Small parameters do not have the biases of specific built models. Polish interpretation of it with current tools , maybe 🤔. It involves some work via prompting
Discussion
I’ve used 3b parameters to dissect entire case law studies and it worked fine