Developer Runs AI Language Model on 2007 Sony PSP With Just 64MB of RAM
A developer successfully ran a working language model on a Sony PSP-2000, a 2007 handheld device with a 333 MHz processor and 64 MB of RAM, generating English text at one to two tokens per second. The model used was Karpathy's stories15M, compressed via int8 quantization to roughly 17 MB and implemented in about 1,100 lines of pure C code. The experiment highlights a broader trend of shrinking hardware requirements for useful AI models, with separate projects demonstrating a 90M-parameter tool-calling model running on smartphones and Raspberry Pi devices. Advances in quantization techniques, such as Unsloth Dynamic 3.0 GGUFs, are also enabling models like Qwen3.8-27B to run at 6.2 GB — 89% smaller than the original — while retaining around 72% accuracy. Practical use cases for these tiny local models include request routing, classification, and constrained code completion, where they can reduce reliance on costly frontier AI APIs.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in