117 words
numpy is fast, really fast. All that is thanks to it’s written in C under the hood. But probably you knew that. Thing you might didn’t know is that it’s also using SIMD.
SIMD - (S I M D) are special processor instructions added during ~2011. They operate on multiple variables at once. Instead of doing a sum of array one by one, SIMD enables us to do 4 items plus 4 items at the same time. Instructions are called AVX2. If you tried running local llm on older machine this is why you couldn’t. Because it doesn’t have AVX2 which is improtrant for machine learning tasks.