llama.go
Meta's LLaMA large language model inference in pure Golang using only CPU. No GPU needed.
It will stress all CPU cores using FP32 math - so you'll need at least 32Gb RAM for 7B model.
AVX2/AVX-512 and ARM NEON optimizations will come later. More details come a bit later too...
