This seems like something that can be tackled with a fine tune, or distillation. That said, you'll get far with injecting documentation of the framework into t

This seems like something that can be tackled with a fine tune, or distillation. That said, you'll get far with injecting documentation of the framework into the context to fill the gaps in pretraining data.