Linear Attention

Linearizing pretrained large language models (LLMs) primarily relies on intra-layer hybrid attention mechanisms to alleviate the quadratic complexity of standard softmax attention. Existing methods perform token routing based on sliding-window partitions, resulting in position-based selection and fails to capture token-specific global importance. Meanwhile, linear attention further suffers from distribution shift caused by learnable feature maps that distort pretrained feature magnitudes. Motivated by these limitations, we propose STILL, an intra-layer hybrid linearization framework for efficiently linearizing LLMs. STILL introduces a Self-Saliency Score with strong local-global consistency, enabling accurate token selection using sliding-window computation, and retains salient tokens for sparse softmax attention while summarizing the remaining context via linear attention. To preserve pretrained representations, we design a Norm-Preserved Feature Map (NP-Map) that decouples feature direction from magnitude and reinjects pretrained norms. We further adopt a unified training-inference architecture with chunk-wise parallelization and delayed selection to improve hardware efficiency. Experiments show that STILL matches or surpasses the original pretrained model on commonsense and general reasoning tasks, and achieves up to a 86.2% relative improvement over prior linearized attention methods on long-context benchmarks. Code is available at this URL.

In this paper, we introduce Exact Flow Linear Attention~(EFLA), an exact-flow formulation of delta-rule linear attention. We show that the delta-rule update can be interpreted as an explicit Euler discretization of an underlying continuous-time system. EFLA replaces this first-order update with the exact closed-form flow. By exploiting the rank-1 structure of the dynamics matrix, both the matrix exponential and the input integral collapse to a simple update that preserves delta-rule linear attention's algebraic structure, parameter count, linear-time complexity, and chunkwise parallelism. This attention mechanism removes the Euler discretization error of the delta-rule dynamics without introducing additional parameters. Experiments on robustness tests, language modeling benchmarks, and the MAD synthetic benchmark show that EFLA improves stability under corrupted and high-energy inputs, reduces perplexity, and achieves stronger downstream performance compared to SSM and Euler-style baselines. These results establish exact-flow integration as a principled and scalable update mechanism for delta-rule linear attention.

Position information is essential for language modeling. In softmax transformers, Rotary Position Embeddings (RoPE) encode positions through fixed-angle rotations, while in linear transformers, order is handled via input-dependent (selective) gating that decays past key-value associations. Selectivity has generally been shown to improve language-related tasks. Inspired by this, we introduce Selective RoPE, an input-dependent rotary embedding mechanism, that generalizes RoPE, and enables rotation in arbitrary angles for both linear and softmax transformers. We show that softmax attention already performs a hidden form of these rotations on query-key pairs, uncovering an implicit positional structure. We further show that in state-space models and gated linear transformers, the real part manages forgetting while the imaginary part encodes positions through rotations. We validate our method by equipping gated transformers with Selective RoPE, demonstrating that its input-dependent rotations improve performance in language modeling and on difficult sequence tasks like copying, state tracking, and retrieval.

Attention mechanisms are effective for multivariate time series modeling because they adaptively aggregate information across timestamps and variables. Their quadratic cost, however, limits their use in multivariate and spatio-temporal systems. Linear attention reduces this cost by replacing nonlinear attention mappings with forms that support associative computation. Replacing exponential weighting may weaken the amplification of high-score positions and further bias weakly selective distributions toward uniform aggregation. We propose an entropy-calibrated linear attention module and instantiate it as the ELinFormer forecaster. Our operator decomposes aggregation into a uniform component and a query-dependent correction, whose magnitude is calibrated by a query-specific temperature. We derive a second-order KL proxy from centered query-key scores. The Taylor expansion reduces the proxy to the squared-score energy divided by the squared temperature. This score energy can be computed from shared products of the centered keys without forming the full pairwise score matrix, and the same products support the linear attention output. We design a smooth temperature rule to calibrate the KL proxy: it keeps low-energy residuals vanishing, approximately calibrates moderate-energy queries to a shared KL level, and allows the proxy to continue growing for high-energy queries. Experiments on five PEMS benchmarks show competitive forecasting accuracy. Controlled comparisons with vanilla attention and Linformer demonstrate a favorable accuracy--memory trade-off, while analysis on PEMS08 shows that the proposed rational temperature rule and query-specific calibration achieve lower MAE than a linear rule or shared-temperature alternatives. These results support entropy calibration as an effective mechanism for preserving selective deviations in linear attention.

Scaling large recommendation systems requires advancing three major frontiers: processing longer user histories, expanding candidate sets, and increasing model capacity. While promising, transformers' computational cost scales quadratically with the user sequence length and linearly with the number of candidates. This trade-off makes it prohibitively expensive to expand candidate sets or increase sequence length at inference, despite the significant performance improvements. We introduce LIME, a novel architecture that resolves this trade-off. Through two key innovations, LIME fundamentally reduces computational complexity. First, low-rank ``link embeddings" enable pre-computation of attention weights by decoupling user and candidate interactions, making the inference cost nearly independent of candidate set size. Second, a linear attention mechanism, LIME-XOR, reduces the complexity with respect to user sequence length from quadratic ($O(N^2)$) to linear ($O(N)$). Experiments on public and industrial datasets show LIME achieves near-parity with state-of-the-art transformers but with a 10$\times$ inference speedup on large candidate sets or long sequence lengths. When tested on a major recommendation platform, LIME improved user engagement while maintaining minimal inference costs with respect to candidate set size and user history length, establishing a new paradigm for efficient and expressive recommendation systems.

Standard attention mechanisms are not well suited to stereo matching. Global attention scales quadratically and provides no explicit matching constraint, while local attention is efficient but loses long-range correspondences. We propose MatchAttention, an attention mechanism that embeds an explicit matching constraint into attention by treating the relative position between a query and its matched key as a learnable component of attention sampling. Centering a small contiguous sampling window on this learnable relative position enforces the matching constraint and supports long-range correspondence at strictly linear attention complexity. A differentiable contiguous attention sampling (CAS) operator enables sub-pixel accuracy, and cascaded MatchAttention blocks iteratively refine the relative positions through residual connections. We instantiate MatchAttention as a hierarchical coarse-to-fine stereo network with two variants. MatchAttentionXL targets accuracy and MatchAttentionRT targets real-time edge inference. MatchAttentionXL achieves state-of-the-art accuracy on Middlebury V3 and top results across KITTI 2012/2015 and ETH3D. MatchAttentionRT runs at 9.3 ms on RTX 4060 Ti and 79.1 ms on Jetson Orin NX 16 GB at 1024 x 512, making it the first stereo model to deliver real-time edge inference without sacrificing zero-shot generalization. The code is available at https://github.com/TingmanYan/MatchAttention.

Recent progress in LLM agents has advanced the prospect of autonomous research. Yet whether AI can complete difficult long-horizon tasks, especially those that advance AI research itself, remains largely unexplored. We present ASI-Arch, a system for AI-driven AI research that autonomously conducts neural architecture research through a closed-loop research-experiment-analyze-update process. Applied to linear attention, ASI-Arch ran 1,773 iterative experiments and discovered 105 state-of-the-art architectures. Its best architecture improves over DeltaNet by nearly three times the gain achieved by Mamba2. Beyond the final performance gains, we analyze the contributions of different parts of the framework in this hard research setting, shedding light on what enables autonomous progress in complex AI research tasks.

Transformers face quadratic complexity and memory issues with long sequences, prompting the adoption of linear attention mechanisms using fixed-size hidden states. However, linear models often suffer from limited recall performance, leading to hybrid architectures that combine linear and full attention layers. Despite extensive hybrid architecture research, the choice of linear attention component has not been deeply explored. We systematically evaluate various linear attention models across generations - vector recurrences to advanced gating mechanisms - both standalone and hybridized. To enable this comprehensive analysis, we trained and open-sourced 72 models: 36 at 340M parameters (20B tokens) and 36 at 1.3B parameters (100B tokens), covering six linear attention variants across five hybridization ratios. Benchmarking on standard language modeling and recall tasks reveals that superior standalone linear models do not necessarily excel in hybrids. While language modeling remains stable across linear-to-full attention ratios, recall significantly improves with increased full attention layers, particularly below a 3:1 ratio. Our study highlights selective gating, hierarchical recurrence, and controlled forgetting as critical for effective hybrid models. We recommend architectures such as HGRN-2 or GatedDeltaNet with a linear-to-full ratio between 3:1 and 6:1 to achieve Transformer-level recall efficiently. Our models are open-sourced at https://huggingface.co/collections/m-a-p/hybrid-linear-attention-research-686c488a63d609d2f20e2b1e.

Attention is the critical component of a transformer. Yet the quadratic computational complexity of vanilla full attention in the input size and the inability of its linear attention variant to focus have been challenges for computer vision tasks. We provide a mathematical definition of generalized attention and formulate both vanilla softmax attention and linear attention within the general framework. We prove that generalized attention disperses, that is, as the number of keys tends to infinity, the query assigns equal weights to all keys. Motivated by the dispersion property and recent development of Mamba form of attention, we design Scalable and Efficient Mamba like Attention (SEMA) which utilizes token localization to avoid dispersion and maintain focusing, complemented by theoretically consistent arithmetic averaging to capture global aspect of attention. We support our approach on Imagenet-1k where classification results show that SEMA is a scalable and effective alternative beyond linear attention, outperforming recent vision Mamba models on increasingly larger scales of images at similar model parameter sizes.

Transformer-based models have made remarkable progress in image restoration (IR) tasks. However, the quadratic complexity of self-attention in Transformer hinders its applicability to high-resolution images. Existing methods mitigate this issue with sparse or window-based attention, yet inherently limit global context modeling. Linear attention, a variant of softmax attention, demonstrates promise in global context modeling while maintaining linear complexity, offering a potential solution to the above challenge. Despite its efficiency benefits, vanilla linear attention suffers from a significant performance drop in IR, largely due to the low-rank nature of its attention map. To counter this, we propose Rank Enhanced Linear Attention (RELA), a simple yet effective method that enriches feature representations by integrating a lightweight depthwise convolution. Building upon RELA, we propose an efficient and effective Vision Transformer, named LAformer. LAformer eliminates hardware-inefficient operations such as softmax and window shifting, enabling efficient processing of high-resolution images. Extensive experiments across 7 IR tasks and 21 benchmarks demonstrate that LAformer outperforms SOTA methods and offers significant computational advantages. Furthermore, we extend LAformer to diffusion-based and flow-based visual generation, showcasing its strong potential as a competitive alternative to DiT and SiT. Code and models are available at https://github.com/shallowdream204/LAformer.

The Transformer architecture has achieved remarkable success in natural language processing and high-level vision tasks over the past few years. However, the inherent complexity of self-attention is quadratic to the size of the image, leading to unaffordable computational costs for high-resolution vision tasks. In this paper, we introduce Concertormer, featuring a novel Concerto Self-Attention (CSA) mechanism designed for image deblurring. The proposed CSA divides self-attention into two distinct components: one emphasizes generally global and another concentrates on specifically local correspondence. By retaining partial information in additional dimensions independent from the self-attention calculations, our method effectively captures global contextual representations with complexity linear to the image size. To effectively leverage the additional dimensions, we present a Cross-Dimensional Communication module, which linearly combines attention maps and thus enhances expressiveness. Moreover, we amalgamate the two-staged Transformer design into a single stage using the proposed gated-dconv MLP architecture. While our primary objective is single-image motion deblurring, extensive quantitative and qualitative evaluations demonstrate that our approach performs favorably against the state-of-the-art methods in other tasks, such as deraining and deblurring with JPEG artifacts. The source codes and trained models will be made available to the public.