You can now run a 90M conversational LLM on the Sony PSP (hardware from 2004). Doesn't get more local than this.

A developer has successfully run a 90M conversational LLM on a Sony PSP, hardware from 2004. This achievement pushes the limits of local AI, demonstrating that the PSP can handle such a model, albeit with slow inference speeds of 0.5-0.6 tokens per second, leading to 1-3 minute reply times. While not practically useful, the 90M model can generate basic content like poems and short stories, and sometimes provides correct answers to simple questions, showcasing its capabilities on vintage hardware.