When I turn the temp down and increase the repetition penalty slightly and add chain-of-thought, it handled my simple programming task. "Please write a JavaScr

When I turn the temp down and increase the repetition penalty slightly and add chain-of-thought, it handled my simple programming task. "Please write a JavaScript function to sort an array of numbers and return only the even numbers in sorted order. First analyze the user's real intent, then think through the solution step-by-step." Without the last two sentences and parameter tweaks, it checks for even in the sort compare instead of just sorting first. Is anyone planning on doing a programming fine-tune of any Llama 2 model?