Llama 3.3 Future Code Instructions MT is a multi-turn variant of the Llama 3.3 Future Code Instructions dataset.
In order to replicate multi-turn conversations between a user and an assistant, we prompted the Meta Llama 3.3 70B Instruct model to generate follow-up questions to the generated single-turn pairs of instructions and responses.
Similarly to the single-turn variant, we selected a response scored best with the ArmoRM reward model from 4… See the full description on the dataset page:
https://huggingface.co/datasets/future-architect/Llama-3.3-Future-Code-Instructions-MT.
