PS5LM runs Qwen 3.8 on a jailbroken PS5 at ~20 tok/s via llama.cpp and Vulkan
lxfater · x · 2026-10-15
PS5LM is an open-source project porting llama.cpp to a jailbroken PlayStation 5, running GGUF models natively on the console's GPU via Vulkan. The quantized Qwen3.8-27B generates 20 tok/s using 11.44 GiB of device memory with 64k context. It ships a TV dashboard showing VRAM/power/thermals, a model store, USB model loading, and an OpenAI-compatible API so any device on the LAN can chat via browser, with inference fully on-device. This fork (PS5LM Plus) adds the native GPU app on top of Mert Cobanov's original payload.
Related event: Jailbroken PS5 runs Qwen 27B locally at 20 tok/s via llama.cpp port(11 posts)→
More from Fun
- 'You need high pain tolerance to review Gemini 4,' researcher quips amid error screens — gleech · 2026-10-15
- Claude's safeguards triggered while animating Gwern's AI takeover story — burny_tech · 2026-10-15
- User lets Grok bot run 20 hours straight, burning the weekly limit on a demo — RachelVT42 · 2026-10-15
- Voxel pagoda world built in 33 minutes with Gemini agents and code-generated audio — aitrendz_xyz · 2026-10-15
- Engineers mock AI CAD hype: people who never designed a part preaching hardware's future — IgorCarron · 2026-10-15
- AI-Assisted Decomp Directory Shuts Down Amid Scoring Scandal; banteg Calls Repo Privatization a Mistake — banteg · 2026-10-15