Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem

Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agent LLMs. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME (ROME is Obviously an Agentic Model), an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-based Policy Alignment (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of the ALE infrastructure.

Tracing the Flow of Knowledge From Science to Technology Using Deep Learning

We develop a language similarity model suitable for working with patents and scientific publications at the same time. In a horse race-style evaluation, we subject eight language (similarity) models to predict credible Patent-Paper Citations. We find that our Pat-SPECTER model performs best, which is the SPECTER2 model fine-tuned on patents. In two real-world scenarios (separating patent-paper-pairs and predicting patent-paper-pairs) we demonstrate the capabilities of the Pat-SPECTER. We finally test the hypothesis that US patents cite papers that are semantically less similar than in other large jurisdictions, which we posit is because of the duty of candor. The model is open for the academic community and practitioners alike.

CLO: Efficient LLM Inference System with CPU-Light KVCache Offloading via Algorithm-System Co-Design

The growth of million-token LLMs exposes the scalability limits of inference systems, where the KVCache dominates memory usage and data transfer overhead. Recent offloading systems migrate the KVCache to CPU memory and incorporate top-k attention to reduce the volume of data transferred from the CPU, while further applying system-level optimizations such as on-GPU caching and prefetching to lower transfer overhead. However, they overlook the CPU bottleneck in three aspects: (1) substantial overhead of fine-grained dynamic cache management performed on the CPU side, (2) significant transfer overhead from poor PCIe bandwidth utilization caused by heavy gathering operations at the CPU side, and (3) GPU runtime bubbles introduced by coarse-grained CPU-centric synchronization. To address these challenges, we propose CLO, a CPU-light KVCache offloading system via algorithm-system co-design. CLO features: (1) a coarse-grained head-wise approximate on-GPU caching strategy with negligible cache management cost, (2) seamless combination of data prefetching and on-GPU persistent caching for lower transfer overhead, (3) a zero-copy transfer engine to fully exploit PCIe bandwidth, and a GPU-centric synchronization method to eliminate GPU stalls. Evaluation on two widely-used LLMs demonstrates that CLO achieves comparable accuracy to state-of-the-art systems, while substantially minimizing CPU overhead, fully utilizing PCIe bandwidth, thus improving decoding throughput by 9.3%-66.6%. Our results highlight that algorithm-system co-design is essential for memory-constrained LLM inference on modern GPU platforms. We open source CLO at https://github.com/CommediaJW/CLO.

Virtual Width Networks

We introduce Virtual Width Networks (VWN), a framework that delivers the benefits of wider representations without incurring the quadratic cost of increasing the hidden size. VWN decouples representational width from backbone width, expanding the embedding space while keeping backbone compute nearly constant. In our large-scale experiment, an 8-times expansion accelerates optimization by over 2 times for next-token and 3 times for next-2-token prediction. The advantage amplifies over training as both the loss gap grows and the convergence-speedup ratio increases, showing that VWN is not only token-efficient but also increasingly effective with scale. Moreover, we identify an approximately log-linear scaling relation between virtual width and loss reduction, offering an initial empirical basis and motivation for exploring virtual-width scaling as a new dimension of large-model efficiency.

User Hesitation and Negative Transfer in Multi-Behavior Recommendation

Multi-behavior recommendation aims to integrate users' interactions across various behavior types (e.g., view, favorite, add-to-cart, purchase) to more comprehensively characterize user preferences. However, existing methods lack in-depth modeling when dealing with interactions that generate only auxiliary behaviors without triggering the target behavior. In fact, these weak signals contain rich latent information and can be categorized into two types: (1) positive weak signals-items that have not triggered the target behavior but exhibit frequent auxiliary interactions, reflecting users' hesitation tendencies toward these items; and (2) negative weak signals-auxiliary behaviors that result from misoperations or interaction noise, which deviate from true preferences and may cause negative transfer effects. To more effectively identify and utilize these weak signals, we propose a recommendation framework focused on weak signal learning, termed HNT. Specifically, HNT models weak signal features from two dimensions: positive and negative effects. By learning the characteristics of auxiliary behaviors that lead to target behaviors, HNT identifies similar auxiliary behaviors that did not trigger the target behavior and constructs a hesitation set of related items as weak positive samples to enhance preference modeling, thereby capturing users' latent hesitation intentions. Meanwhile, during auxiliary feature fusion, HNT incorporates latent negative transfer effect modeling to distinguish and suppress interference caused by negative representations through item similarity learning. Experiments on three real-world datasets demonstrate that HNT improves HR@10 and NDCG@10 by 12.57% and 14.37%, respectively, compared to the best baseline methods.

Generative Inversion for Property-Targeted Materials Design: Application to Shape Memory Alloys

The design of shape memory alloys (SMAs) with high transformation temperatures and large mechanical work output remains a longstanding challenge in functional materials engineering. Here, we introduce a data-driven framework based on generative adversarial network (GAN) inversion for the inverse design of high-performance SMAs. By coupling a pretrained GAN with a property prediction model, we perform gradient-based latent space optimization to directly generate candidate alloy compositions and processing parameters that satisfy user-defined property targets. The framework is experimentally validated through the synthesis and characterization of five NiTi-based SMAs. Among them, the Ni$_{49.8}$Ti$_{26.4}$Hf$_{18.6}$Zr$_{5.2}$ alloy achieves a high transformation temperature of 404 $^\circ$C, a large mechanical work output of 9.9 J/cm$^3$, a transformation enthalpy of 43 J/g , and a thermal hysteresis of 29 {\deg}C, outperforming existing NiTi alloys. The enhanced performance is attributed to a pronounced transformation volume change and a finely dispersed of Ti$_2$Ni-type precipitates, enabled by sluggish Zr and Hf diffusion, and semi-coherent interfaces with localized strain fields. This study demonstrates that GAN inversion offers an efficient and generalizable route for the property-targeted discovery of complex alloys.

Kimi K2: Open Agentic Intelligence

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.

When Brain Foundation Model Meets Cauchy-Schwarz Divergence: A New Framework for Cross-Subject Motor Imagery Decoding

Decoding motor imagery (MI) electroencephalogram (EEG) signals, a key non-invasive brain-computer interface (BCI) paradigm for controlling external systems, has been significantly advanced by deep learning. However, MI-EEG decoding remains challenging due to substantial inter-subject variability and limited labeled target data, which necessitate costly calibration for new users. Many existing multi-source domain adaptation (MSDA) methods indiscriminately incorporate all available source domains, disregarding the large inter-subject differences in EEG signals, which leads to negative transfer and excessive computational costs. Moreover, while many approaches focus on feature distribution alignment, they often neglect the explicit dependence between features and decision-level outputs, limiting their ability to preserve discriminative structures. To address these gaps, we propose a novel MSDA framework that leverages a pretrained large Brain Foundation Model (BFM) for dynamic and informed source subject selection, ensuring only relevant sources contribute to adaptation. Furthermore, we employ Cauchy-Schwarz (CS) and Conditional CS (CCS) divergences to jointly perform feature-level and decision-level alignment, enhancing domain invariance while maintaining class discriminability. Extensive evaluations on two benchmark MI-EEG datasets demonstrate that our framework outperforms a broad range of state-of-the-art baselines. Additional experiments with a large source pool validate the scalability and efficiency of BFM-guided selection, which significantly reduces training time without sacrificing performance.

KunLunBaizeRAG: Reinforcement Learning Driven Inference Performance Leap for Large Language Models

This paper introduces KunLunBaizeRAG, a reinforcement learning-driven reasoning framework designed to enhance the reasoning capabilities of large language models (LLMs) in complex multi-hop question-answering tasks. The framework addresses key limitations of traditional RAG, such as retrieval drift, information redundancy, and strategy rigidity. Key innovations include the RAG-driven Reasoning Alignment (RDRA) mechanism, the Search-Think Iterative Enhancement (STIE) mechanism, the Network-Local Intelligent Routing (NLR) mechanism, and a progressive hybrid training strategy. Experimental results demonstrate significant improvements in exact match (EM) and LLM-judged score (LJ) across four benchmarks, highlighting the framework's robustness and effectiveness in complex reasoning scenarios.

ReVeal: Self-Evolving Code Agents via Iterative Generation-Verification

Recent advances in reinforcement learning (RL) with verifiable outcome rewards have significantly improved the reasoning capabilities of large language models (LLMs), especially when combined with multi-turn tool interactions. However, existing methods lack both meaningful verification signals from realistic environments and explicit optimization for verification, leading to unreliable self-verification. To address these limitations, we propose ReVeal, a multi-turn reinforcement learning framework that interleaves code generation with explicit self-verification and tool-based evaluation. ReVeal enables LLMs to autonomously generate test cases, invoke external tools for precise feedback, and improves performance via a customized RL algorithm with dense, per-turn rewards. As a result, ReVeal fosters the co-evolution of a model's generation and verification capabilities through RL training, expanding the reasoning boundaries of the base model, demonstrated by significant gains in Pass@k on LiveCodeBench. It also enables test-time scaling into deeper inference regimes, with code consistently evolving as the number of turns increases during inference, ultimately surpassing DeepSeek-R1-Zero-Qwen-32B. These findings highlight the promise of ReVeal as a scalable and effective paradigm for building more robust and autonomous AI agents.