arXiv subjects

Longwen Wang

Publications and source records attributed to Longwen Wang.

3 recordsLinked to original sources

arXiv2026

LinearKV: One Cached State Suffices for Position-Independent Caching in Hybrid LLMs

LLM serving is increasingly accelerated by position-independent caching (PIC). Existing PIC methods, however, are built for full-attention models, where a token-indexed KV cache underlies its core operations: matching reusable token chunks, concatenating their KV entries, and selectively recomputing a few tokens to restore cross-chunk context. Hybrid LLMs break these primitives---they replace most attention layers with linear recurrences that expose only a fixed-size state, leaving no token-indexed KV to concatenate or to locally repair. This raises a natural question: can PIC benefit hybrid models, and what would it take? We present LinearKV, a training-free hybrid-PIC framework. Its key insight is a \emph{decoupled initialization}: each linear layer maps its $K$ matched local states to a single initial state, while full-attention layers concatenate their KV as before. LinearKV is therefore compatible with existing PIC methods, reusing their token selection and recomputation as-is. Under this framework, we find that a \emph{single cached state} suffices as the linear layer's initializer. The algebraically principled alternative---composing all $K$ cached states into the exact full-prefix state, as concurrent work HYPIC does---is unnecessary and, on some architectures, even harmful. We compare the two across three hybrid models and three PIC selectors. On the two GDN models the two tie, both recovering most of full quality (up to $92\%$); on the Mamba-2 model, exact composition instead collapses under every selector---under EPIC, for instance, it recovers only $46.6\%$ of full quality, versus $86.8\%$ for a single cached block initializer. A single state initializer is also cheaper, cutting time-to-first-token to $0.46\times$ full prefill versus a further $5$--$17\%$ overhead for exact composition; results hold across LongBench QA and RULER at 8K--32K.

cs.AI↗

arXiv2026

Beyond Binary: Turning Partial Success into Dense Verifiable Rewards for Reinforcement Learning in Code Generation

Effective reward design is a central challenge in Reinforcement Learning (RL) for code generation. Mainstream test-suite-level outcome rewards enforce functional correctness but induce sparsity, while external Reward Models (RMs) provide dense supervision at the cost of misalignment and additional overhead. Since code evaluation naturally yields multiple test-case-level outcomes, partial success, i.e., passing a subset of test cases, offers an intrinsic, verifiable source of dense supervision. In this paper, we propose VeRPO (Verifiable Dense Reward Policy Optimization), an RL framework that systematically turns verifiable partial success into reliable dense rewards. We analyze partial-success rewards using a weighted sum formulation, theoretically identifying a critical cardinality bias that causes policy updates to disproportionately favor gains from easy-test successes over progress on frontier tests. Based on this, VeRPO introduces a dynamic, density-calibrated local reward that explicitly corrects this bias and provides robust dense supervision from partial success. To enhance alignment with end-to-end functional correctness, VeRPO further integrates the local dense reward with global execution outcomes. Extensive experiments across diverse benchmarks and settings demonstrate that VeRPO outperforms outcome-driven and RM-based baselines, achieving up to +8.83 pass@1 gain with negligible time cost (< 0.02%) and zero GPU memory overhead.

cs.LG↗

arXiv2024

Optimizing Federated Graph Learning with Inherent Structural Knowledge and Dual-Densely Connected GNNs

Federated Graph Learning (FGL) is an emerging technology that enables clients to collaboratively train powerful Graph Neural Networks (GNNs) in a distributed manner without exposing their private data. Nevertheless, FGL still faces the challenge of the severe non-Independent and Identically Distributed (non-IID) nature of graphs, which possess diverse node and edge structures, especially across varied domains. Thus, exploring the knowledge inherent in these structures becomes significantly crucial. Existing methods, however, either overlook the inherent structural knowledge in graph data or capture it at the cost of significantly increased resource demands (e.g., FLOPs and communication bandwidth), which can be detrimental to distributed paradigms. Inspired by this, we propose FedDense, a novel FGL framework that optimizes the utilization efficiency of inherent structural knowledge. To better acquire knowledge of diverse and underexploited structures, FedDense first explicitly encodes the structural knowledge inherent within graph data itself alongside node features. Besides, FedDense introduces a Dual-Densely Connected (DDC) GNN architecture that exploits the multi-scale (i.e., one-hop to multi-hop) feature and structure insights embedded in the aggregated feature maps at each layer. In addition to the exploitation of inherent structures, we consider resource limitations in FGL, devising exceedingly narrow layers atop the DDC architecture and adopting a selective parameter sharing strategy to reduce resource costs substantially. We conduct extensive experiments using 15 datasets across 4 different domains, demonstrating that FedDense consistently surpasses baselines by a large margin in training performance, while demanding minimal resources.

cs.LG↗