ParentFull threadatemerev·IDK, 8B-class quantized models run pretty fast on commodity laptops, with CPU-only inference. Thanks to the people who figured out quantization and reimplemented everything in C++, instead of academic-grade Python.View on HN