OpenJev

OpenJev是一个本地运行的AI模型实验工具，允许用户在自己的GPU上直接读取模型选择概率，或通过生成JSON的方式获取选项分布1。该工具的核心特性是所有输入数据均保留在本地浏览器中，不向外部服务器传输1。

工具默认采用MiniCPM 2B模型，用户可在低端设备上切换至Qwen3 0.6B，同时支持更大的4B模型选项1。模型权重来自Hugging Face，并缓存在浏览器本地1。OpenJev采用GGUF量化权重通过wllama实现运行1。首次加载时可能需要等待数分钟1。

OpenJev is an experimental tool for running AI models locally, enabling users to read model selection probabilities directly on their own GPU or obtain option distributions through JSON generation 1. The tool operates entirely within the browser, ensuring that all input data remains on the user's device without leaving the local environment 1.

The platform defaults to the MiniCPM 2B model but offers flexibility for different hardware configurations 1. Users with lower-end devices can switch to the Qwen3 0.6B model, while a larger 4B model option is also available 1. Model weights are sourced from Hugging Face and cached locally in the browser 1. The implementation utilizes GGUF quantized weights through wllama to achieve this local execution capability 1. Initial loading may require several minutes on first use 1.