AttnCache: Accelerating Self-Attention Inference for LLM Prefill via Attention Cache

Large Language Models (LLMs) are widely used in generative applications such as chatting, code generation, and reasoning. However, many realworld workloads such as classification, question answering, recommendation, and text embedding rely solely on the prefill stage of inference, where the model encodes input sequences without performing autoregressive decoding. In these prefill only scenarios, the self-attention computation becomes the primary performance bottleneck due to its quadratic complexity with respect to sequence length. In this paper, we observe that semantically different sentences often produce similar attention maps across layers and heads. Building on this insight, we propose AttnCache, a framework that accelerates the prefill stage of LLM inference by retrieving and reusing similar attention maps. Based on an attention map memorization database, AttnCache employs efficient caching and similarity search techniques to identify and reuse pre-cached attention maps during inference, thereby reducing the computational overhead of self-attention. Experimental results show that AttnCache achieves an average of 1.2x end-to-end and 2x attention speedup on CPU, and 1.6x end-to-end and 3x attention speedup on GPU, with negligible accuracy degradation.

NeuronMM: High-Performance Matrix Multiplication for LLM Inference on AWS Trainium

AI accelerators, customized to AI workloads, provide cost-effective and high-performance solutions for training and inference. Trainium, an AI accelerator recently developed by Amazon Web Services (AWS), provides an attractive option for LLM training and inference through its heterogeneous architecture. However, leveraging Trainium architecture for high performance can be challenging because of its systolic array architecture and special requirement on data layout. In this paper, we design high-performance matrix multiplication (matmul), a critical compute kernel, for LLM inference on Trainium. We introduce a series of techniques customized to Trainium based on kernel fusion and novel caching strategies to reduce data movement across the software-managed memory hierarchy, maximize SRAM bandwidth, and avoid expensive matrix transpose. Evaluating with nine datasets and four recent LLMs, we show that our system largely outperforms the state-of-the-art matmul implemented by AWS on Trainium: at the level of matmul kernel, it achieves an average 1.35x speedup (up to 2.22x), which translates to an average 1.66x speedup (up to 2.49x) for end-to-end LLM inference.

TWC-SLAM: Multi-Agent Cooperative SLAM with Text Semantics and WiFi Features Integration for Similar Indoor Environments

Multi-agent cooperative SLAM often encounters challenges in similar indoor environments characterized by repetitive structures, such as corridors and rooms. These challenges can lead to significant inaccuracies in shared location identification when employing point cloud-based techniques. To mitigate these issues, we introduce TWC-SLAM, a multi-agent cooperative SLAM framework that integrates text semantics and WiFi signal features to enhance location identification and loop closure detection. TWC-SLAM comprises a single-agent front-end odometry module based on FAST-LIO2, a location identification and loop closure detection module that leverages text semantics and WiFi features, and a global mapping module. The agents are equipped with sensors capable of capturing textual information and detecting WiFi signals. By correlating these data sources, TWC-SLAM establishes a common location, facilitating point cloud alignment across different agents' maps. Furthermore, the system employs loop closure detection and optimization modules to achieve global optimization and cohesive mapping. We evaluated our approach using an indoor dataset featuring similar corridors, rooms, and text signs. The results demonstrate that TWC-SLAM significantly improves the performance of cooperative SLAM systems in complex environments with repetitive architectural features.

Latent Harmony: Synergistic Unified UHD Image Restoration via Latent Space Regularization and Controllable Refinement

Ultra-High Definition (UHD) image restoration faces a trade-off between computational efficiency and high-frequency detail retention. While Variational Autoencoders (VAEs) improve efficiency via latent-space processing, their Gaussian constraint often discards degradation-specific high-frequency information, hurting reconstruction fidelity. To overcome this, we propose Latent Harmony, a two-stage framework that redefines VAEs for UHD restoration by jointly regularizing the latent space and enforcing high-frequency-aware reconstruction.In Stage One, we introduce LH-VAE, which enhances semantic robustness through visual semantic constraints and progressive degradation perturbations, while latent equivariance strengthens high-frequency reconstruction.Stage Two jointly trains this refined VAE with a restoration model using High-Frequency Low-Rank Adaptation (HF-LoRA): an encoder LoRA guided by a fidelity-oriented high-frequency alignment loss to recover authentic details, and a decoder LoRA driven by a perception-oriented loss to synthesize realistic textures. Both LoRA modules are trained via alternating optimization with selective gradient propagation to preserve the pretrained latent structure.At inference, a tunable parameter {\alpha} enables flexible fidelity-perception trade-offs.Experiments show Latent Harmony achieves state-of-the-art performance across UHD and standard-resolution tasks, effectively balancing efficiency, perceptual quality, and reconstruction accuracy.

A Survey of Reinforcement Learning for Large Reasoning Models

In this paper, we survey recent advances in Reinforcement Learning (RL) for reasoning with Large Language Models (LLMs). RL has achieved remarkable success in advancing the frontier of LLM capabilities, particularly in addressing complex logical tasks such as mathematics and coding. As a result, RL has emerged as a foundational methodology for transforming LLMs into LRMs. With the rapid progress of the field, further scaling of RL for LRMs now faces foundational challenges not only in computational resources but also in algorithm design, training data, and infrastructure. To this end, it is timely to revisit the development of this domain, reassess its trajectory, and explore strategies to enhance the scalability of RL toward Artificial SuperIntelligence (ASI). In particular, we examine research applying RL to LLMs and LRMs for reasoning abilities, especially since the release of DeepSeek-R1, including foundational components, core problems, training resources, and downstream applications, to identify future opportunities and directions for this rapidly evolving area. We hope this review will promote future research on RL for broader reasoning models. Github: https://github.com/TsinghuaC3I/Awesome-RL-for-LRMs

SparK: Query-Aware Unstructured Sparsity with Recoverable KV Cache Channel Pruning

Long-context inference in large language models (LLMs) is increasingly constrained by the KV cache bottleneck: memory usage grows linearly with sequence length, while attention computation scales quadratically. Existing approaches address this issue by compressing the KV cache along the temporal axis through strategies such as token eviction or merging to reduce memory and computational overhead. However, these methods often neglect fine-grained importance variations across feature dimensions (i.e., the channel axis), thereby limiting their ability to effectively balance efficiency and model accuracy. In reality, we observe that channel saliency varies dramatically across both queries and positions: certain feature channels carry near-zero information for a given query, while others spike in relevance. To address this oversight, we propose SPARK, a training-free plug-and-play method that applies unstructured sparsity by pruning KV at the channel level, while dynamically restoring the pruned entries during attention score computation. Notably, our approach is orthogonal to existing KV compression and quantization techniques, making it compatible for integration with them to achieve further acceleration. By reducing channel-level redundancy, SPARK enables processing of longer sequences within the same memory budget. For sequences of equal length, SPARK not only preserves or improves model accuracy but also reduces KV cache storage by over 30% compared to eviction-based methods. Furthermore, even with an aggressive pruning ratio of 80%, SPARK maintains performance with less degradation than 5% compared to the baseline eviction method, demonstrating its robustness and effectiveness. Our code will be available at https://github.com/Xnhyacinth/SparK.

NTIRE 2025 Image Shadow Removal Challenge Report

This work examines the findings of the NTIRE 2025 Shadow Removal Challenge. A total of 306 participants have registered, with 17 teams successfully submitting their solutions during the final evaluation phase. Following the last two editions, this challenge had two evaluation tracks: one focusing on reconstruction fidelity and the other on visual perception through a user study. Both tracks were evaluated with images from the WSRD+ dataset, simulating interactions between self- and cast-shadows with a large number of diverse objects, textures, and materials.

Abstract, Align, Predict: Zero-Shot Stance Detection via Cognitive Inductive Reasoning

Zero-shot stance detection (ZSSD) aims to identify the stance of text toward previously unseen targets, a setting where conventional supervised models often fail due to reliance on labeled data and shallow lexical cues. Inspired by human cognitive reasoning, we propose the Cognitive Inductive Reasoning Framework (CIRF), which abstracts transferable reasoning schemas from unlabeled text and encodes them as concept-level logic. To integrate these schemas with input arguments, we introduce a Schema-Enhanced Graph Kernel Model (SEGKM) that dynamically aligns local and global reasoning structures. Experiments on SemEval-2016, VAST, and COVID-19-Stance benchmarks show that CIRF establishes new state-of-the-art results, outperforming strong ZSSD baselines by 1.0, 4.5, and 3.3 percentage points in macro-F1, respectively, and achieving comparable accuracy with 70\% fewer labeled examples. We will release the full code upon publication.

MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention

We introduce MiniMax-M1, the world's first open-weight, large-scale hybrid-attention reasoning model. MiniMax-M1 is powered by a hybrid Mixture-of-Experts (MoE) architecture combined with a lightning attention mechanism. The model is developed based on our previous MiniMax-Text-01 model, which contains a total of 456 billion parameters with 45.9 billion parameters activated per token. The M1 model natively supports a context length of 1 million tokens, 8x the context size of DeepSeek R1. Furthermore, the lightning attention mechanism in MiniMax-M1 enables efficient scaling of test-time compute. These properties make M1 particularly suitable for complex tasks that require processing long inputs and thinking extensively. MiniMax-M1 is trained using large-scale reinforcement learning (RL) on diverse problems including sandbox-based, real-world software engineering environments. In addition to M1's inherent efficiency advantage for RL training, we propose CISPO, a novel RL algorithm to further enhance RL efficiency. CISPO clips importance sampling weights rather than token updates, outperforming other competitive RL variants. Combining hybrid-attention and CISPO enables MiniMax-M1's full RL training on 512 H800 GPUs to complete in only three weeks, with a rental cost of just $534,700. We release two versions of MiniMax-M1 models with 40K and 80K thinking budgets respectively, where the 40K model represents an intermediate phase of the 80K training. Experiments on standard benchmarks show that our models are comparable or superior to strong open-weight models such as the original DeepSeek-R1 and Qwen3-235B, with particular strengths in complex software engineering, tool utilization, and long-context tasks. We publicly release MiniMax-M1 at https://github.com/MiniMax-AI/MiniMax-M1.

Athena: Enhancing Multimodal Reasoning with Data-efficient Process Reward Models

We present Athena-PRM, a multimodal process reward model (PRM) designed to evaluate the reward score for each step in solving complex reasoning problems. Developing high-performance PRMs typically demands significant time and financial investment, primarily due to the necessity for step-level annotations of reasoning steps. Conventional automated labeling methods, such as Monte Carlo estimation, often produce noisy labels and incur substantial computational costs. To efficiently generate high-quality process-labeled data, we propose leveraging prediction consistency between weak and strong completers as a criterion for identifying reliable process labels. Remarkably, Athena-PRM demonstrates outstanding effectiveness across various scenarios and benchmarks with just 5,000 samples. Furthermore, we also develop two effective strategies to improve the performance of PRMs: ORM initialization and up-sampling for negative data. We validate our approach in three specific scenarios: verification for test time scaling, direct evaluation of reasoning step correctness, and reward ranked fine-tuning. Our Athena-PRM consistently achieves superior performance across multiple benchmarks and scenarios. Notably, when using Qwen2.5-VL-7B as the policy model, Athena-PRM enhances performance by 10.2 points on WeMath and 7.1 points on MathVista for test time scaling. Furthermore, Athena-PRM sets the state-of-the-art (SoTA) results in VisualProcessBench and outperforms the previous SoTA by 3.9 F1-score, showcasing its robust capability to accurately assess the correctness of the reasoning step. Additionally, utilizing Athena-PRM as the reward model, we develop Athena-7B with reward ranked fine-tuning and outperforms baseline with a significant margin on five benchmarks.