IndexCache: Accelerating Sparse Attention via Cross-Layer Index Reuse

Long-context agentic workflows have emerged as a defining use case for large language models, making attention efficiency critical for both inference speed and serving cost. Sparse attention addresses this challenge effectively, and DeepSeek Sparse Attention (DSA) is a representative production-grade solution: a lightweight lightning indexer selects the top-k most relevant tokens per query, reducing core attention from $O(L^2)$ to $O(Lk)$. However, the indexer itself retains $O(L^2)$ complexity and must run independently at every layer, despite the fact that the resulting top-k selections are highly similar across consecutive layers. We present IndexCache, which exploits this cross-layer redundancy by partitioning layers into a small set of Full layers that run their own indexers and a majority of Shared layers that simply reuse the nearest Full layer's top-k indices. We propose two complementary approaches to determine and optimize this configuration. Training-free IndexCache applies a greedy search algorithm that selects which layers to retain indexers by directly minimizing language modeling loss on a calibration set, requiring no weight updates. Training-aware IndexCache introduces a multi-layer distillation loss that trains each retained indexer against the averaged attention distributions of all layers it serves, enabling even simple interleaved patterns to match full-indexer accuracy. Experimental results on a 30B DSA model show that IndexCache can remove 75% of indexer computations with negligible quality degradation, achieving up to 1.82$\times$ prefill speedup and 1.48$\times$ decode speedup compared to standard DSA. These positive results are further confirmed by our preliminary experiments on the production-scale GLM-5 model (Figure 1).

T2S-Bench & Structure-of-Thought: Benchmarking and Prompting Comprehensive Text-to-Structure Reasoning

Think about how human handles complex reading tasks: marking key points, inferring their relationships, and structuring information to guide understanding and responses. Likewise, can a large language model benefit from text structure to enhance text-processing performance? To explore it, in this work, we first introduce Structure of Thought (SoT), a prompting technique that explicitly guides models to construct intermediate text structures, consistently boosting performance across eight tasks and three model families. Building upon this insight, we present T2S-Bench, the first benchmark designed to evaluate and improve text-to-structure capabilities of models. T2S-Bench includes 1.8K samples across 6 scientific domains and 32 structural types, rigorously constructed to ensure accuracy, fairness, and quality. Evaluation on 45 mainstream models reveals substantial improvement potential: the average accuracy on the multi-hop reasoning task is only 52.1%, and even the most advanced model achieves 58.1% node accuracy in end-to-end extraction. Furthermore, on Qwen2.5-7B-Instruct, SoT alone yields an average +5.7% improvement across eight diverse text-processing tasks, and fine-tuning on T2S-Bench further increases this gain to +8.6%. These results highlight the value of explicit text structuring and the complementary contributions of SoT and T2S-Bench. Dataset and eval code have been released at https://t2s-bench.github.io/T2S-Bench-Page/.

GLM-5: from Vibe Coding to Agentic Engineering

We present GLM-5, a next-generation foundation model designed to transition the paradigm of vibe coding to agentic engineering. Building upon the agentic, reasoning, and coding (ARC) capabilities of its predecessor, GLM-5 adopts DSA to significantly reduce training and inference costs while maintaining long-context fidelity. To advance model alignment and autonomy, we implement a new asynchronous reinforcement learning infrastructure that drastically improves post-training efficiency by decoupling generation from training. Furthermore, we propose novel asynchronous agent RL algorithms that further improve RL quality, enabling the model to learn from complex, long-horizon interactions more effectively. Through these innovations, GLM-5 achieves state-of-the-art performance on major open benchmarks. Most critically, GLM-5 demonstrates unprecedented capability in real-world coding tasks, surpassing previous baselines in handling end-to-end software engineering challenges. Code, models, and more information are available at https://github.com/zai-org/GLM-5.

SADA: Stability-guided Adaptive Diffusion Acceleration

Diffusion models have achieved remarkable success in generative tasks but suffer from high computational costs due to their iterative sampling process and quadratic attention costs. Existing training-free acceleration strategies that reduce per-step computation cost, while effectively reducing sampling time, demonstrate low faithfulness compared to the original baseline. We hypothesize that this fidelity gap arises because (a) different prompts correspond to varying denoising trajectory, and (b) such methods do not consider the underlying ODE formulation and its numerical solution. In this paper, we propose Stability-guided Adaptive Diffusion Acceleration (SADA), a novel paradigm that unifies step-wise and token-wise sparsity decisions via a single stability criterion to accelerate sampling of ODE-based generative models (Diffusion and Flow-matching). For (a), SADA adaptively allocates sparsity based on the sampling trajectory. For (b), SADA introduces principled approximation schemes that leverage the precise gradient information from the numerical ODE solver. Comprehensive evaluations on SD-2, SDXL, and Flux using both EDM and DPM++ solvers reveal consistent $\ge 1.8\times$ speedups with minimal fidelity degradation (LPIPS $\leq 0.10$ and FID $\leq 4.5$) compared to unmodified baselines, significantly outperforming prior methods. Moreover, SADA adapts seamlessly to other pipelines and modalities: It accelerates ControlNet without any modifications and speeds up MusicLDM by $1.8\times$ with $\sim 0.01$ spectrogram LPIPS.

A Multi-Scale Spatial-Temporal Network for Wireless Video Transmission

Deep joint source-channel coding (DeepJSCC) has shown promise in wireless transmission of text, speech, and images within the realm of semantic communication. However, wireless video transmission presents greater challenges due to the difficulty of extracting and compactly representing both spatial and temporal features, as well as its significant bandwidth and computational resource requirements. In response, we propose a novel video DeepJSCC (VDJSCC) approach to enable end-to-end video transmission over a wireless channel. Our approach involves the design of a multi-scale vision Transformer encoder and decoder to effectively capture spatial-temporal representations over long-term frames. Additionally, we propose a dynamic token selection module to mask less semantically important tokens from spatial or temporal dimensions, allowing for content-adaptive variable-length video coding by adjusting the token keep ratio. Experimental results demonstrate the effectiveness of our VDJSCC approach compared to digital schemes that use separate source and channel codes, as well as other DeepJSCC schemes, in terms of reconstruction quality and bandwidth reduction.

Efficient Single Image Super-Resolution with Entropy Attention and Receptive Field Augmentation

Transformer-based deep models for single image super-resolution (SISR) have greatly improved the performance of lightweight SISR tasks in recent years. However, they often suffer from heavy computational burden and slow inference due to the complex calculation of multi-head self-attention (MSA), seriously hindering their practical application and deployment. In this work, we present an efficient SR model to mitigate the dilemma between model efficiency and SR performance, which is dubbed Entropy Attention and Receptive Field Augmentation network (EARFA), and composed of a novel entropy attention (EA) and a shifting large kernel attention (SLKA). From the perspective of information theory, EA increases the entropy of intermediate features conditioned on a Gaussian distribution, providing more informative input for subsequent reasoning. On the other hand, SLKA extends the receptive field of SR models with the assistance of channel shifting, which also favors to boost the diversity of hierarchical features. Since the implementation of EA and SLKA does not involve complex computations (such as extensive matrix multiplications), the proposed method can achieve faster nonlinear inference than Transformer-based SR models while maintaining better SR performance. Extensive experiments show that the proposed model can significantly reduce the delay of model inference while achieving the SR performance comparable with other advanced models.

Diff-Shadow: Global-guided Diffusion Model for Shadow Removal

We propose Diff-Shadow, a global-guided diffusion model for high-quality shadow removal. Previous transformer-based approaches can utilize global information to relate shadow and non-shadow regions but are limited in their synthesis ability and recover images with obvious boundaries. In contrast, diffusion-based methods can generate better content but ignore global information, resulting in inconsistent illumination. In this work, we combine the advantages of diffusion models and global guidance to realize shadow-free restoration. Specifically, we propose a parallel UNets architecture: 1) the local branch performs the patch-based noise estimation in the diffusion process, and 2) the global branch recovers the low-resolution shadow-free images. A Reweight Cross Attention (RCA) module is designed to integrate global contextural information of non-shadow regions into the local branch. We further design a Global-guided Sampling Strategy (GSS) that mitigates patch boundary issues and ensures consistent illumination across shaded and unshaded regions in the recovered image. Comprehensive experiments on three publicly standard datasets ISTD, ISTD+, and SRD have demonstrated the effectiveness of Diff-Shadow. Compared to state-of-the-art methods, our method achieves a significant improvement in terms of PSNR, increasing from 32.33dB to 33.69dB on the SRD dataset. Codes will be released.

E5-V: Universal Embeddings with Multimodal Large Language Models

Multimodal large language models (MLLMs) have shown promising advancements in general visual and language understanding. However, the representation of multimodal information using MLLMs remains largely unexplored. In this work, we introduce a new framework, E5-V, designed to adapt MLLMs for achieving universal multimodal embeddings. Our findings highlight the significant potential of MLLMs in representing multimodal inputs compared to previous approaches. By leveraging MLLMs with prompts, E5-V effectively bridges the modality gap between different types of inputs, demonstrating strong performance in multimodal embeddings even without fine-tuning. We propose a single modality training approach for E5-V, where the model is trained exclusively on text pairs. This method demonstrates significant improvements over traditional multimodal training on image-text pairs, while reducing training costs by approximately 95%. Additionally, this approach eliminates the need for costly multimodal training data collection. Extensive experiments across four types of tasks demonstrate the effectiveness of E5-V. As a universal multimodal model, E5-V not only achieves but often surpasses state-of-the-art performance in each task, despite being trained on a single modality.

MoRA: High-Rank Updating for Parameter-Efficient Fine-Tuning

Low-rank adaptation is a popular parameter-efficient fine-tuning method for large language models. In this paper, we analyze the impact of low-rank updating, as implemented in LoRA. Our findings suggest that the low-rank updating mechanism may limit the ability of LLMs to effectively learn and memorize new knowledge. Inspired by this observation, we propose a new method called MoRA, which employs a square matrix to achieve high-rank updating while maintaining the same number of trainable parameters. To achieve it, we introduce the corresponding non-parameter operators to reduce the input dimension and increase the output dimension for the square matrix. Furthermore, these operators ensure that the weight can be merged back into LLMs, which makes our method can be deployed like LoRA. We perform a comprehensive evaluation of our method across five tasks: instruction tuning, mathematical reasoning, continual pretraining, memory and pretraining. Our method outperforms LoRA on memory-intensive tasks and achieves comparable performance on other tasks.

Improving Bracket Image Restoration and Enhancement with Flow-guided Alignment and Enhanced Feature Aggregation

In this paper, we address the Bracket Image Restoration and Enhancement (BracketIRE) task using a novel framework, which requires restoring a high-quality high dynamic range (HDR) image from a sequence of noisy, blurred, and low dynamic range (LDR) multi-exposure RAW inputs. To overcome this challenge, we present the IREANet, which improves the multiple exposure alignment and aggregation with a Flow-guide Feature Alignment Module (FFAM) and an Enhanced Feature Aggregation Module (EFAM). Specifically, the proposed FFAM incorporates the inter-frame optical flow as guidance to facilitate the deformable alignment and spatial attention modules for better feature alignment. The EFAM further employs the proposed Enhanced Residual Block (ERB) as a foundational component, wherein a unidirectional recurrent network aggregates the aligned temporal features to better reconstruct the results. To improve model generalization and performance, we additionally employ the Bayer preserving augmentation (BayerAug) strategy to augment the multi-exposure RAW inputs. Our experimental evaluations demonstrate that the proposed IREANet shows state-of-the-art performance compared with previous methods.