【北京，4月2日】全球知名的移动互联网公司APUS与人工智能企业新旦智能携手，近日在国内开创性地开源了首个基于千亿级MoE（Mixture of Experts）架构的大模型。这一重大突破不仅标志着我国在大规模预训练模型技术领域迈出了新的一步，也为行业普及和效能提升树立了新的标杆。
据悉，APUS与新旦智能共同研发的APUS-xDAN大模型4.0（MoE）在GitHub上公开源代码，使得广大开发者和研究者得以共享这一先进技术。该模型的一大亮点在于，它能够在4090等相对低端的算力芯片上流畅运行，实现了高效率与低成本的完美平衡。据官方数据显示，APUS-xDAN大模型4.0在综合性能上已达到国际顶尖的GPT-4的90%，展现出强大的计算效能和应用潜力。
这一开源举措对于推动我国人工智能领域的创新和发展具有重要意义。它降低了大规模模型应用的门槛，使得更多企业和个人能够利用低端硬件进行复杂的人工智能运算，从而推动AI技术在更广泛的产业中落地应用。同时，APUS与新旦智能的强强联合，也展示了我国在AI领域的技术实力和开放合作的态度，有望在全球范围内引发新一轮的AI技术研发热潮。
APUS创始人兼CEO李涛表示：“我们希望通过开源APUS-xDAN大模型4.0（MoE），激发更多开发者和企业的创新活力，共同推动AI技术的普惠与进步。我们相信，这样的开放合作将为整个行业带来更大的价值。”新旦智能也对此表示，开源千亿级MoE架构大模型是公司践行技术创新与开放共享理念的重要实践，期待与业界同仁共同探索AI的无限可能。
此次APUS与新旦智能的联合开源行动，无疑为我国乃至全球的AI研发开辟了新的道路，将为人工智能技术的广泛应用和产业升级注入新的动力。
英语如下：
**News Title:** “APUS and Xin旦 Intelligence Create a Miracle: China’s First Open-Source Trillion-Level MoE Large Model Emerges”
**Keywords:** APUS, Trillion MoE, Open-Source Large Model
**News Content:**
**Beijing, April 2** — Renowned mobile internet company APUS, in collaboration with artificial intelligence firm Xin旦 Intelligence, recently made history in China by open-sourcing the country’s first large model based on a trillion-scale MoE (Mixture of Experts) architecture. This significant breakthrough signifies a new stride in China’s development of large-scale pre-training model technology and sets a new benchmark for industry adoption and performance enhancement.
It is reported that the jointly developed APUS-xDAN Large Model 4.0 (MoE) by APUS and Xin旦 Intelligence has released its source code on GitHub, allowing a wide range of developers and researchers to access this advanced technology. A standout feature of this model is its ability to run smoothly on relatively low-end computing chips like the 4090, achieving an optimal balance between efficiency and cost-effectiveness. According to official data, the APUS-xDAN Large Model 4.0 has reached 90% of the comprehensive performance of the world-leading GPT-4, demonstrating remarkable computational efficiency and application potential.
This open-source initiative holds significant importance for fostering innovation and development in China’s AI sector. It lowers the barrier for the application of large-scale models, enabling more企业和 individuals to conduct complex AI computations using affordable hardware, thereby facilitating the broader adoption of AI technology across various industries. Furthermore, the partnership between APUS and Xin旦 Intelligence showcases China’s technological prowess and commitment to open collaboration in AI, potentially sparking a new wave of AI research and development globally.
APUS founder and CEO, Li Tao, commented, “By open-sourcing the APUS-xDAN Large Model 4.0 (MoE), we aim to ignite innovation among developers and enterprises, collectively advancing AI technology’s accessibility and progress. We believe that such open cooperation will bring immense value to the entire industry.” Xin旦 Intelligence echoed this sentiment, stating that open-sourcing the trillion-scale MoE architecture large model is a critical embodiment of the company’s commitment to technological innovation and open sharing, and they look forward to exploring AI’s endless possibilities with industry peers.
This joint open-source endeavor by APUS and Xin旦 Intelligence undoubtedly paves a new path for AI research both in China and globally, injecting fresh momentum into the widespread application and upgrading of AI technologies.
【来源】https://mp.weixin.qq.com/s/x_XSfcDBL7A4zb43vDFSMg
Views: 1