I like the latest qwen2.5 (https://nexaai.com/Qwen/Qwen2.5-0.5B-Instruct/gguf-q4_0/read...). It was just released last week. It is one of the best small langauge models right now according to benchmarks. And it is small and fast!