Super nice story on the matmul optimization that gave 810 gflops for 512x512. Thanks for the write up and the contributions to llama.cpp and the community more

Super nice story on the matmul optimization that gave 810 gflops for 512x512. Thanks for the write up and the contributions to llama.cpp and the community more broadly.