LLaMA Now Goes Faster on CPUs

LLaMa Now Goes Faster on CPUs: New matrix multiplication kernels have been developed for llamafile, resulting in improved performance for prompt evaluation time. The improvements are most noticeable for certain weights and specific CPU types such as ARMv8.2+, Intel Alderlake, and AVX512. The new kernels outperform MKL for matrices that fit in L2 cache and offer potential for faster evaluation speed.