cobanov/PS5LM

What it solves

PS5LM turns a jailbroken PlayStation 5 into a private, local server for running Large Language Models (LLMs) and decision models. It allows users to offload AI inference from their PC or cloud services to the console's GPU, providing a dedicated hardware resource for AI tasks that can be accessed by other devices on the same network.

How it works

The project utilizes llama.cpp with a Vulkan backend to run GGUF-format models directly on the PS5's RDNA 2 GPU. It provides a local API server (compatible with OpenAI and Anthropic standards) that allows external devices like laptops or coding tools to send requests to the console. The app includes a graphical interface for the TV to manage model downloads, monitor hardware performance (temperature, power, load), and interact with models directly via the controller.

Who it’s for

This project is for users with a jailbroken PS5 (firmware 7.00 to 13.60) who want to run private AI models locally without relying on cloud accounts, external PCs, or internet connectivity.

Highlights

• Runs models directly on the PS5 GPU using the Vulkan backend.
• Provides OpenAI and Anthropic-compatible API endpoints for integration with external tools.
• Supports decision models for typed, high-speed inference tasks.
• Includes an on-screen dashboard for monitoring system performance and managing model storage.
• Enables image-to-text capabilities for multimodal models.