谷歌发布开源大模型Gemma,性能超越竞品
科技巨头谷歌宣布推出全球性能最强大、轻量级的开源模型系列Gemma,为负责任的AI开发提供支持。
Gemma分为2B(20亿参数)和7B(70亿参数)两种尺寸版本,其中2B版本甚至可以在笔记本电脑上直接运行。谷歌表示,Gemma采用与Gemini模型相同的研究和技术,由Google DeepMind和谷歌其他团队共同开发。
在18个语言理解、推理、数学等关键基准测试中,Gemma在11项测试中超越了Meta Llama-2等更大参数的开源模型。这表明Gemma在自然语言处理、推理和数学问题解决方面具有出色的性能。
谷歌强调,Gemma的开发旨在促进负责任的AI开发。该模型经过精心设计,以最大限度地减少偏见和有害输出的风险。此外,Gemma作为开源模型发布,允许研究人员和开发人员对其进行进一步研究和改进。
Gemma的发布标志着谷歌在开源大模型领域取得了重大进展。这些模型有望为各种应用提供支持,包括自然语言处理、计算机视觉和机器学习。
谷歌表示,他们将继续开发和发布开源大模型,以推动AI技术的发展和负责任的使用。
英语如下:
**Headline: Google’s Open-Source Large Model Gemma Debuts, Outperforming Competitors**
**Keywords:** Open-source model, Outperforms, Runs on a laptop
**Body:**
Google has announced the release of Gemma,a family of open-source large models that are the most powerful and lightweight of their kind, supporting responsible AI development.
Available in 2B (20 billion parameters) and 7B (70 billion parameters) sizes, the 2B version of Gemma can even run on a laptop. Google saysthat Gemma builds on the same research and techniques as its Gemini model, developed by Google DeepMind and other teams across Google.
Across 18 key benchmarks for language understanding, reasoning, math, and more, Gemma outperforms larger parameter open-source models like Meta’s LLaMA-2 on 11 of them. This shows that Gemma has strong performance in natural language processing, reasoning, and solving math problems.
Google emphasizes that Gemma was developed with responsible AI development in mind. The model has been designed to minimize the risk of bias and harmful outputs. Additionally, as an open-source model, Gemma allows researchersand developers to further study and improve it.
The release of Gemma marks a significant step forward for Google in the area of open-source large models. These models promise to power a wide range of applications, including natural language processing, computer vision, and machine learning.
Google says that it will continue to develop and release open-source large models to advance the development and responsible use of AI technology.
【来源】https://www.tmtpost.com/6944205.html
Views: 1