越狱PS5可原生运行大语言模型 PS5LM项目开源发布

PS5LM是一个新发布的开源项目，使用户能够在越狱的PlayStation 5上通过GPU原生运行GGUF格式的大语言模型1。该应用利用Vulkan后端，充分发挥主机GPU的11.4GB共享内存用于模型加载和上下文处理1。

项目支持PlayStation 5固件版本7.00至13.60的越狱设备1。PS5LM提供了与OpenAI和Anthropic API兼容的接口，允许网络上的其他设备连接到主机运行的模型服务1。用户可加载任何GGUF格式的模型，支持的模型包括Qwen 3.5 9B、d1 Omni 600M、d1 3B和Clef Flash等1。此外，应用还支持图像识别功能，兼容Gemma 4、MiniCPM-V和d1等视觉模型1。该项目采用GPL-3.0或更高版本开源许可证发布1，为用户提供了在本地进行私密部署的途径，无需依赖PC、云服务或账户1。

PS5LM, an open-source initiative, enables users to run large language models in GGUF format directly on a jailbroken PlayStation 5's GPU through a Vulkan backend 1. The application supports loading any GGUF model and provides interfaces compatible with both OpenAI's /v1/chat/completions and Anthropic's /v1/messages APIs, allowing devices on a network to connect to the host model service 1. The project requires a jailbroken PS5 running firmware versions 7.00 through 13.60 and leverages the console's GPU, which features 11.4 GB of shared memory for models and context 1.

The PS5LM project supports a range of language models, including Qwen 3.5 9B, d1 Omni 600M, d1 3B, and Clef Flash, as well as image recognition capabilities through models such as Gemma 4, MiniCPM-V, and d1 1. The application enables local, private deployment without requiring a PC, cloud services, or user accounts 1. The project is distributed under the GPL-3.0 or later open-source license 1.