阿联酋 Falcons AI 的 NSFW 分类模型跻身全球顶级开源模型(2025)
英文标题:🔶 UAE-based Falcon AI NSFW classifier among top global open-source models (2025)
总部位于阿联酋哈伊马角的 Falcons AI,其用于识别 NSFW 图像的微调 Vision Transformer(ViT)模型在截至 2025 年 2 月 14 日的 28 天内获得逾 5080 万次 Hugging Face 下载。该模型在平台 1400 多个模型中按月下载量排名第六,下载量约为同期 DeepSeek 全部模型合计的 12 至 13 倍。数据表明,面向内容审核的开源 AI 工具正拥有强劲的全球需求,也反映出阿联酋 AI 研发与开发者生态在开源社区中的影响力正在提升。对开发者和企业而言,这类以 Apache 2.0 许可证发布的模型可为平台治理、媒体发布和合规场景提供可直接部署的图像审核能力。
5.2 KB
1.6 KB
52.2 KB
963 B
18.2 KB
134.0 KB
中文
**#阿联酋 #开源——**总部位于哈伊马角的 AI 公司 Falcons AI 开发的一款 AI 模型,在过去 28 天内于全球领先开源 AI 平台 Hugging Face 的下载量已超过 5080 万次。这款经过微调的 Vision Transformer(ViT)模型专为 NSFW(Not Safe for Work,不适合工作场所观看)图像分类而设计,目前在该平台托管的 1400 多个 AI 模型中下载量排名第六。相比之下,近期备受关注的 AI 模型 Deepseek 在同期录得 370 万次下载。Falcons AI 的成功凸显了总部位于阿联酋的 AI 研究在全球开源社区中日益增长的影响力。
**这意味着什么?——**Falcons AI 开源模型下载量的迅速增长,既表明社区对该开发者的信任,也反映出全球对强大 AI 内容审核工具不断增长的需求。在线平台和企业环境中的内容审核是一个规模庞大且持续增长的应用场景,涵盖企业治理、媒体出版与广播,以及监管应用等多个领域。新一波生成式 AI 图像模型只会进一步推动数字平台上用户生成内容的激增。Falcons AI 的 NSFW 模型切中了市场需求,显然吸引了广泛的全球受众。
以下是关于 Falcons AI 及其 NSFW AI 模型的一些关键细节:
-
由总部位于阿联酋的 Falcons AI 开发的一款 NSFW(不适合工作场所观看)分类模型,在过去 28 天内的下载次数超过 5080 万次(截至 2025 年 2 月 14 日),使其成为全球下载量最高的开源 AI 模型之一。
-
Falcons AI 总部位于哈伊马角,为企业和政府提供 AI 驱动的解决方案,重点聚焦自动化、效率提升和决策支持。该公司还拥有先进的开源模型计划。
-
该 NSFW 模型按过去一个月下载量计算,在 Hugging Face 的 1400 多个模型中排名第六。该模型基于常用于自然语言处理的 Transformer 架构,并将其适配用于图像分类。
-
作为对比,尽管中国模型获得了大规模的全球关注,Falcons AI 模型在同期通过 Hugging Face 开源社区获得的下载量,仍是全部 Deepseek AI 模型合计下载量的 12 至 13 倍。
-
Falcons AI 面向 NSFW 图像的微调 Vision Transformer(ViT)使用 8 万张经过筛选的图像训练,以区分正常内容和露骨内容;模型采用 5e-5 的学习率和 16 的批次大小,以实现最佳性能。
-
该 NSFW 图像分类模型可依据 Apache 2.0 开源许可证下载和使用。
**放眼全局——**阿联酋对其 AI 生态系统的战略投资正在取得成效:来自研发实验室、商业 AI 创业公司、大型企业和独立专家的成果,正开始在全球舞台上产生影响。除阿联酋大型研发实验室备受关注的模型发布,以及 Falcons AI NSFW 模型这样的亮眼案例外,阿联酋生态系统也正因不断增长的开发者资源、社区与全球协作而赢得声誉。
阅读更多关于阿联酋开源模型的内容:
English / 原文
Screenshot of NSFW model (Falcons AI)
#UAE #opensource - An AI model developed by Ras Al Khaimah-based AI company Falcons AI has surpassed 50.8 million downloads in the past 28 days on Hugging Face, a leading global open-source AI platform. The Fine-Tuned Vision Transformer (ViT) model, designed to classify NSFW (Not Safe for Work) images, is currently the sixth most downloaded AI model out of over 1,400 models hosted on the platform. In comparison, Deepseek, one of the most talked-about AI models recently, recorded 3.7 million downloads in the same period. Falcons AI's success highlights the growing influence of UAE-based AI research in the global open-source community.
SO WHAT? - The rapid increase in downloads of Falcons AI’s open-source model underscores both community trust in the developer and the increasing global demand for robust AI-powered content moderation tools. Content moderation for online platforms and corporate environments is a big and growing use case which applies to a variety of domains from corporate governance, to media publishing and broadcasting, to regulatory usage. The new wave of Generative AI image models is only adding to the proliferation of user-generated content across digital platforms. Falcons AI has hit a sweet spot with its NSFW model, clearly appealing to a broad global audience.
Here are some key details about Falcons AI and the NSFW AI model:
-
A NSFW (not safe for work) classification model developed by UAE-based Falcons AI has been downloaded over 50.8 million times in the past 28 days (as of 14-Feb-25) , making it one of the world’s most downloaded open-source AI models.
-
Headquartered in Ras Al Khaimah, Falcons AI provides AI-powered solutions for businesses and governments, focusing on automation, efficiency, and decision-making support. It also has an advanced open-source model programme.
-
Ranked sixth (out of more than 1,400 models) on Hugging Face in number of downloads over the past month, the NSFW model is based on the transformer architecture commonly used in natural language processing but adapted for image classification.
-
By way of comparison, the Falcons AI model has registered 12-13 times more downloads via the Hugging Face open-source community than all Deepseek AI models combine over the same period, despite the Chinese model’s massive global publicity.
-
The Falcons AI Fine-Tuned Vision Transformer (ViT) for NSFW images was trained on 80,000 curated images to distinguish between normal and explicit content, using a learning rate of 5e-5 and a batch size of 16 for optimal performance.
-
The NSFW Image Classification Model is available for download and use under an Apache 2.0 open-source licence.
ZOOM OUT - The UAE’s strategic investment in its AI ecosystem is paying off, with development from R&D labs, commercial AI ventures, corporates and independent experts starting to make an impact on the global stage. In addition to well publicised model releases from big UAE R&D labs and shining stars like the Falcons AI NSFW model, the UAE ecosystem is also earning a reputation for growing developer resources, communities and global collaboration.
Read more about open-source models from the UAE: