This dataset is an adapation of my previous claude_multiround_chat_30k dataset with only the first 30k instruction/response pairs and parsed into an instruct format.
The instructions were generated synethically using a method that can be tenatively described as "multi-instruct." These instructions consist of numerous discrete tasks that the AI has to work its way through, thereby hopefully increasing its comprehension and awareness of complex instructions.
The topics of the instruction ranged… See the full description on the dataset page:
https://huggingface.co/datasets/Norquinal/claude_multi_instruct_30k.
