Old equipment running large language model technology breakthroughs.

The smolmaxxing project enables large language models to run on old devices like Xiaomi TV boxes and iPad Air by customizing the kernel, optimizing memory usage, and streaming weights from flash storage. The technology is not limited to specific devices: 1-bit kernels require only basic NEON support, and flash storage streaming applies to any Linux or Android device as long as the device memory is smaller than the model size. The project requires using Hugging Face's pre-built model files and needs to be paired with a patched llama.cpp library. On Xiaomi TV boxes, the Qwen3.5-2B model processes 7.7 tokens per second, and the Ling-mini-2.0 model processes 5.1 tokens per second. The project emphasizes that all performance data is measured on the device.

AI-written from the reports · Updated yesterday