Training Report of TeleChat3-MoE

TeleChat3-MoE is the latest series of TeleChat large language models, featuring a Mixture-of-Experts (MoE) architecture with parameter counts ranging from 105 billion to over one trillion,trained end-to-end on Ascend NPU cluster. This technical report mainly presents the underlying training infrastructure that enables reliable and efficient scaling to frontier model sizes. We detail systematic methodologies for operator-level and end-to-end numerical accuracy verification, ensuring consistency across hardware platforms and distributed parallelism strategies. Furthermore, we introduce a suite of performance optimizations, including interleaved pipeline scheduling, attention-aware data scheduling for long-sequence training,hierarchical and overlapped communication for expert parallelism, and DVM-based operator fusion. A systematic parallelization framework, leveraging analytical estimation and integer linear programming, is also proposed to optimize multi-dimensional parallelism configurations. Additionally, we present methodological approaches to cluster-level optimizations, addressing host- and device-bound bottlenecks during large-scale training tasks. These infrastructure advancements yield significant throughput improvements and near-linear scaling on clusters comprising thousands of devices, providing a robust foundation for large-scale language model development on hardware ecosystems.

DBAW-PIKAN: Dynamic Balance Adaptive Weight Kolmogorov-Arnold Neural Network for Solving Partial Differential Equations

Physics-informed neural networks (PINNs) have led to significant advancements in scientific computing by integrating fundamental physical principles with advanced data-driven techniques. However, when dealing with problems characterized by multi-scale or high-frequency features, PINNs encounter persistent and severe challenges related to stiffness in gradient flow and spectral bias, which significantly limit their predictive capabilities. To address these issues, this paper proposes a Dynamic Balancing Adaptive Weighting Physics-Informed Kolmogorov-Arnold Network (DBAW-PIKAN), designed to mitigate such gradient-related failure modes and overcome the bottlenecks in function representation. The core of DBAW-PIKAN combines the Kolmogorov-Arnold network architecture, based on learnable B-splines, with an adaptive weighting strategy that incorporates a dynamic decay upper bound. Compared to baseline models, the proposed method accelerates the convergence process and improves solution accuracy by at least an order of magnitude without introducing additional computational complexity. A series of numerical benchmarks, including the Klein-Gordon, Burgers, and Helmholtz equations, demonstrate the significant advantages of DBAW-PIKAN in enhancing both accuracy and generalization performance.

Task Vector in TTS: Toward Emotionally Expressive Dialectal Speech Synthesis

Recent advances in text-to-speech (TTS) have yielded remarkable improvements in naturalness and intelligibility. Building on these achievements, research has increasingly shifted toward enhancing the expressiveness of generated speech, such as dialectal and emotional TTS. However, cross-style synthesis combining both dialect and emotion remains challenging and largely unexplored, mainly due to the scarcity of dialectal data with emotional labels. To address this, we propose Hierarchical Expressive Vector (HE-Vector), a two-stage method for Emotional Dialectal TTS. In the first stage, we construct different task vectors to model dialectal and emotional styles independently, and then enhance single-style synthesis by adjusting their weights, a method we refer to as Expressive Vector (E-Vector). For the second stage, we hierarchically integrate these vectors to achieve controllable emotionally expressive dialect synthesis without requiring jointly labeled data, corresponding to Hierarchical Expressive Vector (HE-Vector). Experimental results demonstrate that HE-Vectors achieve superior performance in dialect synthesis, and promising results in synthesizing emotionally expressive dialectal speech in a zero-shot setting.

Alzheimer's Dementia Detection Using Perplexity from Paired Large Language Models

Alzheimer's dementia (AD) is a neurodegenerative disorder with cognitive decline that commonly impacts language ability. This work extends the paired perplexity approach to detecting AD by using a recent large language model (LLM), the instruction-following version of Mistral-7B. We improve accuracy by an average of 3.33% over the best current paired perplexity method and by 6.35% over the top-ranked method from the ADReSS 2020 challenge benchmark. Our further analysis demonstrates that the proposed approach can effectively detect AD with a clear and interpretable decision boundary in contrast to other methods that suffer from opaque decision-making processes. Finally, by prompting the fine-tuned LLMs and comparing the model-generated responses to human responses, we illustrate that the LLMs have learned the special language patterns of AD speakers, which opens up possibilities for novel methods of model interpretation and data augmentation.

TextRegion: Text-Aligned Region Tokens from Frozen Image-Text Models

Image-text models excel at image-level tasks but struggle with detailed visual understanding. While these models provide strong visual-language alignment, segmentation models like SAM2 offer precise spatial boundaries for objects. To this end, we propose TextRegion, a simple, effective, and training-free framework that combines the strengths of image-text models and SAM2 to generate powerful text-aligned region tokens. These tokens enable detailed visual understanding while preserving open-vocabulary capabilities. They can be directly applied to various downstream tasks, including open-world semantic segmentation, referring expression comprehension, and grounding. We conduct extensive evaluations and consistently achieve superior or competitive performance compared to state-of-the-art training-free methods. Additionally, our framework is compatible with many image-text models, making it highly practical and easily extensible as stronger models emerge. Code is available at: https://github.com/avaxiao/TextRegion.

Towards Explainable Temporal Reasoning in Large Language Models: A Structure-Aware Generative Framework

While large language models (LLMs) show great potential in temporal reasoning, most existing work focuses heavily on enhancing performance, often neglecting the explainable reasoning processes underlying the results. To address this gap, we introduce a comprehensive benchmark covering a wide range of temporal granularities, designed to systematically evaluate LLMs' capabilities in explainable temporal reasoning. Furthermore, our findings reveal that LLMs struggle to deliver convincing explanations when relying solely on textual information. To address challenge, we propose GETER, a novel structure-aware generative framework that integrates Graph structures with text for Explainable TEmporal Reasoning. Specifically, we first leverage temporal knowledge graphs to develop a temporal encoder that captures structural information for the query. Subsequently, we introduce a structure-text prefix adapter to map graph structure features into the text embedding space. Finally, LLMs generate explanation text by seamlessly integrating the soft graph token with instruction-tuning prompt tokens. Experimental results indicate that GETER achieves state-of-the-art performance while also demonstrating its effectiveness as well as strong generalization capabilities. Our dataset and code are available at https://github.com/carryTatum/GETER.

100 Days After DeepSeek-R1: A Survey on Replication Studies and More Directions for Reasoning Language Models

The recent development of reasoning language models (RLMs) represents a novel evolution in large language models. In particular, the recent release of DeepSeek-R1 has generated widespread social impact and sparked enthusiasm in the research community for exploring the explicit reasoning paradigm of language models. However, the implementation details of the released models have not been fully open-sourced by DeepSeek, including DeepSeek-R1-Zero, DeepSeek-R1, and the distilled small models. As a result, many replication studies have emerged aiming to reproduce the strong performance achieved by DeepSeek-R1, reaching comparable performance through similar training procedures and fully open-source data resources. These works have investigated feasible strategies for supervised fine-tuning (SFT) and reinforcement learning from verifiable rewards (RLVR), focusing on data preparation and method design, yielding various valuable insights. In this report, we provide a summary of recent replication studies to inspire future research. We primarily focus on SFT and RLVR as two main directions, introducing the details for data construction, method design and training procedure of current replication studies. Moreover, we conclude key findings from the implementation details and experimental results reported by these studies, anticipating to inspire future research. We also discuss additional techniques of enhancing RLMs, highlighting the potential of expanding the application scope of these models, and discussing the challenges in development. By this survey, we aim to help researchers and developers of RLMs stay updated with the latest advancements, and seek to inspire new ideas to further enhance RLMs.

FBRT-YOLO: Faster and Better for Real-Time Aerial Image Detection

Embedded flight devices with visual capabilities have become essential for a wide range of applications. In aerial image detection, while many existing methods have partially addressed the issue of small target detection, challenges remain in optimizing small target detection and balancing detection accuracy with efficiency. These issues are key obstacles to the advancement of real-time aerial image detection. In this paper, we propose a new family of real-time detectors for aerial image detection, named FBRT-YOLO, to address the imbalance between detection accuracy and efficiency. Our method comprises two lightweight modules: Feature Complementary Mapping Module (FCM) and Multi-Kernel Perception Unit(MKP), designed to enhance object perception for small targets in aerial images. FCM focuses on alleviating the problem of information imbalance caused by the loss of small target information in deep networks. It aims to integrate spatial positional information of targets more deeply into the network,better aligning with semantic information in the deeper layers to improve the localization of small targets. We introduce MKP, which leverages convolutions with kernels of different sizes to enhance the relationships between targets of various scales and improve the perception of targets at different scales. Extensive experimental results on three major aerial image datasets, including Visdrone, UAVDT, and AI-TOD,demonstrate that FBRT-YOLO outperforms various real-time detectors in terms of performance and speed.

Virtual-mask Informed Prior for Sparse-view Dual-Energy CT Reconstruction

Sparse-view sampling in dual-energy computed tomography (DECT) significantly reduces radiation dose and increases imaging speed, yet is highly prone to artifacts. Although diffusion models have demonstrated potential in effectively handling incomplete data, most existing methods in this field focus on the image do-main and lack global constraints, which consequently leads to insufficient reconstruction quality. In this study, we propose a dual-domain virtual-mask in-formed diffusion model for sparse-view reconstruction by leveraging the high inter-channel correlation in DECT. Specifically, the study designs a virtual mask and applies it to the high-energy and low-energy data to perform perturbation operations, thus constructing high-dimensional tensors that serve as the prior information of the diffusion model. In addition, a dual-domain collaboration strategy is adopted to integrate the information of the randomly selected high-frequency components in the wavelet domain with the information in the projection domain, for the purpose of optimizing the global struc-tures and local details. Experimental results indicated that the present method exhibits excellent performance across multiple datasets.

Finding the Sweet Spot: Preference Data Construction for Scaling Preference Optimization

Iterative data generation and model retraining are widely used to align large language models (LLMs). It typically involves a policy model to generate on-policy responses and a reward model to guide training data selection. Direct Preference Optimization (DPO) further enhances this process by constructing preference pairs of chosen and rejected responses. In this work, we aim to \emph{scale up} the number of on-policy samples via repeated random sampling to improve alignment performance. Conventional practice selects the sample with the highest reward as chosen and the lowest as rejected for DPO. However, our experiments reveal that this strategy leads to a \emph{decline} in performance as the sample size increases. To address this, we investigate preference data construction through the lens of underlying normal distribution of sample rewards. We categorize the reward space into seven representative points and systematically explore all 21 ($C_7^2$) pairwise combinations. Through evaluations on four models using AlpacaEval 2, we find that selecting the rejected response at reward position $\mu - 2\sigma$ rather than the minimum reward, is crucial for optimal performance. We finally introduce a scalable preference data construction strategy that consistently enhances model performance as the sample scale increases.