Recently I was scrolling through HF to try a very small model. Fired up Qwen 0.5 B and surprisingly for my purposes it did between that even Lllama 2 7B. That was very surprising to me.