谷歌近日宣布了一项重大技术创新，推出了开源大模型系列Gemma，该模型在性能上超越了包括Meta的Llama-2在内的多个竞品，树立了新的行业标杆。Gemma模型分为20亿参数的2B版本和70亿参数的7B版本，其轻量级设计的2B版本竟然可以在普通的笔记本电脑上顺畅运行，这无疑大大降低了大规模语言模型的使用门槛。
据谷歌透露，Gemma模型借鉴了 Gemini 的研究和技术，由Google DeepMind与谷歌的其他团队共同研发，其核心目标是推动负责任的AI开发。谷歌表示，该模型在18项关键的语言理解、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，展现出卓越的性能和效率。
这一创新标志着谷歌在AI领域的又一重大突破，Gemma模型的开源特性将为全球的开发者和研究者提供更强大、更易获取的工具，有望推动AI技术的进一步普及和应用。谷歌的这一举措不仅强化了其在AI领域的领导地位，也为行业设立了新的性能和责任标准。
英语如下：
**News Title:** “Google Unveils Gemma, a Lightweight AI Model Outperforming Meta’s Llama-2 in Open Source Realm”
**Keywords:** Google Gemma, Open-Source Large Model, Outperforms Llama-2
**News Content:** Google recently announced a major technological innovation with the launch of its open-source large model series, Gemma, which surpasses several competitors, including Meta’s Llama-2, setting new industry standards. The Gemma model comes in two versions – a 2 billion parameter 2B edition and a 7 billion parameter 7B edition. Impressively, the lightweight 2B version can run smoothly on a standard laptop, significantly lowering the barrier to entry for using large language models.
According to Google, the Gemma model draws on the research and technology of Gemini, jointly developed by Google DeepMind and other Google teams, with a core focus on promoting responsible AI development. The company claims that in 18 key language understanding, reasoning, and math benchmark tests, Gemma outscored larger open-source models like Meta’s Llama-2 in 11, demonstrating exceptional performance and efficiency.
This innovation signifies another major breakthrough for Google in the AI domain. The open-source nature of the Gemma model will provide global developers and researchers with more powerful and accessible tools, potentially accelerating the adoption and application of AI technology. Not only does this move reinforce Google’s leadership in AI, but it also sets new benchmarks for performance and responsibility within the industry.
【来源】https://www.tmtpost.com/6944205.html
Views: 1