Auto optimization challenge

[beartell]

With the cpu, gpu, ram, disk, operating system, cpu architecture in mind,
And also bear in mind the use case, scenario : rag, image generation, summarization, completion, etc

Can we calculate and optimize the best performing parameters of running llamafile and llamafile(s) on a system ?
Context length, ngl, dynamic batching,etc...

Can we write an optimizer for that or we can continue our manual,heuristic efforts...