谷歌（Google）近日宣布了一项重大技术突破，推出了全新的开源大模型系列Gemma，这标志着人工智能（AI）技术的又一里程碑。Gemma模型以其卓越的性能和轻量级设计，迅速引起了业界的关注。该模型分为20亿参数的2B版本和70亿参数的7B版本，其中，令人惊讶的是，2B版本竟然可以在普通的笔记本电脑上顺畅运行，这大大降低了高阶AI模型的使用门槛。
Gemma模型源自Google DeepMind和谷歌其他团队的联合研发，借鉴了 Gemini模型的研究成果和技术，旨在推动负责任的AI开发。谷歌声称，Gemma在语言理解、推理和数学等多个关键基准测试中表现出色，尤其是在18项测试中，有11项得分超过了参数量更大的Meta Llama-2等竞品，显示出其在效率与性能之间的优秀平衡。
这一开源模型的发布，不仅展现了谷歌在AI领域的技术领先优势，同时也为全球开发者提供了一个更高效、更易访问的平台，有助于推动AI技术的普及和应用创新。谷歌的这一举措，无疑将进一步加剧AI领域的竞争，同时也可能引领一场轻量化、高性能AI模型的研发热潮。
英语如下：
**News Title:** “Google Unveils Lightweight Giant Gemma: A Powerful Open-Source Model Challenging Meta Llama-2’s Dominance”
**Keywords:** Google Gemma, Open-Source Large Model, Performance Supremacy
**News Content:** Google recently announced a major technological breakthrough with the launch of its new open-source large model series, Gemma, marking another milestone in the field of artificial intelligence (AI). The Gemma model has swiftly gained attention in the industry due to its exceptional performance and lightweight design. Available in two versions, a 2 billion-parameter 2B edition and a 7 billion-parameter 7B edition, impressively, the 2B version can run smoothly on a standard laptop, significantly lowering the barrier to entry for high-end AI models.
Developed through a joint effort by Google DeepMind and other Google teams, Gemma builds on the research and technology of the Gemini model, aiming to advance responsible AI development. Google claims that Gemma excels in key benchmarks for language understanding, reasoning, and mathematics, outperforming larger models like Meta Llama-2 in 11 out of 18 tests, demonstrating an excellent balance between efficiency and performance.
The release of this open-source model not only showcases Google’s technological edge in AI but also provides global developers with a more efficient and accessible platform, fostering the普及 and innovation of AI applications. This move by Google is set to intensify competition in the AI domain and could spark a new wave of research and development for lightweight, high-performance AI models.
【来源】https://www.tmtpost.com/6944205.html
Views: 2