谷歌（Google）近日宣布推出其最新开源大模型系列Gemma，这一模型在性能上展现出显著优势，超越了包括Meta的Llama-2在内的多个竞品。Gemma模型分为两个版本，分别为20亿参数的2B版本和70亿参数的7B版本。值得注意的是，即便是2B版本，也能够在普通的笔记本电脑上流畅运行，这无疑极大地降低了AI模型应用的硬件门槛。
Gemma模型源于Google DeepMind与谷歌其他团队的联合研发，其核心技术与 Gemini 模型一脉相承，特别强调了对负责任的AI开发的重视。谷歌声称，Gemma在18个关键的语言理解、推理和数学基准测试中，有11项得分超过了参数量更大的Meta Llama-2等开源模型，显示出其在效率和性能上的优越性。
这一创新成果不仅标志着谷歌在AI领域的又一重大突破，同时也为全球开发者提供了一个强大且易于使用的工具，他们可以在自己的项目中利用Gemma，推动AI技术的广泛应用和持续发展。谷歌的这一开源举措，无疑将对AI研究和应用的生态产生积极影响，进一步促进技术的普惠与进步。
英语如下：
News Title: “Google Launches Gemma, an Open-Source Large Model Outperforming Meta’s Llama-2, Pioneering a New Era in Lightweight AI Innovation”
Keywords: Google Gemma, Open-source model, Performance surpassing
News Content: Google recently announced the release of its latest open-source large model series, Gemma, which demonstrates a significant performance advantage over competitors, including Meta’s Llama-2. The Gemma model is available in two versions, a 2 billion-parameter 2B version and a 7 billion-parameter 7B version. Remarkably, even the 2B version can run smoothly on a standard laptop, significantly lowering the hardware门槛 for AI model application.
Developed through a collaborative effort between Google DeepMind and other Google teams, Gemma is rooted in the same technological lineage as the Gemini model, with a particular emphasis on responsible AI development. Google claims that Gemma outperforms larger parameter-count open-source models like Meta’s Llama-2 in 11 of 18 key language understanding, reasoning, and mathematics benchmarks, showcasing its superiority in efficiency and performance.
This innovation not only signifies another major milestone for Google in the AI domain but also provides global developers with a powerful and user-friendly tool. They can now leverage Gemma in their projects, fostering the widespread adoption and continuous development of AI technology. Google’s open-source initiative is poised to have a positive impact on the AI research and application ecosystem, further promoting the democratization and advancement of technology.
【来源】https://www.tmtpost.com/6944205.html
Views: 1