【全球科技讯】今天凌晨，科技巨头马斯克旗下的大模型公司 xAI 带来了震撼业界的公告，正式开源其拥有3140亿参数的混合专家（MoE）模型——Grok-1。这一壮举使得Grok-1一举成为目前公开参数量最大的语言模型，开启了人工智能领域的全新篇章。
据机器之心报道，Grok-1的基础训练基于海量文本数据，但未针对特定任务进行微调，展现出强大的泛化能力。模型设计独特，每个token上的激活权重约为25%，这在提高效率的同时，也保证了模型的精准度。值得注意的是，Grok-1是在2023年10月采用JAX库和Rust语言构建的自定义训练堆栈从零开始训练的，展现了xAI在技术创新上的决心和实力。
xAI秉持开放精神，遵循Apache 2.0许可证，将Grok-1的权重和网络架构全面开源，这一举措无疑将极大地推动人工智能研究和应用的发展，为全球科研人员提供宝贵的资源和学习平台。此次开源行动，预示着人工智能技术的普惠性将更进一步，全球开发者将有机会共同探索和利用Grok-1的潜力，加速AI技术在各个领域的落地应用。
英语如下：
**News Title:** “Musk’s xAI Stuns with Open-Source Release: Grok-1, a 3140 Billion-Parameter Giant, Leads the New Era of Large Language Models”
**Keywords:** Elon Musk, Grok-1, Open-Source Large Model
**News Content:**
**Global Tech Update** – Early this morning, tech giant Elon Musk’s big model company xAI made a groundbreaking announcement in the industry, officially open-sourcing its massive Mixed-Expert (MoE) model, Grok-1, with an astounding 3140 billion parameters. This move has instantly positioned Grok-1 as the largest publicly disclosed parameter count language model, ushering in a new chapter in the realm of artificial intelligence.
According to Machine Mind, Grok-1’s foundational training is based on extensive text data, without fine-tuning for specific tasks, demonstrating exceptional generalization capabilities. The model’s unique design features approximately 25% activation weights on each token, striking a balance between efficiency and precision. Notably, Grok-1 was trained from scratch in October 2023 using a custom training stack built with the JAX library and Rust programming language, showcasing xAI’s commitment to innovation and technological prowess.
Adhering to an open spirit, xAI has released Grok-1’s weights and network architecture under the Apache 2.0 license. This move is set to significantly advance AI research and application development, providing invaluable resources and a learning platform for researchers worldwide. This open-source initiative foreshadows a more inclusive future for AI technology, enabling global developers to collectively explore and harness Grok-1’s potential, accelerating the adoption of AI across various sectors.
【来源】https://mp.weixin.qq.com/s/hvt5zwoazDx26KOaKuTs_w
Views: 1