2 ms·

So how is this compared to just running "small" models locally via Ollama or other free inference engines?