This dataset is a variant of the Multi-Turn-Insurance-Underwriting dataset, in which models do not get access to any tools except a code interpreter and a pointer to the relevant file system.
This helps us analyze how well models explore their environments.
This diagram shows the architecture of how we create the dataset, with assistant responses interleaved with questions, ending with a… See the full description on the dataset page:
https://huggingface.co/datasets/snorkelai/Multi-Turn-Insurance-Underwriting-Code-Gen.
