llama2.cpp

Forked from Karpathy repo llama2.c

I implementled a modification over llama2.c to use Eigen for fast matrix multiplication. I did it trying to minimze the modifications over the original C version.

Eigen requires C++. Then to compile use:

g++ -I. -Ofast -fopenmp run.cpp  -lm  -o run

This version is not faster in my old computer but using Eigen you can use AVX2 (among others) set of instructions that could speedup the inference.

License

MIT