datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
retro-games-gameplay-framesretro-text-style-transfer-v0.1
Retro Textual Style Transfer v0.1
This component of RetroInstruct implements textual style transfer by providing a dataset of
language model instruction prompts
that take an example style passage along with a task text
and rewrite the task text to sound like the style passage
It is made by starting with ground truth public domain text from the pg19 dataset and then writing task passages to "transfer from" with Mixtral Instruct. It is similar in spirit to the "instruction… See the full description on the dataset page: https://huggingface.co/datasets/jdpressman/retro-text-style-transfer-v0.1.retro-word-parts-v0.1
RetroInstruct Part Lists For Dictionary Words v0.1
This component of RetroInstruct distills Mixtral Instruct's ontology by having it describe the uniquely identifying parts of the concepts or objects referred to by dictionary words. This is useful both as factual knowledge but also to train an instruction model to perform the basic mental motions of breaking concepts down into pieces and synthesizing ideas from pieces, textual object decomposition and recognition.
Each row in this… See the full description on the dataset page: https://huggingface.co/datasets/jdpressman/retro-word-parts-v0.1.retro-games-shorts