ForesightKV: Optimizing KV Cache Eviction for Reasoning Models by Learning Long-Term Contribution

Recently, large language models (LLMs) have shown remarkable reasoning abilities by producing long reasoning traces. However, as the sequence length grows, the key-value (KV) cache expands linearly, incurring significant memory and computation costs. Existing KV cache eviction methods mitigate this issue by discarding less important KV pairs, but often fail to capture complex KV dependencies, resulting in performance degradation. To better balance efficiency and performance, we introduce ForesightKV, a training-based KV cache eviction framework that learns to predict which KV pairs to evict during long-text generations. We first design the Golden Eviction algorithm, which identifies the optimal eviction KV pairs at each step using future attention scores. These traces and the scores at each step are then distilled via supervised training with a Pairwise Ranking Loss. Furthermore, we formulate cache eviction as a Markov Decision Process and apply the GRPO algorithm to mitigate the significant language modeling loss increase on low-entropy tokens. Experiments on AIME2024 and AIME2025 benchmarks of three reasoning models demonstrate that ForesightKV consistently outperforms prior methods under only half the cache budget, while benefiting synergistically from both supervised and reinforcement learning approaches.

DexTac: Learning Contact-aware Visuotactile Policies via Hand-by-hand Teaching

For contact-intensive tasks, the ability to generate policies that produce comprehensive tactile-aware motions is essential. However, existing data collection and skill learning systems for dexterous manipulation often suffer from low-dimensional tactile information. To address this limitation, we propose DexTac, a visuo-tactile manipulation learning framework based on kinesthetic teaching. DexTac captures multi-dimensional tactile data-including contact force distributions and spatial contact regions-directly from human demonstrations. By integrating these rich tactile modalities into a policy network, the resulting contact-aware agent enables a dexterous hand to autonomously select and maintain optimal contact regions during complex interactions. We evaluate our framework on a challenging unimanual injection task. Experimental results demonstrate that DexTac achieves a 91.67% success rate. Notably, in high-precision scenarios involving small-scale syringes, our approach outperforms force-only baselines by 31.67%. These results underscore that learning multi-dimensional tactile priors from human demonstrations is critical for achieving robust, human-like dexterous manipulation in contact-rich environments.

Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts

Hybrid Transformer architectures, which combine softmax attention blocks and recurrent neural networks (RNNs), have shown a desirable performance-throughput tradeoff for long-context modeling, but their adoption and studies are hindered by the prohibitive cost of large-scale pre-training from scratch. Some recent studies have shown that pre-trained softmax attention blocks can be converted into RNN blocks through parameter transfer and knowledge distillation. However, these transfer methods require substantial amounts of training data (more than 10B tokens), and the resulting hybrid models also exhibit poor long-context performance, which is the scenario where hybrid models enjoy significant inference speedups over Transformer-based models. In this paper, we present HALO (Hybrid Attention via Layer Optimization), a pipeline for distilling Transformer models into RNN-attention hybrid models. We then present HypeNet, a hybrid architecture with superior length generalization enabled by a novel position encoding scheme (named HyPE) and various architectural modifications. We convert the Qwen3 series into HypeNet using HALO, achieving performance comparable to the original Transformer models while enjoying superior long-context performance and efficiency. The conversion requires just 2.3B tokens, less than 0.01% of their pre-training data

SG-CADVLM: A Context-Aware Decoding Powered Vision Language Model for Safety-Critical Scenario Generation

Autonomous vehicle safety validation requires testing on safety-critical scenarios, but these events are rare in real-world driving and costly to test due to collision risks. Crash reports provide authentic specifications of safety-critical events, offering a vital alternative to scarce real-world collision trajectory data. This makes them valuable sources for generating realistic high-risk scenarios through simulation. Existing approaches face significant limitations because data-driven methods lack diversity due to their reliance on existing latent distributions, whereas adversarial methods often produce unrealistic scenarios lacking physical fidelity. Large Language Model (LLM) and Vision Language Model (VLM)-based methods show significant promise. However, they suffer from context suppression issues where internal parametric knowledge overrides crash specifications, producing scenarios that deviate from actual accident characteristics. This paper presents SG-CADVLM (A Context-Aware Decoding Powered Vision Language Model for Safety-Critical Scenario Generation), a framework that integrates Context-Aware Decoding with multi-modal input processing to generate safety-critical scenarios from crash reports and road network diagrams. The framework mitigates VLM hallucination issues while enabling the simultaneous generation of road geometry and vehicle trajectories. The experimental results demonstrate that SG-CADVLM generates critical risk scenarios at a rate of 84.4% compared to 12.5% for the baseline methods, representing an improvement of 469%, while producing executable simulations for autonomous vehicle testing.

Explaining Synergistic Effects in Social Recommendations

In social recommenders, the inherent nonlinearity and opacity of synergistic effects across multiple social networks hinders users from understanding how diverse information is leveraged for recommendations, consequently diminishing explainability. However, existing explainers can only identify the topological information in social networks that significantly influences recommendations, failing to further explain the synergistic effects among this information. Inspired by existing findings that synergistic effects enhance mutual information between inputs and predictions to generate information gain, we extend this discovery to graph data. We quantify graph information gain to identify subgraphs embodying synergistic effects. Based on the theoretical insights, we propose SemExplainer, which explains synergistic effects by identifying subgraphs that embody them. SemExplainer first extracts explanatory subgraphs from multi-view social networks to generate preliminary importance explanations for recommendations. A conditional entropy optimization strategy to maximize information gain is developed, thereby further identifying subgraphs that embody synergistic effects from explanatory subgraphs. Finally, SemExplainer searches for paths from users to recommended items within the synergistic subgraphs to generate explanations for the recommendations. Extensive experiments on three datasets demonstrate the superiority of SemExplainer over baseline methods, providing superior explanations of synergistic effects.

PEAfowl: Perception-Enhanced Multi-View Vision-Language-Action for Bimanual Manipulation

Bimanual manipulation in cluttered scenes requires policies that remain stable under occlusions, viewpoint and scene variations. Existing vision-language-action models often fail to generalize because (i) multi-view features are fused via view-agnostic token concatenation, yielding weak 3D-consistent spatial understanding, and (ii) language is injected as global conditioning, resulting in coarse instruction grounding. In this paper, we introduce PEAfowl, a perception-enhanced multi-view VLA policy for bimanual manipulation. For spatial reasoning, PEAfowl predicts per-token depth distributions, performs differentiable 3D lifting, and aggregates local cross-view neighbors to form geometrically grounded, cross-view consistent representations. For instruction grounding, we propose to replace global conditioning with a Perceiver-style text-aware readout over frozen CLIP visual features, enabling iterative evidence accumulation. To overcome noisy and incomplete commodity depth without adding inference overhead, we apply training-only depth distillation from a pretrained depth teacher to supervise the depth-distribution head, providing perception front-end with geometry-aware priors. On RoboTwin 2.0 under domain-randomized setting, PEAfowl improves the strongest baseline by 23.0 pp in success rate, and real-robot experiments further demonstrate reliable sim-to-real transfer and consistent improvements from depth distillation. Project website: https://peafowlvla.github.io/.

Graph-Anchored Knowledge Indexing for Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) has emerged as a dominant paradigm for mitigating hallucinations in Large Language Models (LLMs) by incorporating external knowledge. Nevertheless, effectively integrating and interpreting key evidence scattered across noisy documents remains a critical challenge for existing RAG systems. In this paper, we propose GraphAnchor, a novel Graph-Anchored Knowledge Indexing approach that reconceptualizes graph structures from static knowledge representations into active, evolving knowledge indices. GraphAnchor incrementally updates a graph during iterative retrieval to anchor salient entities and relations, yielding a structured index that guides the LLM in evaluating knowledge sufficiency and formulating subsequent subqueries. The final answer is generated by jointly leveraging all retrieved documents and the final evolved graph. Experiments on four multi-hop question answering benchmarks demonstrate the effectiveness of GraphAnchor, and reveal that GraphAnchor modulates the LLM's attention to more effectively associate key information distributed in retrieved documents. All code and data are available at https://github.com/NEUIR/GraphAnchor.

LongCat-Flash-Thinking-2601 Technical Report

We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves state-of-the-art performance among open-source models on a wide range of agentic benchmarks, including agentic search, agentic tool use, and tool-integrated reasoning. Beyond benchmark performance, the model demonstrates strong generalization to complex tool interactions and robust behavior under noisy real-world environments. Its advanced capability stems from a unified training framework that combines domain-parallel expert training with subsequent fusion, together with an end-to-end co-design of data construction, environments, algorithms, and infrastructure spanning from pre-training to post-training. In particular, the model's strong generalization capability in complex tool-use are driven by our in-depth exploration of environment scaling and principled task construction. To optimize long-tailed, skewed generation and multi-turn agentic interactions, and to enable stable training across over 10,000 environments spanning more than 20 domains, we systematically extend our asynchronous reinforcement learning framework, DORA, for stable and efficient large-scale multi-environment training. Furthermore, recognizing that real-world tasks are inherently noisy, we conduct a systematic analysis and decomposition of real-world noise patterns, and design targeted training procedures to explicitly incorporate such imperfections into the training process, resulting in improved robustness for real-world applications. To further enhance performance on complex reasoning tasks, we introduce a Heavy Thinking mode that enables effective test-time scaling by jointly expanding reasoning depth and width through intensive parallel thinking.

Blockchain-Based Spectrum Resource Securitization via Semi-Fungible Token-Lock

As 6G networks evolve, spectrum assets require flexible, dynamic, and efficient utilization, motivating blockchain based spectrum securitization. Existing approaches based on ERC404 style hybrid token models rely on frequent minting and burning during asset transfers, which disrupt token identity continuity and increase on chain overhead. This paper proposes the Semi Fungible Token Lock (SFT Lock) method, a lock/unlock based mechanism that preserves NFT identity and historical traceability while enabling fractional ownership and transferability. By replacing mint/burn operations with deterministic state transitions, SFT Lock ensures consistent lifecycle representation of spectrum assets and significantly reduces on chain operations. Based on this mechanism, a modular smart contract architecture is designed to support spectrum authorization, securitization, and sharing, and a staking mechanism is introduced to enhance asset liquidity. Experimental results on a private Ethereum network demonstrate that, compared with ERC404 style hybrid token models, the proposed method achieves substantial gas savings while maintaining functional correctness and traceability.

Language-Coupled Reinforcement Learning for Multilingual Retrieval-Augmented Generation

Multilingual retrieval-augmented generation (MRAG) requires models to effectively acquire and integrate beneficial external knowledge from multilingual collections. However, most existing studies employ a unitive process where queries of equivalent semantics across different languages are processed through a single-turn retrieval and subsequent optimization. Such a ``one-size-fits-all'' strategy is often suboptimal in multilingual settings, as the models occur to knowledge bias and conflict during the interaction with the search engine. To alleviate the issues, we propose LcRL, a multilingual search-augmented reinforcement learning framework that integrates a language-coupled Group Relative Policy Optimization into the policy and reward models. We adopt the language-coupled group sampling in the rollout module to reduce knowledge bias, and regularize an auxiliary anti-consistency penalty in the reward models to mitigate the knowledge conflict. Experimental results demonstrate that LcRL not only achieves competitive performance but is also appropriate for various practical scenarios such as constrained training data and retrieval over collections encompassing a large number of languages. Our code is available at https://github.com/Cherry-qwq/LcRL-Open.