Edit: Running the DeepSeek-R1-Distill-Llama-8B-Q8_0 gives me about 3t/s and destroys my system performance on the base m4 mini. Trying the Q4_K_M model next.