北京，4月2日——APUS集团与战略伙伴新旦智能共同宣布，双方成功研发并开源了国内首个基于千亿参数的Mixture of Experts（MoE）架构大模型，该模型命名为APUS-xDAN大模型4.0。这一开创性的举措标志着中国在大规模预训练模型领域的又一重大突破，为AI技术的普及与效能提升树立了新的标杆。
据悉，APUS-xDAN大模型4.0（MoE）不仅在性能上达到了与GPT-4相当的90%综合水平，更难能可贵的是，该模型能够在4090等相对低端的算力芯片上流畅运行，极大地降低了行业应用的门槛，实现了从高端计算资源到普惠计算的转变。这一成果的开源，将为全球开发者提供一个高效、易用的AI研发平台，推动人工智能技术在更广泛的领域和场景中得到应用。
APUS与新旦智能的此次合作，不仅展现了中国企业在AI技术研发上的实力与创新精神，同时也彰显了开源精神对于科技进步的推动作用。这一千亿级MoE架构大模型的开源，有望激发更多的创新实践，促进全球AI生态的繁荣与发展。
APUS集团与新旦智能表示，将持续投入资源，与全球开发者共享技术成果，共同推动AI技术的边界不断拓展，为构建更智能、更普惠的未来贡献力量。这一里程碑式的事件，无疑将在中国乃至全球的AI发展历程中写下浓墨重彩的一笔。
英语如下：
**News Title:** “APUS and Xin旦 Intelligence Collaborate for a Domestic First: Trillion-Level MoE Large Model Open Sourced, Efficiently Operable on Low-end Chips”
**Keywords:** APUS, Trillion MoE, Open-Source Large Model
**News Content:**
**Title:** APUS and Xin旦 Intelligence Unveil the Open-Source of China’s First Trillion-Level MoE Architecture Large Model, Paving the Way for AI Innovation
**Beijing, April 2** — APUS Group and strategic partner Xin旦 Intelligence jointly announced the successful development and open-sourcing of the country’s first large model based on a Trillion-parameter Mixture of Experts (MoE) architecture, named APUS-xDAN Large Model 4.0. This groundbreaking move signifies another significant breakthrough in China’s large-scale pre-training model field and sets a new benchmark for AI technology’s普及 and performance enhancement.
It is reported that the APUS-xDAN Large Model 4.0 (MoE) achieves a 90% composite level comparable to GPT-4, and notably, it can run smoothly on relatively low-end computing chips like the 4090, significantly lowering the threshold for industry adoption and transitioning from high-end to accessible computing resources. By open-sourcing this achievement, it will provide a high-efficiency, user-friendly AI development platform for global developers, promoting the application of artificial intelligence technology across a broader range of fields and scenarios.
This collaboration between APUS and Xin旦 Intelligence not only demonstrates the strength and innovative spirit of Chinese enterprises in AI R&D but also highlights the role of open-source spirit in advancing科技进步. The open-source of this trillion-level MoE architecture large model is expected to inspire more innovative practices, fostering the prosperity and development of the global AI ecosystem.
Both APUS Group and Xin旦 Intelligence have committed to continued resource investment, sharing technological achievements with global developers, and jointly pushing the boundaries of AI technology to contribute to a smarter and more inclusive future. This milestone event will undoubtedly leave a profound mark in China’s and the world’s AI development history.
【来源】https://mp.weixin.qq.com/s/x_XSfcDBL7A4zb43vDFSMg
Views: 1