1 post
MicroLLM Lab runs seven 25M-360M parameter language models in the browser via WebGPU, on-device with no server, and benchmarks their speed and accuracy.