and effectiveness. These findings validate AMoPO's capability to achieve dimension-aware preference alignment, highlighting its superiority. Our codes and datasets are available at https://github.com/Javkonline/AMoPO.

Hi-VAE: Efficient Video Autoencoding with Global and Detailed Motion

Recent breakthroughs in video autoencoders (Video AEs) have advanced video generation, but existing methods fail to efficiently model spatio-temporal redundancies in dynamics, resulting in suboptimal compression factors. This shortfall leads to excessive training costs for downstream tasks. To address this, we introduce Hi-VAE, an efficient video autoencoding framework that hierarchically encode coarse-to-fine motion representations of video dynamics and formulate the decoding process as a conditional generation task. Specifically, Hi-VAE decomposes video dynamics into two latent spaces: Global Motion, capturing overarching motion patterns, and Detailed Motion, encoding high-frequency spatial details. Using separate self-supervised motion encoders, we compress video latents into compact motion representations to reduce redundancy significantly. A conditional diffusion decoder then reconstructs videos by combining hierarchical global and detailed motions, enabling high-fidelity video reconstructions. Extensive experiments demonstrate that Hi-VAE achieves a high compression factor of 1428$\times$, almost 30$\times$ higher than baseline methods (e.g., Cosmos-VAE at 48$\times$), validating the efficiency of our approach. Meanwhile, Hi-VAE maintains high reconstruction quality at such high compression rates and performs effectively in downstream generative tasks. Moreover, Hi-VAE exhibits interpretability and scalability, providing new perspectives for future exploration in video latent representation and generation.