Benchmark fast-llama against llama.cpp and mistral.rs with different amount of cpu cores and different gpus.