metadata
license: pddl
A tiny example dataset consisting of four books dedicated to the open domain in JSONL format:
- Alice in Wonderland - Lewis Caroll
- Dracula - Bram Stoker
- The Wonderful Wizard of Oz - L. Frank Baum
- The Count of Monte Cristo - Alexandre Dumas & Auguste Maquet
All works are open domain, thus this dataset is also dedicated to the open domain.
The dataset has been made to have extremely long context lengths, ideally as close to 2048 at possible without cutting off chunks in strange places. Each chunk is also overlapped by 100 characters from the prior chunk.
Example:
There's an example LORA model here, which also includes training instructions for repoduction: https://huggingface.co/Blackroot/Llama2-13B-Lora-Tiny-Opendomain-Example