使用微信扫一扫分享到朋友圈
使用微信扫一扫进入小程序分享活动
作为业界最具影响力的开源盛会之一,中国开源年会 COSCon 自 2015 年由开源社首次发起,便以独特的中立社区定位、跨组织跨项目的广泛覆盖面,吸引着全球企业、高校、开源组织与开发者的目光 —— 它不同于一般行业大会,更像一场"开源人的集体奔赴",让国内外贡献者能在此碰撞思想、共筑生态。
从 2016 年"源力觉醒"点燃中国开源的星火,到 2024 年"Open Source Open Life"让开源深度融入衣食住行,再到 2025 年"众智开源 Open Source, Open Intelligence"让众智在开放中汇聚 —— 十一载春秋,中国开源年会 COSCon 见证着开源从技术圈的"小众实践",成长为驱动产业创新的"核心力量"。
如今,COSCon 走进第十一届。当技术范式加速更迭、AI 重塑生产方式、全球协作进入新阶段,我们正式宣布本届主题:开源下一代,OpenSource:The Next Generation —— 开源的下一个时代,将由下一代开源技术、下一代开源人才、下一代开源社区、下一代开源商业共同定义。不是等待未来发生,而是用开源的方式,去打开下一代。
从第一届到第十一届,是一个阶段的沉淀,更是一个新阶段的开始。我们诚邀开源社区、开发者和开源爱好者们加入我们,一起成为 COSCon'26 的共建者,成为"下一代"的一部分。
As one of the most influential open-source events globally, the China Open Source Conference (COSCon) has drawn attention from enterprises, universities, open-source organizations and developers worldwide since its launch by KAIYUANSHE in 2015. With its unique neutral community positioning and broad coverage across organizations and projects, it stands apart from regular industry conferences — more like a "collective gathering of open-source enthusiasts" where contributors at home and abroad exchange ideas and co-build the ecosystem.
From "Awakening of Open Source Power" in 2016 that ignited the spark of China's open-source movement, to "Open Source Open Life" in 2024 that integrated open source deeply into daily life, to "Open Source, Open Intelligence" in 2025 that pooled collective wisdom in the open — over eleven years, COSCon has witnessed open source evolve from a "niche practice" in the tech circle to a "core driver" of industrial innovation.
Now COSCon enters its 11th edition. As technology paradigms shift faster, AI reshapes how we produce, and global collaboration enters a new stage, we are proud to announce this year's theme: OpenSource: The Next Generation — the next era of open source will be defined by the next generation of open-source technology, talent, community and business. Not waiting for the future to happen, but opening the next generation the open-source way.
From the first edition to the eleventh, this is the settling of one chapter and the beginning of the next. We sincerely invite the open-source community, developers and enthusiasts to join us — to become a co-builder of COSCon'26, and part of "the next generation."
大模型与 Coding Agent 已经将代码编写压缩至分钟级,AI 时代技术人才的核心竞争力正在向「完整生产交付与持续运维闭环」延伸。本场分享从 Sealos 演进为桌面原生 AI 云工作台的一线实践出发,剖析两类新人才的崛起:一是以业务研发兼管全栈上线的 2–20 人精简团队与接单工作室,二是借助自然语言配置跑通 9000+ 开源项目的超级个体。分享将拆解如何通过专属 Ops Agent、模板参数自动填充(Auto-fill)、部署方案确认卡与同集群私有内网画布,将容器镜像构建、数据库高可用配网、AI 网关接入与线上排障收敛为自然语言交互,让单兵与小团队直接具备全建制产研运交付力。
The Kubernetes era made declarative orchestration central to cloud-native infrastructure. AI-native infrastructure advances this approach by making models, accelerator memory, inference replicas, and token delivery central design concerns. For small teams, this evolution also demands a simpler operational footprint. GPUWorks explores an independently designed control plane for multiple GPU hosts, requiring neither Kubernetes nor OpenStack. Its approximately 45 MB compressed management package embeds a web interface, model gateway, storage, observability, and Linux agents. Self-contained Go binaries provide Linux and macOS builds for x86-64 and ARM64, with no separate Python environment, database service, or monitoring stack required for management. It supports NVIDIA, Huawei Ascend, Hygon, PPU, and Kunlun accelerators across owned, remote, and rented hosts. Capability checks and runtime adapters accommodate supported accelerator families. Deployment recipes capture reusable configuration knowledge; hardware and memory checks expose constraints before launch. Shared configuration resolution keeps deployment previews aligned with execution, while integrated status and logs support diagnosis. These design choices bring model requirements and service readiness into everyday operations, with the aim of reducing manual preparation and infrastructure maintenance. Attendees will gain reusable design principles for lightweight AI-native infrastructure and a framework for evaluating dependency footprint, multi-host correctness, and maintenance cost. The open-source value lies in sharing management patterns, node interfaces, and validation methods across an ecosystem of complementary infrastructure approaches.
Q1:在 AI 时代,你这一层最不能被闭源或单一厂商锁死的能力是什么?
Q2:开源项目的护城河会从代码转向什么——可验证性、数据、标准,还是社区信任?
Q3:发起 AI Infra 开源项目的人、12 个月内可执行开源建议是什么?
Blade AI 能把一句对话目标变成一次完整的混沌实验:规划场景、执行注入、验证效果、恢复并出报告。它的规划与取证由 LLM 完成,而 LLM 一定会出错——所以我们不承诺正确的模型,只承诺错误有界、可见、可恢复。分享将拆解四道护栏:写操作仅限人工立法的允许集;故障创建超时先与实时集群核对再放行重试;效果裁决由模型做出、但被关进封闭词汇集,“无法确认”会扣住故障直到恢复执行;恢复独立于 Agent 进程,由集群内定时载体幂等重放。内容源于真实演练,面向想把 Agent 引入生产运维的平台与 SRE 团队。
人工智能正在改变药物研发的方式,但从设计出一个具有活性的分子,到真正成为临床可用的药物,仍面临稳定性、膜通透性和体内递送等诸多挑战。本次分享将结合镜像多肽、环肽和双环肽的研究实践,介绍人工智能如何突破天然分子的设计边界,通过手性转换、拓扑设计和非天然氨基酸等策略,探索更广阔的分子空间。
标题并非写错,是 AI Naive, 而非 AI Native. 在本次演讲中,作者希望通过一年来十几个软件项目 AI Coding 的实战经验,介绍这种 AI Naive 软件开发方法:用 "Naive"(空杯) 的心态,在 "Naive"(陌生) 的领域,做出不"Naive" 的软件产品。个人部分软件作品发布在 ainaive.com 上,所有项目代码也都开源在 github 上,可参见个人 github: https://github.com/hutusi . 部分经历曾写了两篇博客来介绍:《关于AI编程的一些体会和设想》 https://hutusi.com/weeklies/24 《如何让AI写好代码》 https://hutusi.com/weeklies/25
面向 RISC-V AI 生态中接口不统一、框架适配成本高与硬件性能释放不足等挑战,本报告围绕 RVUK 统一微内核接口,介绍其厂商中立的 C ABI、分层架构与可插拔内核机制,结合矩阵乘法、低比特量化及 torq-tile 适配实践,阐述如何连接底层硬件优化与上层框架需求。在此基础上,结合开源工作规划,探讨以 PyTorch 为切入点,逐步推进多框架接入、编译器协同与接口标准化的路径,推动高性能微内核跨框架复用,促进 RISC-V AI 软件生态的开放协作与共同建设。
本演讲以 "RISC-V AI 软件栈落地:从首个算子补丁到开源生态共建" 为题,分享我们围绕 RISC-V 开展的开源开放协作实践。从向 oneDNN 提交首个 RISC-V 算子补丁起步,我们逐步探索出 "标准共建、代码贡献、生态协作" 并行的开放路径:在指令集、编译器、算子库、推理框架到检索应用的多层软件栈上持续深耕,累计合入 60 余项补丁、沉淀 8 篇规范、联合 45 家伙伴共建生态。面向 AI,重点推进 oneDNN RISC-V 原生算子、faiss/Milvus/Knowhere 检索栈与 ktransformers 推理优化,实现从算子到 RAG 的端到端落地,让开源协作真正转化为可复用的算力红利。诚邀更多开发者同行,共筑 RISC-V 开源生态的下一程。
LingCage 在 riscv64 路径中实现 Rust KVM 后端、Image+FDT 直接启动,并基于 AIA 使用内核 APLIC 与 IMSIC;其快照、克隆与 vsock 通道面向短生命周期 Agent 负载。本议题对照 x86_64 的 bzImage/ACPI 与 aarch64 的 GICv3,拆解 riscv64 的机器抽象、vCPU/中断状态和 snapshot 实现,说明构建检查与实机验证现状,讨论硬件要求及向 rust-vmm 上游回馈的计划,为 RISC-V 软件生态提供可复用 crate。
从一个人的登台,到7个项目组的亮相
那些差点放弃的瞬间
关键时刻拉了一把的人
群聊之外的隐藏剧情
素未谋面的老朋友
还没实现、等你加入的脑洞
一个你可以共建的社区团聚~
问:请问购买门票是否可以开具发票?如何申请?
答:您好,购买门票可以开具发票。请您发送以下信息到 contact@kaiyuanshe.org:
发票类型:普通发票 / 专用发票
开票内容:培训费 / 会议费 / 咨询费
门票订单号(可在个人中心 - 订单与发票处查看):
订单金额:
开票信息:
我们将在活动结束后,收到邮件的 10 个工作日内为您开具发票。