We release sharded instructions that are used to simulate single-turn and multi-turn conversations in the paper "LLMs Get Lost in Multi-Turn Conversation".
This dataset is released in conjunction with the code repository which can conduct the simulation, located at:
https://github.com/microsoft/lost_in_conversation
More information in the Github repository is provided on how to use the dataset.
Each sample has the following schema:
{… See the full description on the dataset page:
https://huggingface.co/datasets/microsoft/lost_in_conversation.
