news pappernews papper

今日,全球瞩目的科技巨头埃隆·马斯克旗下公司 xAI 宣布,他们开发的一款名为 Grok-1 的人工智能模型正式开源。这款模型拥有高达 3140 亿参数,是当前参数量最大的开源大语言模型。这一消息无疑将在人工智能领域掀起一股开源热潮。

Grok-1 模型的发布标志着开源社区在人工智能领域的又一重要里程碑。该模型基于混合专家(MoE)架构,其核心技术在于能够在给定 token 上激活高达 25% 的权重,从而实现高效的学习和推理。此外,xAI 团队还透露,他们将于 2023 年 10 月使用 JAX 库和 Rust 语言的自定义训练堆栈从零开始重新训练 Grok-1。

在遵守 Apache 2.0 许可证的前提下,xAI 不仅开源了 Grok-1 的权重,还开放了其网络架构,这一举动无疑将极大地促进人工智能技术的共享和创新。开源社区成员将能够访问并研究 Grok-1 的源代码,这不仅有助于他们理解模型的内部工作原理,还能够在此基础上进行改进和优化。

马斯克一直以来都是开源运动的积极倡导者,他认为开放和共享是推动技术进步的关键。Grok-1 的开源不仅体现了马斯克的理念,也为全球的人工智能研究和应用带来了新的可能性。随着更多开发者和研究者的加入,我们有理由相信,这将加速人工智能技术的成熟和应用,并带来更多的社会和经济价值。

英文翻译内容:

Title: xAI Releases Grok-1 Model, Leading a New Era in Open Source Large-Language Models

Keywords: AI Model, Open Source Community, Technological Innovation

News content:
Today, Elon Musk’s company xAI announced the official open-sourcing of Grok-1, an AI model with 314 billion parameters, the largest currently available for open source. This announcement is set to ignite a new wave of enthusiasm within the open-source community.

Grok-1, built on a MoE architecture, excels in activating up to 25% of its weights per token, enabling efficient learning and inference. xAI also plans to retrain Grok-1 from scratch using their custom training stack based on JAX and Rust in October 2023.

By releasing Grok-1’s weights and network architecture under the Apache 2.0 license, xAI is not only sharing the model’s source code but also providing a platform for the community to understand and improve upon the technology. This move is expected to accelerate the maturation and application of AI technology, bringing with it significant social and economic benefits.

【来源】https://mp.weixin.qq.com/s/hvt5zwoazDx26KOaKuTs_w

Views: 1

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注