- [] update prompt given to the model (before the actual question/answer) to match prompt used in training
- [] figure out how to merge lora adapter and classifier head with base model
- [] add new examples
- [] update interface so that one can select types of problems from the examples