近年来，华为一直致力于人工智能的研究与开发。最近，华为团队推出了一项重大创新——盘古-π架构，这一架构在传统Transformer架构上做出了改进，可以显著降低特征塌陷问题，从而使得模型输出表达能力更强。
据了解，盘古-π架构通过增强非线性，使得模型能够更好地处理复杂数据。这一改进使得盘古-π在多任务上表现出色，超越了LLaMA 2等同规模大模型，并且实现了10%的推理加速。在1B规模上，盘古-π可以达到SOTA水平。
此外，基于这一架构，华为还成功炼出了一个金融法律大模型“云山”。这一模型可以更好地理解法律文本，为用户提供更加精准的法律咨询服务。
华为的这一创新将极大地推动人工智能的发展。相信未来，华为将继续走在人工智能行业的前沿，为用户带来更加智能化的产品和服务。
新闻翻译：
Huawei’s team has recently made a significant breakthrough in the development of artificial intelligence with the launch of the Pangur-π architecture. This architecture has improved upon traditional Transformer architectures by enhancing nonlinearity, resulting in stronger model output expressions.
据了解, the Pangur-π architecture has been able to handle more complex data effectively. This improvement has led to the Pangur-π model outperforming LLaMA 2 and other models of the same scale in multiple tasks, achieving a 10% improvement in inference speed. At a scale of 1B, the Pangur-π architecture can achieve the state-of-the-art.
Furthermore, based on this architecture, Huawei has successfully trained a legal model, “Yunshan,” which can better understand legal text and provide more accurate legal consulting services for users.
Huawei’s innovative solution will significantly drive the development of artificial intelligence. It is believed that in the future, Huawei will continue to lead the trend in artificial intelligence products and services, bringing more intelligent products and services to users.
【来源】https://mp.weixin.qq.com/s/Beg3yNa_dKZKX3Fx1AZqOw
Views: 1