Hao Mark Chen, Zhiwen Mo, Royson Lee, Qianzhou Wang, Da Li, Shell Xu Hu, Wayne Luk, Timothy Hospedales 等 9 位作者
Among parallel decoding paradigms, diffusion large language models (dLLMs) have emerged as a promising candidate that balances generation quality and throughput. However, their integration with Mixture-of-Experts (MoE) architectures is constrained by an expert explosion: as the number of tokens generated in parallel increases, the number of distinct experts activated grows nearly linearly. This results in substantial memory traffic that pushes inference into a memory-bound regime, negating the efficiency gains of both MoE and parallel decoding. To address this challenge, we propose Dynamic Expert Sharing (DES), a novel technique that shifts MoE optimization from token-centric pruning and conventional expert skipping methods to sequence-level coreset selection. To maximize expert reuse, DES identifies a compact, high-utility set of experts to satisfy the requirements of an entire parallel decoding block. We introduce two innovative selection strategies: (1) Intra-Sequence Sharing (DES-Seq), which adapts optimal allocation to the sequence level, and (2) Saliency-Aware Voting (DES-Vote), a novel mechanism that allows tokens to collectively elect a coreset based on aggregated router weights. Extensive experiments on MoE dLLMs demonstrate that DES reduces unique expert activations by over 55% and latency by up to 38%, while retaining 99% of vanilla accuracy, effectively decoupling memory overhead from the degree of parallelism.
Jixuan Ruan, Zhuo Cui, Zhengding Hu, Zhongkai Yu, Xiang Fang, Yue Guan, Jason Ludmir, Phillip E. Weinberg 等 13 位作者
Large Language Models (LLMs) offer an adaptable alternative to traditional quantum compilation heuristics, which bottleneck progress by demanding costly manual redesigns whenever hardware evolves. However, naive single-pass LLM approaches fail under the unique challenges introduced by quantum compilation. We introduce Qubrio, a practical agentic framework that significantly reduces heuristic redesigns through a three-level decomposition. By (1) partitioning programs into sequential operation stages, (2) deploying specialized agents for placement, routing and optimization, and (3) using disaggregated feedback for precise error correction, our system reliably navigates complex physical constraints. Beyond serving as a direct compiler, our LLM framework also uncovers novel strategies that facilitate convoy-style shuttling, substantially enhancing hardware concurrency without sacrificing fidelity. We further integrate these strategies into the previous state-of-the-art (SOTA) compiler, recovering a significant fraction of the LLM's performance advantages. On realistic workloads, our compiler achieves significant improvements in hardware runtime and program fidelity over the SOTA baseline PowerMove, seamlessly adapting to new hardware capabilities. Qubrio features an open user interface for practical use, and its code is publicly available at https://github.com/crosaa/Qubrio.
Aoxiang Fan, Corentin Dumery, Nicolas Talabot, Pascal Fua
Computing accurate geometry from multi-view images is a fundamental problem in computer vision. Recent feed-forward (FF) models jointly estimate 3D geometry and camera parameters, but they typically suffer from geometry distortion caused by reconstruction ambiguity, even when ground-truth camera parameters are supplied. In this paper, we study the multi-view stereo (MVS) problem with known camera parameters and propose a novel approach that bridges conventional MVS and FF methods. Rather than casting MVS as a sequence-to-one mapping that predicts depth only for a single reference view, we reformulate it as a sequence-to-sequence task, akin to FF models, that jointly predicts geometry for all input views. We introduce a global transformer-based architecture with two components that explicitly exploit camera-induced priors: ray-map embeddings that inject camera parameters into image patch tokens, making the transformer camera-aware, and a unified global cost volume that replaces conventional per-view cost volumes to jointly capture 3D structure across all views. Extensive experiments on multiple public benchmarks show our approach achieves state-of-the-art performance, surpassing both MVS and FF reconstruction baselines.
Byrne, Gluesing-Luerssen, and Ravagnani determined the number of codewords with a prescribed nonzero support in a linear maximum sum-rank distance code whose blocks have a common number of columns. Their formula depends only on the code parameters and the support dimensions. Evaluating it for arbitrary parameters gives a family of necessary nonnegativity conditions for existence. We prove their conjecture that this family is equivalent to the condition at the vector of total dimension $d+1$ obtained by filling the longest blocks first, where $d$ is the distance parameter. The resulting explicit criterion applies over every finite field $\mathbb F_q$, for all row numbers $n_i\leqslant m$ and $1\leqslant d<N$, where $m$ is the common number of columns and $N$ is the total number of rows. It also covers the boundary cases in which this vector fills each of its nonzero blocks completely, not explicitly included in the conjecture's displayed parameter range. The proof uses a generating function and an exchange identity to reduce the problem to vectors obtained by filling the longest blocks first. We then estimate alternating sums of elementary symmetric polynomials, with a separate argument for $q=2$. The dual conditions admit the same reduction. For codes that exist, the joint condition follows from sphere-packing bounds for the code and its dual. An infinite family of parameters satisfies the joint condition but admits no such linear code.
Bo Li Peking University, Mingda Wang Hebei University of Technology, Gexiang Fang, Shikun Zhang Peking University, Wei Ye Peking University
We revisit retrieval-augmented generation (RAG) by embedding retrieval control directly into generation. Instead of treating retrieval as an external intervention, we express retrieval decisions within token-level decoding, enabling end-to-end coordination without additional controllers or classifiers. Under the paradigm of Retrieval as Generation, we propose GRIP (Generation-guided Retrieval with Information Planning), a unified framework in which the model regulates retrieval behavior through control-token emission. Central to GRIP is Self-Triggered Information Planning, which allows the model to decide when to retrieve, how to reformulate queries, and when to terminate, all within a single autoregressive trajectory. This design tightly couples retrieval and reasoning and supports dynamic multi-step inference with on-the-fly evidence integration. To supervise these behaviors, we construct a structured training set covering answerable, partially answerable, and multi-hop queries, each aligned with specific token patterns. Experiments on five QA benchmarks show that GRIP surpasses strong RAG baselines and is competitive with GPT-4o while using substantially fewer parameters. Code and resources are provided in the supplementary materials.
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