Zeng Guoyang, CTO of Mianbi Intelligence, pointed out that on-device models are the key path for AI implementation. Its original methodology, the "Model Wi...
Zeng Guoyang, CTO of Mianbi Intelligence, pointed out that on-device models are the key path for AI
implementation. Its original methodology, the "Model Wind Tunnel," can predict full training performance in small-scale experiments and proposes the "Wall-Facing Law" based on "knowledge density": knowledge density doubles every 3.5 months. The MiniCPM performance of 2B parameters outperforms competing 8B models of the same period. FaceWall has already completed chip adaptation for Qualcomm, MediaTek, Intel, Nvidia, AMD, and others. The newly released BitCPM-CANN model series can install up to six times the number of models on the same memory on Huawei Ascend chips. The full-duplex, full-modal model MiniCPM-o4.5 supports real-time interrupts and mood adjustments. The team developed the world's first production-grade training framework written entirely by AI, ForgeTrain, and introduced a behavioral pattern library to achieve a "tacit understanding system" that requires no opening of the interface.
面壁智能CTO曾国洋指出，端侧模型是AI落地的关键路径。
其原创方法论"模型风洞"可在小规模实验中预测完整训练效果，并基于"知识密度"提出"面壁定律"：知识密度每3.5个月翻一番。
2B参数的MiniCPM表现优于同期8B竞品。
面壁已完成高通、联发科、英特尔、英伟达、AMD等芯片适配，新发布的BitCPM-CANN模型系列可在华为昇腾芯片上让同一内存多装约6倍模型。
全双工全模态模型MiniCPM-o4.5支持实时打断与情绪调整。
团队开发了全球首个完全由AI编写的生产级训练框架ForgeTrain，并引入行为模式库实现无需开口的"默契系统"。
面壁智能CTO曾国洋在专访中将端侧模型视为AI落地的关键路径，并介绍模型训练、芯片适配和多模态交互方面的进展。材料称，MiniCPM、BitCPM-CANN、MiniCPM-o4.5及ForgeTrain已形成相关技术布局。
材料将模型能力提升归因于“模型风洞”和以知识密度衡量效率的“面壁定律”，并称知识密度每3.5个月翻一番、2B参数MiniCPM优于同期8B竞品。这些结论均来自企业公众号及受访者表述。
Aioga判断，端侧路线的核心价值可能在于以较小模型、硬件适配和实时交互降低部署约束。不过，“优于竞品”“全球首个”及效率提升等说法目前只有来源方材料支持，仍需独立测试与公开技术细节验证。
若材料所述能力在不同硬件和真实任务中得到验证，端侧AI可能获得更高的部署密度与交互灵活性。值得关注的是，覆盖多类芯片、支持实时打断和情绪调整，并不等同于已实现规模化商业落地。 后续应关注模型风洞的预测方法、知识密度的计算口径、MiniCPM对比测试条件，以及BitCPM-CANN“同一内存多装约6倍模型”的基线与适用范围，同时核验ForgeTrain和行为模式库的实际生产表现。
The readable text on this page was extracted from the public source and organized with attribution, publication time and the original link. Copyright remains with the original author and publisher.
Ingestion channel: Summary aggregation · Source domain: mp.weixin.qq.com
Source: 公众号：面壁智能（MiniCPM）
Original link: Open original source
Aioga archive: Open intelligence page
Content record: summary-fallback · Updated: 2026-07-14T02:29:00.000Z
统一接入主流 AI 模型 API，为开发、测试与生产环境提供稳定调用入口。
立即访问 api.w173.com
Scan to open this articleAioga aggregates global AI updates and preserves source information for verification and citation.