MambaCSP is a hybrid-attention state space model that replaces transformer-based backbones with a linear-time Mamba architecture for channel state prediction. By adding lightweight patch‑mixer attention layers, it captures long‑range dependencies while maintaining hardware efficiency. Experiments on MISO‑OFDM show 9‑12% higher accuracy, 3× faster throughput, 2.6× lower VRAM usage, and 2.9× faster inference compared to LLM‑based methods.
By Aladin Djuhera, Haris Gacanin, Holger Boche
arXiv:2609.37362v1 Announce Type: new
Abstract: Large language model (LLM) routing aims to assign each query to the most suitable model from a heterogeneous candidate pool, improving the quality--eff...
By Guannan Lai, Han-Jia Ye
arXiv:2609.08189v1 Announce Type: new
Abstract: Dynamic layer routing reduces the inference cost of Large Language Models (LLMs) by learning to skip layers for individual tokens. Existing methods, ho...
By Hongjin Lin, Wentao Wan, Keze Wang
arXiv:2606. 09514v1 Announce Type: new Abstract: Large language models (LLMs) incur high inference cost due to their depth and parameter scale.
By Yuhua Zhou, Shaoqi Yu, Shichao Weng, Changhai Zhou, Mingze Yin, Fei Yang, Aimin Pan
TurboMamba is a new language‑model architecture that combines a global Mamba‑3 model with a small Transformer to achieve faster inference, faster training, and better token prediction. It introduces three lightweight enhancements—direct access to the previous block’s last token, a small correction to next‑token scores, and a learned mixture with recent‑token frequencies—to improve local predictions. In experiments on FineWeb‑Edu, TurboMamba outperforms Transformer, Mamba‑3, Jamba, and Zamba, achieving the lowest mean test prediction loss and significantly higher training throughput and batch latency speedups.
By Hanxin Zhang
arXiv:2607. 11796v1 Announce Type: new Abstract: Selective state-space models such as Mamba route information through a bank of first-order modes whose input coupling is set by a learned selection mechanism.
By Raktim Bhattacharya
arXiv:2608. 05872v1 Announce Type: cross Abstract: Standard Large Language Models (LLMs) execute layers sequentially.
By Pawe{\l} Batorski, Abtin Pourhadi, Akylgali Aitaza, Przemys{\l}aw Spurek, Paul Swoboda
CRISP (Cliff-awaRe Input-adaptive Sparse Prefilling) is a new method for long-context LLM inference that replaces costly quadratic attention prefilling with a dynamic, input-adaptive sparse routing scheme. It introduces a structural proxy, C_struct, to directly read routing decisions from the proxy attention map, eliminating the need for pooled matrix multiplication and KL divergence. Additionally, CRISP addresses the post-softmax mass cliff by using a sink-aware threshold based on the noise floor, theoretically reducing background noise accumulation to O(n). Empirical results on InfiniteBench, RULER, and LongBench show that CRISP outperforms existing sparse methods and can match or exceed exact dense attention, achieving up to a 5.30× speedup at 512k tokens and significant gains on retrieval-heavy tasks.
By Huu Huy Nguyen, Chien Van Nguyen, Franck Dernoncourt, Ryan A. Rossi, Linh Ngo Van, Jieyang Chen, Thien Huu Nguyen
Elastic Spectral State Space Models (ES-SSM) are a train‑once, export‑many sequence modeling framework that achieves elasticity by spectrally approximating the state‑space operator. The method builds on Hankel spectral filtering, using fixed spectral channels to represent long‑range token mixing and combining input‑adaptive gates with budget dropout to enable reliable deployment across different resource budgets. ES‑SSM is evaluated on byte‑level language modeling, Long Range Arena, Speech Commands V2, and offline reinforcement learning, showing that a single trained model can be truncated to competitive compact models while maintaining smooth quality‑cost curves across a wide range of truncation levels.
By Dachuan Song, Junyu Yin, Zechen Hu, Xuan Wang
The paper investigates how attention dynamics evolve across recurrent depth in language models, finding that attention support stabilizes early while hidden states and outputs take longer. It proposes WISE, a training‑free method that uses full attention in early steps and then reuses the discovered sparse working set for later steps, preserving performance on multi‑hop QA tasks. Experiments show that WISE maintains quality up to 2K context, offers measurable speedups, and highlights the importance of recurrent discovery of attention support.
By Ke Wan, Chen Chen
arXiv:2606. 06467v1 Announce Type: cross Abstract: Long-context inference in modern LLMs is increasingly constrained by decoding efficiency, especially in reasoning-heavy settings where models generate long intermediate chains of thought.
By Yutao Sun, Yanqi Zhang, Li Dong, Jianyong Wang, Furu Wei
Multi-State Linear Attention (MSLA) improves long‑sequence modeling by compressing context into separate memory states, but its aggregation only uses mean alignment, ignoring distribution within segments. MERLIN (MomEnt Routing for multi‑state LINear attention) enhances this by tracking first and second moments of each segment’s keys, enabling queries to recover both mean and variance of alignment for more content‑aware aggregation. Experiments on Gated DeltaNet‑1.3B and Mamba‑2‑780M demonstrate that MERLIN consistently outperforms existing aggregation methods on long‑context tasks, achieving significant accuracy gains while using less memory.
By Hee Seung Hwang, Sanghyuk Chun, Xindi Wu, Olga Russakovsky, Zhiwei Deng