testingcatalog· @testingcatalog · X·· 19 小时前AI 评分
Atomic Agent 演示云端模型规划、本地模型执行 28 步任务
Testing Catalog 转发 Atomic 的演示:云端模型只做一次规划调用,本地模型负责 28 个任务步骤,本地模型运行在 Atomic 基于 llama.cpp 的分支上并使用 Google 的 TurboQuant,公司测试称 TurboQuant 可让长对话内存占用最多降低 80%。该应用支持 Apple Silicon Mac、Windows 和 Linux,云端模型与 Fusion 使用用户自己的 API key,可在 atomicagent.io 试用。
In Atomic's demo, a cloud model made one planning call, and local models handled 28 task steps. Local models run on Atomic's fork of llama.cpp with Google's TurboQuant, and the company reports that TurboQuant reduced memory use for long chats by up to 80% in their tests.
The app is available for Apple Silicon Macs, Windows, and Linux. Cloud models and Fusion use the user's own API key.
Test it out 👀
atomicagent.io
来源:testingcatalog · x.com