programming models

[Sankyuubigan]

I would like to make changes to my python projects. my projects have about 100 thousand tokens. i would like to have the neural network test itself, at least read the logs after startup, so that i don't have to manually do basic routine things.

I would like to know what benchmarks you use to determine the best local models for coding.

As I understand, you need to have several models, for different types of tasks. but then you need benchmarks to understand which model to use at the moment, today, and always have an updated list of current ones for today.

As I understand it, there are models that are better at creating a task plan with promt (so you don't get stuck with endless code editing).

And other llm can execute those changes according to that plan. (models that have a large context window).

as I understand “devstral small 2505 gguf q6” is quite a good option at the moment?