Hybrid large language models (LLMs) have emerged as a promising way to combine the modeling capability of the Attention mechanism with the efficiency of state space models (SSMs). However, most existing hybrid LLMs still inherit a static design philosophy: once the Attention-to-SSM layout is chosen, every request must follow the same hybrid execution. This one-layout-fits-all design is inefficient for real-world serving workloads, where different requests rely on contextual information in different ways. This paper presents Morph, a serving system that co-designs the model architecture and runtime for request-adaptive hybrid LLM inference. Its model architecture pairs Attention and SSM layers within each block and uses prefill-oriented gating to select a request-specific layout that remains fixed throughout decoding. Its serving runtime preserves batching efficiency under divergent layer choices through selective batching and enables correct prefix reuse through hybrid-aware context caching. Experiments on six datasets show that Morph achieves a better quality-efficiency trade-off than existing static hybrid baselines. System-level evaluation further shows that Morph improves the serving throughput by up to 3.87x over baselines while substantially reducing inference latency.
Haoxi Li, Sikai Bai, Jianfei Ma, Qinglin Hou, Xiaocheng Lu, Tao Han, Jie Zhang, Song Guo
Learning from experience is essential for building the next generation of agents that continually expand their capabilities. Pretrained vision-language models (VLMs) already internalize rich world knowledge, yet this knowledge is rarely leveraged to drive policy improvement through online interaction. Realizing this potential efficiently remains challenging, as policy updates shift the agent's interaction distribution, requiring the world model to adapt accordingly, while continually updating large pretrained models is computationally expensive. To address this challenge, we introduce Dwell, a model-based reinforcement learning (RL) framework that reframes a VLM as a recurrent latent transition operator, endowing it with the ability to predict, adjust, and act in latent space. Leveraging the VLM's own representations and its dual role as both world model and policy, the agent continually learns from real interactions to anticipate and evaluate action outcomes, and from latent imagination to improve its behavior. Across five interactive visual environments, Dwell achieves the strongest overall performance among trainable methods while substantially reducing the compute required for imagination, showing that effective world models for VLM agents need not reconstruct the future; they need to predict futures that are useful for action. Our code is publicly available at https://github.com/Tom111-lhx/Dwell
Yuanming Shao, Yuming Feng, Weizhe Zhang, Bin Xiao, Song Guo, Zejun Wang
Payment channel hubs (PCHs) enable private cross-chain payments, but existing protocols remain difficult to extend across heterogeneous blockchains. Their payment flows often depend on how particular blockchains represent transactions, authorize channel updates, and settle balances. Some protocols do not explain how their channels are implemented on actual blockchains. Supporting a new ledger or execution model requires a corresponding channel realization and may also require protocol adaptation and security analysis. We present UniPCH, a framework with two layers that separates native channel execution from private cross-chain payment logic. The lower UniPC layer defines common state transitions and security requirements. It separates task authorization from confirmation of the complete channel state. We provide realizations for UTXO ledgers and account ledgers that support smart contracts. At the upper layer, we develop selective re-signing blind conditional signatures (S-RBCS) to coordinate conditional authorization across hubs. Using S-RBCS, we construct a payment protocol that coordinates different channels in four phases. Our analysis establishes local UniPC security, griefing resistance, economic atomicity, and session unlinkability. We implement a Go prototype and evaluate the running time and bandwidth of each phase under different network conditions and system parameters. Results show modest computational overhead from the common interfaces. Most evaluated configurations run faster than comparable schemes under network delay. A full refresh ablation shows that reusing valid task evidence reduces computation and communication costs in the account realization.
Haoyue Zhang, Hualei Zhang The Hong Kong University of Science and Technology, Xiaosong Ma Hong Kong Polytechnic University, Jie Zhang Hong Kong University of Science and Technology, Song Guo Department of Computer Science and Engineering, Hong Kong University of Science and Technology
Large Language Models (LLMs) exhibit enhanced capabilities by Chain-of-Thought reasoning. However, the extended reasoning sequences introduce significant GPU memory overhead due to increased key-value (KV) cache. Existing KV cache compression methods mitigate memory bottlenecks but struggle in long reasoning tasks. In this paper, we analyze attention patterns in reasoning tasks and reveal a Token Importance Recurrence phenomenon: a large proportion of tokens regain high attention after multiple decoding steps, which is failed to capture by existing works and may lead to unpredictable eviction on such periodically critical tokens. To address this, we propose LazyEviction, an observation window-based lagged eviction framework retaining latent recurring tokens by prioritized eviction based on tokens’ recurrence patterns. Extensive experiments demonstrate that LazyEviction reduces KV cache by 50% 70% while maintaining comparable accuracy, outperforming existing KV cache baselines. Our implementation code can be found at https://github.com/Halo-949/LazyEviction.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Yongjiang Liu, Haoxi Li The Hong Kong University of Science and Technology, Xiaosong Ma Hong Kong Polytechnic University, Jie Zhang Hong Kong University of Science and Technology, Song Guo Department of Computer Science and Engineering, Hong Kong University of Science and Technology
Recent Large Reasoning Models (LRMs) excel at complex reasoning tasks but often suffer from overthinking, generating overly long and redundant reasoning trajectories. To explore its essence, our empirical analysis reveals that LRMs are primarily limited to recognizing task properties (i.e., difficulty levels) like humans before solving the problem, leading to a one-size-fits-all reasoning strategy. This observation motivates a fundamental question: Can we explicitly bootstrap such ability to alleviate overthinking in LRMs? To this end, we propose Think-How-to-Think (TH2T), a novel two-stage fine-tuning strategy that progressively inspires LRMs’ difficulty cognition and redundancy cognition of LRMs. Specifically, we first inject Difficulty Dypnosis into output prefixes as cues for global, prospective reasoning strategy selection, stimulating the model’s sharper sensitivity to task complexity and adaptive control of reasoning depth. Then, we incorporate Redundancy Hypnosis into in-progress reasoning steps, which serve as local, retrospective signals for behavior correction by identifying and eliminating superfluous reasoning detours. Experiments across 7B/14B/32B models demonstrate that TH2T significantly reduces inference costs by over 70% on easy tasks and 40% on complex ones without compromising performance. The resultant models exhibit a nascent ability for difficulty-aware reasoning, effectively mitigating behaviors like excessive reflection and looping, thereby paving the way for more cognitively efficient LRMs.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表