Distillation has emerged as a practical and effective approach to enhance the reasoning capabilities of open-source language models. In this work, we conduct a large-scale empirical study on reasoning data distillation by collecting verified outputs from three state-of-the-art teacher models-AM-Thinking-v1, Qwen3-235B-A22B, and DeepSeek-R1-on a shared corpus of 1.89 million queries. We construct three parallel datasets and analyze their distributions, revealing that AM-Thinking-v1-distilled data exhibits greater token length diversity and lower perplexity. Student models trained on each dataset are evaluated on reasoning benchmarks including AIME2024, AIME2025, MATH500, and LiveCodeBench. The model distilled from AM-Thinking-v1 consistently achieves the best performance (e.g., 84.3 on AIME2024, 72.2 on AIME2025, 98.4 on MATH500, and 65.9 on LiveCodeBench) and demonstrates adaptive output behavior-producing longer responses for harder tasks and shorter ones for simpler tasks. These findings highlight the value of high-quality, verified reasoning traces. We release the AM-Thinking-v1 and Qwen3-235B-A22B distilled datasets to support future research on open and high-performing reasoning-oriented language models. The datasets are publicly available on Hugging Face\footnote{Datasets are available on Hugging Face: \href{https://huggingface.co/datasets/a-m-team/AM-Thinking-v1-Distilled}{AM-Thinking-v1-Distilled}, \href{https://huggingface.co/datasets/a-m-team/AM-Qwen3-Distilled}{AM-Qwen3-Distilled}.}.

While integrating external tools into large language models (LLMs) enhances their ability to access real-time information and domain-specific services, existing approaches focus narrowly on functional tool selection following user instructions, overlooking the context-aware personalization in tool selection. This oversight leads to suboptimal user satisfaction and inefficient tool utilization, particularly when overlapping toolsets require nuanced selection based on contextual factors. To bridge this gap, we introduce ToolSpectrum, a benchmark designed to evaluate LLMs' capabilities in personalized tool utilization. Specifically, we formalize two key dimensions of personalization, user profile and environmental factors, and analyze their individual and synergistic impacts on tool utilization. Through extensive experiments on ToolSpectrum, we demonstrate that personalized tool utilization significantly improves user experience across diverse scenarios. However, even state-of-the-art LLMs exhibit the limited ability to reason jointly about user profiles and environmental factors, often prioritizing one dimension at the expense of the other. Our findings underscore the necessity of context-aware personalization in tool-augmented LLMs and reveal critical limitations for current models. Our data and code are available at https://github.com/Chengziha0/ToolSpectrum.

Large language models struggle with representing and generating rare tokens despite their importance in specialized domains. In this study, we identify neuron structures with exceptionally strong influence on language model's prediction of rare tokens, termed as rare token neurons, and investigate the mechanism for their emergence and behavior. These neurons exhibit a characteristic three-phase organization (plateau, power-law, and rapid decay) that emerges dynamically during training, evolving from a homogeneous initial state to a functionally differentiated architecture. In the activation space, rare token neurons form a coordinated subnetwork that selectively co-activates while avoiding co-activation with other neurons. This functional specialization potentially correlates with the development of heavy-tailed weight distributions, suggesting a statistical mechanical basis for emergent specialization.

Model merging has emerged as a promising technique for enhancing large language models, though its application in large-scale pre-training remains relatively unexplored. In this paper, we present a comprehensive investigation of model merging techniques during the pre-training process. Through extensive experiments with both dense and Mixture-of-Experts (MoE) architectures ranging from millions to over 100 billion parameters, we demonstrate that merging checkpoints trained with constant learning rates not only achieves significant performance improvements but also enables accurate prediction of annealing behavior. These improvements lead to both more efficient model development and significantly lower training costs. Our detailed ablation studies on merging strategies and hyperparameters provide new insights into the underlying mechanisms while uncovering novel applications. Through comprehensive experimental analysis, we offer the open-source community practical pre-training guidelines for effective model merging.

Agents powered by advanced large language models (LLMs) have demonstrated impressive capabilities across diverse complex applications. Recently, Multi-Agent Systems (MAS), wherein multiple agents collaborate and communicate with each other, have exhibited enhanced capabilities in complex tasks, such as high-quality code generation and arithmetic reasoning. However, the development of such systems often relies on handcrafted methods, and the literature on systematic design and optimization of LLM-based MAS remains limited. In this work, we introduce OMAC, a general framework designed for holistic optimization of LLM-based MAS. Specifically, we identify five key optimization dimensions for MAS, encompassing both agent functionality and collaboration structure. Building upon these dimensions, we first propose a general algorithm, utilizing two actors termed the Semantic Initializer and the Contrastive Comparator, to optimize any single dimension. Then, we present an algorithm for joint optimization across multiple dimensions. Extensive experiments demonstrate the superior performance of OMAC on code generation, arithmetic reasoning, and general reasoning tasks against state-of-the-art approaches.

Understanding and predicting urban dynamics is crucial for managing transportation systems, optimizing urban planning, and enhancing public services. While neural network-based approaches have achieved success, they often rely on task-specific architectures and large volumes of data, limiting their ability to generalize across diverse urban scenarios. Meanwhile, Large Language Models (LLMs) offer strong reasoning and generalization capabilities, yet their application to spatial-temporal urban dynamics remains underexplored. Existing LLM-based methods struggle to effectively integrate multifaceted spatial-temporal data and fail to address distributional shifts between training and testing data, limiting their predictive reliability in real-world applications. To bridge this gap, we propose UrbanMind, a novel spatial-temporal LLM framework for multifaceted urban dynamics prediction that ensures both accurate forecasting and robust generalization. At its core, UrbanMind introduces Muffin-MAE, a multifaceted fusion masked autoencoder with specialized masking strategies that capture intricate spatial-temporal dependencies and intercorrelations among multifaceted urban dynamics. Additionally, we design a semantic-aware prompting and fine-tuning strategy that encodes spatial-temporal contextual details into prompts, enhancing LLMs' ability to reason over spatial-temporal patterns. To further improve generalization, we introduce a test time adaptation mechanism with a test data reconstructor, enabling UrbanMind to dynamically adjust to unseen test data by reconstructing LLM-generated embeddings. Extensive experiments on real-world urban datasets across multiple cities demonstrate that UrbanMind consistently outperforms state-of-the-art baselines, achieving high accuracy and robust generalization, even in zero-shot settings.

Large Language Models (LLMs) increasingly incorporate multilingual capabilities, fueling the demand to transfer them into target language-specific models. However, most approaches, which blend the source model's embedding by replacing the source vocabulary with the target language-specific vocabulary, may constrain expressive capacity in the target language since the source model is predominantly trained on English data. In this paper, we propose Semantic Aware Linear Transfer (SALT), a novel cross-lingual transfer technique that recycles embeddings from target language Pre-trained Language Models (PLMs) to transmit the deep representational strengths of PLM-derived embedding to LLMs. SALT derives unique regression lines based on the similarity in the overlap of the source and target vocabularies, to handle each non-overlapping token's embedding space. Our extensive experiments show that SALT significantly outperforms other transfer methods and achieves lower loss with accelerating faster convergence during language adaptation. Notably, SALT obtains remarkable performance in cross-lingual understanding setups compared to other methods. Furthermore, we highlight the scalable use of PLMs to enhance the functionality of contemporary LLMs by conducting experiments with varying architectures.

Recent advancements in source code summarization have leveraged transformer-based pre-trained models, including Large Language Models of Code (LLMCs), to automate and improve the generation of code summaries. However, existing methods often focus on generating a single high-quality summary for a given source code, neglecting scenarios where the generated summary might be inadequate and alternative options are needed. In this paper, we introduce Variational Prefix Tuning (VPT), a novel approach that enhances pre-trained models' ability to generate diverse yet accurate sets of summaries, allowing the user to choose the most suitable one for the given source code. Our method integrates a Conditional Variational Autoencoder (CVAE) framework as a modular component into pre-trained models, enabling us to model the distribution of observed target summaries and sample continuous embeddings to be used as prefixes to steer the generation of diverse outputs during decoding. Importantly, we construct our method in a parameter-efficient manner, eliminating the need for expensive model retraining, especially when using LLMCs. Furthermore, we employ a bi-criteria reranking method to select a subset of generated summaries, optimizing both the diversity and the accuracy of the options presented to users. We present extensive experimental evaluations using widely used datasets and current state-of-the-art pre-trained code summarization models to demonstrate the effectiveness of our approach and its adaptability across models.

This study systematically evaluates 27 frontier Large Language Models on eight biology benchmarks spanning molecular biology, genetics, cloning, virology, and biosecurity. Models from major AI developers released between November 2022 and April 2025 were assessed through ten independent runs per benchmark. The findings reveal dramatic improvements in biological capabilities. Top model performance increased more than 4-fold on the challenging text-only subset of the Virology Capabilities Test over the study period, with OpenAI's o3 now performing twice as well as expert virologists. Several models now match or exceed expert-level performance on other challenging benchmarks, including the biology subsets of GPQA and WMDP and LAB-Bench CloningScenarios. Contrary to expectations, chain-of-thought did not substantially improve performance over zero-shot evaluation, while extended reasoning features in o3-mini and Claude 3.7 Sonnet typically improved performance as predicted by inference scaling. Benchmarks such as PubMedQA and the MMLU and WMDP biology subsets exhibited performance plateaus well below 100%, suggesting benchmark saturation and errors in the underlying benchmark data. The analysis highlights the need for more sophisticated evaluation methodologies as AI systems continue to advance.

Machine learning has demonstrated transformative potential for database operations, such as query optimization and in-database data analytics. However, dynamic database environments, characterized by frequent updates and evolving data distributions, introduce concept drift, which leads to performance degradation for learned models and limits their practical applicability. Addressing this challenge requires efficient frameworks capable of adapting to shifting concepts while minimizing the overhead of retraining or fine-tuning. In this paper, we propose FLAIR, an online adaptation framework that introduces a new paradigm called \textit{in-context adaptation} for learned database operations. FLAIR leverages the inherent property of data systems, i.e., immediate availability of execution results for predictions, to enable dynamic context construction. By formalizing adaptation as $f:(\mathbf{x} \,| \,C_t) \to \mathbf{y}$, with $C_t$ representing a dynamic context memory, FLAIR delivers predictions aligned with the current concept, eliminating the need for runtime parameter optimization. To achieve this, FLAIR integrates two key modules: a Task Featurization Module for encoding task-specific features into standardized representations, and a Dynamic Decision Engine, pre-trained via Bayesian meta-training, to adapt seamlessly using contextual information at runtime. Extensive experiments across key database tasks demonstrate that FLAIR outperforms state-of-the-art baselines, achieving up to 5.2x faster adaptation and reducing error by 22.5% for cardinality estimation.