Refactor prompt template formation

[vaibhavad]

Currently prompt template design is implemented here, which means the package needs to be edited for any new model or a different template.

It should be shifted to training script

[vaibhavad]

Can be done in this way

class CustomModel(LLM2Vec):
    def prepare_for_tokenization(self, text):
        text = (
                "<start_of_turn>user " + text.strip() + "<end_of_turn> \n\n"
                + "<start_of_turn>model "
            )
        return text