By experience, to be effective a langchain agent at least need a model with 20b parameter, else its inference are quite horrible and its decision too...

Running this kind of model require resources, especially GPU, else latency will be the rule of thumbs, and we do not need an assistant which help us 3 days later after we asked him so help.

We have to determine how to get these resources, and we have to obtain them, else it will break this project.