If you like what you got there, use qwen2.5-coder
You will not believe the results, i did it and I did not believe the results. I have a preference for the 7B but the 14B also does really well, smaller than 7B needs a bit more encouragement or a more focused reasoning data set than what I used.