So when I use safetensors files am I basically running the model at FP16? Do I need to find GGUF files if i want to use say Q8/Q6?