Shuguang Wang

ByteDance Seed, Jiaze Chen, Tiantian Fan et al. · bytedance

We introduce Seed1.5-Thinking, capable of reasoning through thinking before responding, resulting in improved performance on a wide range of benchmarks. Seed1.5-Thinking achieves 86.7 on AIME 2024, 55.0 on Codeforces and 77.3 on GPQA, demonstrating excellent reasoning abilities in STEM and coding. Beyond reasoning tasks, the method demonstrates notable generalization across diverse domains. For instance, it surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability. Compared to other state-of-the-art reasoning models, Seed1.5-Thinking is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters. As part of our effort to assess generalized reasoning, we develop two internal benchmarks, BeyondAIME and Codeforces, both of which will be publicly released to support future research. Model trial link: https://www.volcengine.com/experience/ark.

Shiyi Kong, Minyan Lu, Jun Ai et al.

Modern software systems become too complex to be tested and validated. Detecting software partial failures in complex systems at runtime assist to handle software unintended behaviors, avoiding catastrophic software failures and improving software runtime availability. These detection techniques aim to find the manifestation of faults before they finally lead to unavoidable failures, thus supporting following runtime fault tolerant techniques. We review the state of the art articles and find that the content failures account for the majority of all kinds of software failures, but its detection methods are rarely studied. In this work, we propose a novel failure detection indicator based on the software runtime dynamic execution information for software content failures. The runtime information is recorded during software execution, then transformed to a measure named runtime entropy and finally fed into machine learning models. The machine learning models are built to classify the intended and unintended behaviors of the objected software systems. A series of controlled experiments on several open source projects are conducted to prove the feasibility of the method. We also evaluate the accuracy of machine learning models built in this work.

Yangtao Deng, Xiang Shi, Zhuo Jiang et al.

Large-scale distributed model training requires simultaneous training on up to thousands of machines. Faulty machine detection is critical when an unexpected fault occurs in a machine. From our experience, a training task can encounter two faults per day on average, possibly leading to a halt for hours. To address the drawbacks of the time-consuming and labor-intensive manual scrutiny, we propose Minder, an automatic faulty machine detector for distributed training tasks. The key idea of Minder is to automatically and efficiently detect faulty distinctive monitoring metric patterns, which could last for a period before the entire training task comes to a halt. Minder has been deployed in our production environment for over one year, monitoring daily distributed training tasks where each involves up to thousands of machines. In our real-world fault detection scenarios, Minder can accurately and efficiently react to faults within 3.6 seconds on average, with a precision of 0.904 and F1-score of 0.893.

Jinkun Lin, Ziheng Jiang, Zuquan Song et al.

Large language model (LLM) training is one of the most demanding distributed computations today, often requiring thousands of GPUs with frequent synchronization across machines. Such a workload pattern makes it susceptible to stragglers, where the training can be stalled by few slow workers. At ByteDance we find stragglers are not trivially always caused by hardware failures, but can arise from multiple complex factors. This work aims to present a comprehensive study on the straggler issues in LLM training, using a five-month trace collected from our ByteDance LLM training cluster. The core methodology is what-if analysis that simulates the scenario without any stragglers and contrasts with the actual case. We use this method to study the following questions: (1) how often do stragglers affect training jobs, and what effect do they have on job performance; (2) do stragglers exhibit temporal or spatial patterns; and (3) what are the potential root causes for stragglers?

Diandian Gu, Jing Lin, Gaohong Liu et al.

We present Seed3D 2.0, an advanced 3D content generation system built on Seed3D 1.0, with substantial improvements across generation fidelity, simulation-ready capabilities, and application coverage. For geometry, a coarse-to-fine two-stage pipeline decouples global structure learning from high-frequency detail recovery, while a locality-aware VAE achieves higher spatial compression and more efficient decoding. For texture and material generation, we replace the cascaded pipeline of Seed3D 1.0 with a unified PBR model that directly generates multi-view albedo and metallic-roughness maps, enhanced by Mixture-of-Experts scaling and VLM-based semantic conditioning for improved material precision and visual fidelity. Beyond single-object generation, Seed3D 2.0 introduces a simulation-ready model suite comprising scene layout planning, part-aware decomposition, and training-free articulation generation, enabling coherent scene construction and part-level physical interaction across physics and graphics engines. A large-scale human preference study against five recent commercial models shows that Seed3D 2.0 achieves consistent win rates of 69.0% to 89.9% in textured 3D asset generation. Seed3D 2.0 is available on https://exp.volcengine.com/ark/vision?_vtm_=0.0.c70961.d701978.0&mode=vision&modelId=doubao-seed3d-2-0-260328&tab=Gen3D

Borui Wan, Gaohong Liu, Zuquan Song et al.

The training scale of large language models (LLMs) has reached tens of thousands of GPUs and is still continuously expanding, enabling faster learning of larger models. Accompanying the expansion of the resource scale is the prevalence of failures (CUDA error, NaN values, job hang, etc.), which poses significant challenges to training stability. Any large-scale LLM training infrastructure should strive for minimal training interruption, efficient fault diagnosis, and effective failure tolerance to enable highly efficient continuous training. This paper presents ByteRobust, a large-scale GPU infrastructure management system tailored for robust and stable training of LLMs. It exploits the uniqueness of LLM training process and gives top priorities to detecting and recovering failures in a routine manner. Leveraging parallelisms and characteristics of LLM training, ByteRobust enables high-capacity fault tolerance, prompt fault demarcation, and localization with an effective data-driven approach, comprehensively ensuring continuous and efficient training of LLM tasks. ByteRobust is deployed on a production GPU platform and achieves 97% ETTR for a three-month training job on 9,600 GPUs.

Yangtao Deng, Lei Zhang, Qinlong Wang et al.

Reliability is essential for ensuring efficiency in LLM training. However, many real-world reliability issues remain difficult to resolve, resulting in wasted resources and degraded model performance. Unfortunately, today's collective communication libraries operate as black boxes, hiding critical information needed for effective root cause analysis. We propose Mycroft, a lightweight distributed tracing and root cause analysis system designed to address previously hidden reliability issues in collective communication. Mycroft's key idea is to trace collective communication states and leverage internal control and data dependencies to resolve reliability problems in LLM training. Mycroft has been deployed at ByteDance for over six months to debug collective communication related issues at runtime. It detected anomalies within 15 seconds in 90% of cases and identified the root cause within 20 seconds in 60% of cases. We also conducted extensive fault injection experiments to demonstrate Mycroft's capability and efficiency.

Jie Cao, Ha Nguyen, Selim Yavuz et al.

Large Language Model (LLM) simulations, where LLMs act as students with varying approaches to learning tasks, can support teachers' noticing of student thinking. However, simulations using zero- or few-shot prompting often yield inauthentic knowledge and language, directing teachers to unrealistic reasoning. We evaluate three approaches (Fine-tuning, Multi-agent, and Direct Preference Optimization; DPO) to improve the authenticity and pedagogical utility of simulated students. All approaches improve cognitive and linguistic authenticity, compared with few-shot prompts. Interviews with elementary mathematics pre-service teachers and researchers (\textit{n} = 8) reveal distinct pedagogical affordances. The fine-tuned model produces realistic, brief responses but limits opportunities to extend students' thinking. Meanwhile, the multi-agent and DPO approaches generate explicit reasoning behind student strategies. We discuss implications for designing LLM simulations that balance authenticity with instructional utility for teacher learning.

Jiashi Feng, Xiu Li, Jing Lin et al.

Developing embodied AI agents requires scalable training environments that balance content diversity with physics accuracy. World simulators provide such environments but face distinct limitations: video-based methods generate diverse content but lack real-time physics feedback for interactive learning, while physics-based engines provide accurate dynamics but face scalability limitations from costly manual asset creation. We present Seed3D 1.0, a foundation model that generates simulation-ready 3D assets from single images, addressing the scalability challenge while maintaining physics rigor. Unlike existing 3D generation models, our system produces assets with accurate geometry, well-aligned textures, and realistic physically-based materials. These assets can be directly integrated into physics engines with minimal configuration, enabling deployment in robotic manipulation and simulation training. Beyond individual objects, the system scales to complete scene generation through assembling objects into coherent environments. By enabling scalable simulation-ready content creation, Seed3D 1.0 provides a foundation for advancing physics-based world simulators. Seed3D 1.0 is now available on https://console.volcengine.com/ark/region:ark+cn-beijing/experience/vision?modelId=doubao-seed3d-1-0-250928&tab=Gen3D

Rui Li, Xiaoyun Zhi, Jinxin Chi et al.

Large Language Models (LLMs) have become a cornerstone of modern AI, driving breakthroughs in natural language processing and expanding into multimodal jobs involving images, audio, and video. As with most computational software, it is important to distinguish between ordinary runtime performance and startup overhead. Prior research has focused on runtime performance: improving training efficiency and stability. This work focuses instead on the increasingly critical issue of startup overhead in training: the delay before training jobs begin execution. Startup overhead is particularly important in large, industrial-scale LLMs, where failures occur more frequently and multiple teams operate in iterative update-debug cycles. In one of our training clusters, more than 3.5% of GPU time is wasted due to startup overhead alone. In this work, we present the first in-depth characterization of LLM training startup overhead based on real production data. We analyze the components of startup cost, quantify its direct impact, and examine how it scales with job size. These insights motivate the design of Bootseer, a system-level optimization framework that addresses three primary startup bottlenecks: (a) container image loading, (b) runtime dependency installation, and (c) model checkpoint resumption. To mitigate these bottlenecks, Bootseer introduces three techniques: (a) hot block record-and-prefetch, (b) dependency snapshotting, and (c) striped HDFS-FUSE. Bootseer has been deployed in a production environment and evaluated on real LLM training workloads, demonstrating a 50% reduction in startup overhead.

Shuguang Wang, Yuanjing Wang

Wang and Wang (2025) proposed the Task-Driven Fixation Network (TDFN) based on the fixation mechanism, which leverages low-resolution information along with high-resolution details near fixation points to accomplish specific visual tasks. The model employs reinforcement learning to generate fixation points. However, training reinforcement learning models is challenging, particularly when aiming to generate pixel-level accurate fixation points on high-resolution images. This paper introduces an improved fixation point generation method by leveraging the difference between the reconstructed image and the input image to train the fixation point generator. This approach directs fixation points to areas with significant differences between the reconstructed and input images. Experimental results demonstrate that this method achieves highly accurate fixation points, significantly enhances the network's classification accuracy, and reduces the average number of required fixations to achieve a predefined accuracy level.

Shuguang Wang, Yuanjing Wang

This paper presents a novel neural network architecture featuring automatic fixation point selection, designed to efficiently address complex tasks with reduced network size and computational overhead. The proposed model consists of: a low-resolution channel that captures low-resolution global features from input images; a high-resolution channel that sequentially extracts localized high-resolution features; and a hybrid encoding module that integrates the features from both channels. A defining characteristic of the hybrid encoding module is the inclusion of a fixation point generator, which dynamically produces fixation points, enabling the high-resolution channel to focus on regions of interest. The fixation points are generated in a task-driven manner, enabling the automatic selection of regions of interest. This approach avoids exhaustive high-resolution analysis of the entire image, maintaining task performance and computational efficiency.

Shiyi Kong, Jun Ai, Minyan Lu et al.

Modern software systems have become increasingly complex, which makes them difficult to test and validate. Detecting software partial anomalies in complex systems at runtime can assist with handling unintended software behaviors, avoiding catastrophic software failures and improving software runtime availability. These detection techniques aim to identify the manifestation of faults (anomalies) before they ultimately lead to unavoidable failures, thus, supporting the following runtime fault-tolerant techniques. In this work, we propose a novel anomaly detection method named DistAD, which is based on the distribution of software runtime dynamic execution traces. Unlike other existing works using key performance indicators, the execution trace is collected during runtime via intrusive instrumentation. Instrumentation are controlled following a sampling mechanism to avoid excessive overheads. Bi-directional Long Short-Term Memory (Bi-LSTM), an architecture of Recurrent Neural Network (RNN) is used to achieve the anomaly detection. The whole framework is constructed under a One-Class Neural Network (OCNN) learning mode which can help eliminate the limits of lacking for enough labeled samples and the data imbalance issues. A series of controlled experiments are conducted on a widely used database system named Cassandra to prove the validity and feasibility of the proposed method. Overheads brought about by the intrusive probing are also evaluated. The results show that DistAD can achieve more than 70% accuracy and 90% recall (in normal states) with no more than 2 times overheads compared with unmonitored executions.