Google 发布了基于 Gemini 的开源 AI 模型 Gemma。这些模型基于与 Google 闭源的 Gemini 模型相同的研究和技术，但 Gemma 模型更小。Google 公开了两个不同规模的 AI 模型，分别是 Gemma 2B 和 Gemma 7B，它们分别具有 20 亿和 70 亿个参数。这两个模型在全球范围内可以在 Kaggle、Hugging Face、Nvidia NeMo 和 Google Vertex AI 等网站上获取。
与 Google Gemini 相比，Gemma 模型的功能较为简化，但据 Google 称，这两个模型在某些 AI 基准测试中仍然表现优于更大的模型。据该科技巨头称，Gemma 可以在笔记本电脑、台式机或云端上本地运行。该公司将 Gemma 7B 与 Meta 的 Llama-2 7B 和 13B 模型在八个不同的基准测试中进行了比较，根据 Google 的数据，Gemma 在所有测试中表现更好。Google 已发布了一份技术报告，其中详细介绍了 Gemma 及其性能。
两个 Gemma 模型可以用于商业目的，但禁止将其用于某些用途，例如生成有关儿童虐待、虚假信息或煽动暴力的内容。Gemma 还配备了 “负责任的 AI 工具包”，包括安全分类、调试工具和构建 AI 应用的最佳实践。
Gemma 可用于开发 AI 应用程序。开发人员可以添加自己的指南或禁用词列表。据 Google 称，目前 Gemma 在与语言相关的任务中表现最佳，尤其是英语任务。Google 的更高级 Gemini 模型也可供开发人员使用，但 Gemini 模型更为封闭，开发人员只能通过 API 或 Google 的 Vertex AI 平台使用。而 Gemma 则可以自由使用。
Google has released an open-source AI model called Gemma, based on Gemini. These models are built on the same research and technology as Google’s closed-source Gemini model, but Gemma is smaller in size. Google has made available two AI models of different scales, namely Gemma 2B and Gemma 7B, with 2 billion and 7 billion parameters, respectively. Both models can be accessed worldwide on platforms such as Kaggle, Hugging Face, Nvidia NeMo, and Google Vertex AI.
While the Gemma models are less advanced than Google Gemini, Google claims that they still outperform larger models in certain AI benchmarks. According to the tech giant, Gemma can run locally on laptops, desktops, or in the cloud. The company compared Gemma 7B to Meta’s Llama-2 7B and 13B models in eight different benchmarks, and according to Google’s data, Gemma performed better in all tests. Google has published a technical report providing detailed information about Gemma and its performance.
Both Gemma models can be used for commercial purposes, but there are restrictions on their usage, such as generating content related to child abuse, misinformation, or inciting violence. Gemma comes with responsible AI toolkits that include safety classifications, debugging tools, and best practices for building AI applications.
Gemma can be utilized for developing AI applications, and developers can add their own guidelines or lists of prohibited words. According to Google, Gemma currently performs best in language-related tasks, particularly in English. Google’s more advanced Gemini model is also available to developers, but it is more closed-off, and developers can only use it through APIs or Google’s Vertex AI platform. Gemma, on the other hand, can be freely deployed.
微信扫一扫