What is an acceptable token speed (t/s) for your daily local LLM use?

~5 t/s (Very slow - unusable for longer paragraph generation)

~10 t/s (Still slow - typical for CPU-only inference)

~20 t/s (Mostly usable for light chat or short completions)

~40 t/s (Much faster than the average reading speed)

~60 t/s (Smooth for most common tasks)

~100 t/s (Very fast, comfortable for long code snippet generation)

>100 t/s (Feels near-instant when it comes to short paragraphs)