在人工智能领域，昆仑万维近期迈出了一大步，宣布开源其自主研发的2千亿稀疏大模型——天工MoE（Skywork-MoE）。这一消息不仅标志着昆仑万维在AI技术研发上的重大突破，也预示着开源领域将迎来一个重量级的里程碑。
天工MoE的推出，是昆仑万维在深度学习模型领域深入探索的结果。它基于昆仑万维先前开源的Skywork-13B模型的中间检查点进行扩展，首次将MoE Upcycling技术完整应用于开源千亿级MoE大模型之中。MoE Upcycling技术的创新应用，不仅显著提升了模型的性能，而且在降低推理成本方面取得了重大进展。这在当前AI领域中显得尤为珍贵，因为高性能与低成本一直是模型应用中难以兼得的两个目标。
值得注意的是，天工MoE模型的另一大亮点是其强大的兼容性与灵活性。模型支持单台4090服务器进行推理运算，这意味着无论是在大型数据中心还是个人工作站，开发者和研究者都能便捷地使用这一模型，大大降低了AI技术的入门门槛，促进了AI技术的普及与应用。
昆仑万维的这一举动，不仅展现了其在AI技术研发上的深厚积累和创新能力，也为开源社区贡献了宝贵资源。通过开源天工MoE，昆仑万维不仅推动了AI技术的共享与合作，还为全球开发者提供了更高效、更具成本效益的AI解决方案，有望在学术研究、商业应用等多个领域引发新一轮的创新浪潮。
来源：钛媒体
这一新闻事件的发布，不仅在人工智能领域内引起了广泛的关注，也向全球展示了中国企业在AI技术创新与开放合作方面的实力与决心。昆仑万维通过天工MoE的开源，不仅加速了AI技术的迭代与应用，也为未来的智能社会发展奠定了坚实的基础。
英语如下：
Headline: “Kunlun WebTech Opens Sources a 2 Trillion Sparse Large Model Tian Gong MoE, Efficient Inference with Single 4090 Server”
Keywords: Kunlun WebTech, Open-source Large Model, Skywork-MoE
Content: In the realm of artificial intelligence, Kunlun WebTech has taken a significant step forward by announcing the open-source release of its self-developed 2 trillion sparse large model, Tian Gong MoE (Skywork-MoE). This announcement marks a major breakthrough in AI research and development by Kunlun WebTech, signaling a significant milestone in the open-source domain.
The launch of Tian Gong MoE is a result of Kunlun WebTech’s deep exploration in deep learning models. It builds upon the previously open-sourced Skywork-13B model, expanding into the application of MoE Upcycling technology for the first time in an open-source billion-parameter MoE large model. The innovative application of MoE Upcycling technology not only significantly boosts the model’s performance but also makes substantial progress in reducing inference costs. This is particularly valuable in the current AI field, where high performance and low cost have long been a hard-to-achieve duo.
Notably, a standout feature of the Tian Gong MoE model is its robust compatibility and flexibility. The model supports inference computations with a single 4090 server, making it accessible for use in both large data centers and personal workstations. This significantly lowers the entry barrier for AI technology, promoting the widespread adoption and application of AI technology.
Kunlun WebTech’s move not only showcases its deep expertise and innovation in AI technology development but also contributes valuable resources to the open-source community. By open-sourcing Tian Gong MoE, Kunlun WebTech accelerates the sharing and collaboration of AI technology, providing global developers with more efficient and cost-effective AI solutions. This is expected to trigger innovation waves in various fields such as academic research and commercial applications.
Source: Tiemedia
The announcement of this news event has garnered widespread attention within the AI field and demonstrated the capabilities and determination of Chinese enterprises in AI technology innovation and open cooperation. Through the open-source release of Tian Gong MoE, Kunlun WebTech accelerates the iteration and application of AI technology, laying a solid foundation for the future development of intelligent society.
【来源】https://www.tmtpost.com/7114403.html
Views: 1