Workflow
Seed Diffusion Preview
icon
Search documents
字节跳动发布全球最快代码生成AI:2146倍速度碾压传统模型
Sou Hu Cai Jing· 2025-08-08 14:52
Core Insights - The article discusses a groundbreaking advancement in AI code generation technology called "Seed Diffusion Preview," developed by ByteDance's Seed team in collaboration with Tsinghua University's Intelligent Industry Research Institute. This technology significantly enhances the speed of code generation, achieving an impressive rate of 2146 tokens per second on H20 GPUs, which is several times faster than traditional models [2][3][15]. Group 1: Traditional Code Generation Challenges - Traditional code generation models are limited by their autoregressive nature, which requires generating code tokens sequentially, leading to bottlenecks in speed and efficiency [3][4]. - The new Seed Diffusion model overcomes these limitations by employing a discrete state diffusion model, allowing for parallel processing of code generation, akin to a multi-threaded programming approach [5][6]. Group 2: Training Methodology - The training process of Seed Diffusion utilizes a two-stage curriculum learning approach, which gradually develops the model's capabilities from basic recognition to complex code generation [6][7]. - The first stage focuses on noise reduction through masked and edited training processes, while the second stage employs a customized trajectory space diffusion training to optimize the generation paths [8][9]. Group 3: Performance Metrics - Seed Diffusion has demonstrated exceptional performance across various coding benchmarks, achieving 85.2% and 79.4% success rates in foundational programming tests, and 76.0% in real-world coding scenarios [15][16]. - The model also excels in code editing tasks, with scores of 44.4% and 54.3% in relevant benchmarks, indicating its capability to understand and improve existing code structures [17]. Group 4: Industry Impact - The introduction of Seed Diffusion is expected to revolutionize the software development landscape by significantly reducing coding time and costs, allowing developers to focus on higher-level tasks [19][21]. - This technology could lead to a shift in software development practices, encouraging more modular and standardized approaches, as well as altering educational focuses towards algorithmic thinking and system design [24][25]. Group 5: Competitive Landscape - Seed Diffusion establishes a notable competitive advantage over existing models like Mercury Coder and Gemini Diffusion, showcasing superior speed and quality metrics [26][27]. - The open-source strategy adopted by ByteDance may further influence the industry by promoting higher technical standards and fostering innovation among developers [27]. Group 6: Future Challenges - Despite its advancements, Seed Diffusion faces challenges in scaling to more complex coding tasks and ensuring code quality and security in real-world applications [28][29]. - The model's reliance on high-quality training data and the need for user-friendly interfaces are critical areas for ongoing development and improvement [29][30].
AI动态汇总:智谱发布GLM-4.5,蚂蚁数科发布金融推理大模型Agentar-Fin-R1
China Post Securities· 2025-08-06 02:33
- The GLM-4.5 model, developed by Zhipu, integrates reasoning, coding, and intelligent agent capabilities into a single architecture. It employs a hybrid expert framework with 355 billion total parameters, activating only 32 billion parameters per inference to enhance computational efficiency. The training process includes three stages: pretraining on 15 trillion general text tokens, fine-tuning on 8 trillion specialized data, and reinforcement learning for multi-task alignment. The model achieves a 37% performance improvement in complex reasoning tasks through innovations like deep-layer prioritization and grouped query attention mechanisms [12][14][15] - GLM-4.5 ranks third globally in AGI core capability evaluations, with a composite score of 63.2. It outperforms competitors in tasks such as web interaction (26.4% accuracy in BrowseComp) and code repair (64.2 in SWE-bench Verified). The model demonstrates an 80.8% win rate against Qwen3-Coder in 52 real-world programming tasks, despite having half the parameters of DeepSeek-R1, showcasing its superior performance-to-parameter ratio [15][16][19] - The Agentar-Fin-R1 model, launched by Ant Financial, is a financial reasoning model based on the Qwen3 architecture. It features a dual-engine design: the Master Builder engine translates business logic into executable code, while the Agent Group engine uses consensus algorithms for multi-agent decision-making. The model is trained on a domain-specific corpus covering six major financial sectors, achieving a financial knowledge accuracy rate of 92.3% through weighted training algorithms [20][21][23] - Agentar-Fin-R1 excels in financial evaluations, scoring 87.70 in FinEval1.0 and 86.79 in FinanceIQ. It leads in tasks like risk pricing and compliance review, with a score of 69.93 in the Finova evaluation, surpassing larger general-purpose models. Its compliance system improves review efficiency by 90%, and its credit approval module reduces loan processing time from 3 days to 15 minutes while lowering bad debt rates by 18% [23][24][25] - The Goedel-Prover-V2 theorem-proving system, developed by Princeton, Tsinghua, and NVIDIA, uses an 8B/32B parameter model to achieve state-of-the-art results. It employs scaffolded data synthesis, validator-guided self-correction, and model averaging to enhance performance. The system achieves 88.1% Pass@32 accuracy on the MiniF2F benchmark, with the 8B model reaching 83.3% of the performance of the 671B DeepSeek-Prover-V2 while using only 1/100th of the parameters [58][60][61] - Goedel-Prover-V2 demonstrates exceptional efficiency, with its 32B model solving 64 problems in the PutnamBench competition at Pass@64, outperforming the 671B DeepSeek-Prover-V2, which required Pass@1024 to solve 47 problems. The system's iterative self-correction mode improves proof quality with minimal token consumption increase, and its training process is highly efficient, requiring only 12 hours per iteration on 4 H100 GPUs [60][61][63]
产业观察:【AI产业跟踪】字节开源AI Agent Coze
AI Industry Trends - ByteDance has open-sourced its AI Agent "Coze," which supports commercial use and has over 6,000 stars on GitHub, providing a platform for developing intelligent agents without coding[14] - The "Step 3" model by Jieyue features 321 billion total parameters and 38 billion activated parameters, achieving a 300% inference efficiency compared to DeepSeek-R1, with expected revenue of nearly $1 billion in 2025[11] - Ant Group released the financial reasoning model "Agentar-Fin-R1," which outperforms similar models in multiple financial evaluations and is based on a comprehensive financial dataset[16] AI Applications and Platforms - SenseTime launched the "Wuneng" embodied intelligence platform, featuring a multimodal reasoning model that improves cross-modal reasoning accuracy by 5 times compared to Gemini 2.5 Pro[8] - Huawei introduced the AI-Box platform, designed for lightweight edge deployment, supporting local execution of multimodal large models with low power consumption[9] - Tencent's Tairos platform offers modular services for multimodal perception and planning, focusing on enhancing robotic software capabilities[10] AI Model Developments - Zhiyuan released the GLM-4.5 model, which integrates reasoning, programming, and agent capabilities, achieving top performance in global open-source model benchmarks[17] - JD Cloud announced the open-source enterprise-level intelligent agent "JoyAgent," which supports multi-agent collaboration and has been tested in over 20,000 internal applications[18] - ByteDance and Nanjing University developed the CriticLean framework, improving the accuracy of mathematical formalization from 38% to 84%[19] Market Risks - AI software sales are below expectations, leading to adjustments in capital expenditure plans and slower iteration speeds for core AI products[34]
字节Seed发布扩散语言模型,推理速度达2146 tokens/s,比同规模自回归快5.4倍
量子位· 2025-08-01 04:23
闻乐 发自 凹非寺 量子位 | 公众号 QbitAI 用 扩散模型 写代码,不仅像开了倍速,改起来还特别灵活! 字节Seed最新发布扩散语言模型 Seed Diffusion Preview ,这款模型主要聚焦于代码生成领域,它的特别之处在于采用了离散状态扩散技 术,在推理速度上表现出色。 在H20上,它的代码推理速度能达到 2146tokens/s ,比同类的Mercury和Gemini Diffusion等模型快不少,同时 比同等规模的自回归模型 快5.4倍 ,并且在代码编辑任务中更具优势。 Seed Diffusion Preview以 结构化的代码生成 为实验领域,系统性地验证离散扩散技术路线作为下一代语言模型基础框架的可行性。 下面介绍它的具体技术细节。 核心是两阶段训练学习 自回归模型存在串行解码延迟瓶颈,理论上,扩散模型的并行生成潜力和整体性生成的优势可以解决自回归模型推理速度局限这一痛点。 但理论优势与实际效果还是有差距,离散扩散模型在语言任务中的大规模部署仍面临两大核心瓶颈: 归纳偏置冲突 和 推理效率瓶 颈 。 为解决上述问题,Seed Diffusion Preview采用了四项关键的技 ...
英伟达H20算力芯片被曝存在严重安全问题;乐道沈斐截胡理想广告,讽刺友商暗搓搓请水军;罗马仕进入破产程序?内部员工:9月底定生死
雷峰网· 2025-08-01 00:41
Key Points - The article discusses various recent developments in the automotive, technology, and finance sectors, highlighting competitive dynamics and strategic moves by companies like Li Auto, ByteDance, NVIDIA, and JD.com [1][5][24]. Group 1: Automotive Industry - Li Auto launched its new electric SUV, the Li i8, which has sparked competitive responses from other brands, particularly regarding advertising and performance claims [1][9]. - The launch of the Li i8 coincides with the introduction of the LeDao L90, another electric SUV, which is positioned as a strong competitor in the family-oriented electric vehicle market [27]. - Concerns have been raised about the safety and performance claims of the Li i8 after a collision test video was released, leading to disputes with other manufacturers [9][10]. Group 2: Technology Sector - NVIDIA's H20 chip has been flagged for serious security vulnerabilities, prompting a government inquiry into its safety for Chinese users [5]. - ByteDance has clarified its employee retention statistics, stating that the average tenure is around 3 years, countering rumors of high turnover rates [8]. - OpenAI's annual revenue has reportedly surged to $12 billion, with ChatGPT's weekly active users surpassing 700 million, indicating strong market demand for AI services [37]. Group 3: Financial Developments - JD.com announced a voluntary public acquisition offer for German retailer Ceconomy, valuing the company at approximately €22.3 billion (around $26.3 billion) [24]. - Meta is planning to invest heavily in AI infrastructure, with projected capital expenditures reaching between $66 billion and $72 billion by 2025, reflecting a significant commitment to AI development [38].
字节跳动Seed团队发布扩散语言模型,每秒推理速度2146 tokens
news flash· 2025-07-31 12:35
7月31日,字节跳动Seed团队发布实验性扩散语言模型Seed Diffusion Preview。据介绍,其目标是以结构 化的代码生成为实验领域,系统性地验证离散扩散技术路线作为下一代语言模型基础框架的可行性。实 验结果显示,Seed Diffusion Preview代码推理速度可达到2146tokens/s,速度相比同等规模的自回归模型 提升5.4倍。(界面快讯) ...