初创公司 Underdog 近日推出了一款名为“Woof”的语音生成模型,其文件大小仅为 32MB,拥有 3100 万参数 1。该模型已对所有人开放下载,旨在推动本地化及注重隐私保护的 AI 体验 1。在技术实现上,“Woof”被设计为可在浏览器中运行,无需互联网连接即可在大多数设备上工作 1。
为了展示该模型的实际应用效果,Underdog 配套发布了一款复古风格的互动演示游戏 1。这款游戏包含打字机交互功能,能够实时朗读用户输入的内容 1。创始人 Sigil Wen 表示,公司的目标是让用户拥有自己的数据,并计划在数月内让现有设备能够运行最强大的模型 1。他在引用宣言时强调:“We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI.” 1。
此次发布的“Woof”并非 Underdog 的首个轻量化尝试。该公司此前还推出了大小为 167MB 的语音转文本模型“Bark”,以及一款拥有 270 亿参数的笔记本端模型 1。
Startup Underdog has released "Woof," a speech generation model weighing just 32 megabytes, accompanied by a retro-style interactive game that showcases the technology's capabilities 1. The model contains 31 million parameters and is available for open download to anyone interested in utilizing it 1. Its companion demo features a typewriter interface that reads user inputs aloud in real time 1.
Designed to run directly within web browsers, Woof operates offline on most devices, eliminating the need for an internet connection during use 1. Founder Sigil Wen stated that the company’s objective is to enable users to retain ownership of their data and to make the most powerful models compatible with existing hardware within months 1. Highlighting this vision, Wen cited a manifesto declaring, “We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI” 1. This release follows Underdog’s previous innovations, which include the 167-megabyte Bark speech-to-text model and a 27-billion-parameter model tailored for laptops 1.
评论
还没有评论,欢迎留下第一条。