Comment by LorenDB

You might also evaluate Gemma 2B and Qwen 0.5B as alternative tiny models, FWIW.