This dataset was created as a synthetic task to evaluate domain adaptation of LLMs, specifically, to compare the efficacy of in-context learning vs fine-tuning. The task is designed to be almost trivial for a human, but to require complex prompting for an LLM to perform well on it.
In this case, the task is to read a credit card transaction string and extract the merchant name. This is not trivial for a machine, as these strings… See the full description on the dataset page: https://huggingface.co/datasets/crossingminds/credit_card_3k.
