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，这一系列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，其中，令人惊叹的是2B版本竟然可以在普通的笔记本电脑上流畅运行，这无疑极大地降低了高效能AI模型的使用门槛。
据透露，Gemma模型的研发融合了Google DeepMind和谷歌其他团队的智慧，采用了与 Gemini 模型相同的研究和技术，旨在推动负责任的AI开发。谷歌表示，Gemma在语言理解、推理、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项成绩超过了参数量更大的Meta Llama-2，彰显了其在轻量化与性能优化方面的卓越能力。
这一开源模型的发布，不仅展示了谷歌在AI领域的领先地位，也为全球开发者提供了一个更强大、更易获取的工具，有望推动AI技术的广泛应用和创新。谷歌的这一举动无疑将进一步激发社区的创新活力，加速AI技术在各个领域的落地和进步。
英语如下：
News Title: “Google Unveils Lightweight Giant Gemma: Outperforming Meta’s Llama-2, Setting New Standards in the Open-Source Realm”
Keywords: Google release, Gemma model, Performance superiority
News Content: Google recently announced a major technological innovation with the launch of the Gemma series of open-source large models, outperforming several competitors, including Meta’s Llama-2, and establishing new industry benchmarks. The Gemma models come in two versions: a 2-billion-parameter 2B version and a 7-billion-parameter 7B version. Remarkably, the 2B version can run smoothly on a standard laptop, significantly lowering the barrier to entry for high-performance AI models.
Sources reveal that the development of the Gemma model drew on the expertise of both Google DeepMind and other Google teams, leveraging the same research and technology as the Gemini model, with a focus on promoting responsible AI development. Google claims that Gemma excels in key benchmark tests for language understanding, reasoning, and mathematics, surpassing Llama-2 in 11 out of 18 tests, demonstrating its exceptional capability in lightweight design and performance optimization.
This open-source model release not only underscores Google’s leadership in the AI field but also offers global developers a more powerful and accessible tool, potentially fueling the widespread adoption and innovation of AI technology. Google’s move is expected to invigorate the community’s innovative spirit, accelerating the integration and advancement of AI across various sectors.
【来源】https://www.tmtpost.com/6944205.html
Views: 1