Data from the https://huggingface.co/datasets/rickRossie/bluemoon_roleplay_chat_data_300k_messages bluemoon RP dataset.

subset, 6 million tokens.

ShareGPT format. For countering context blindness in professional LLMs. No changes from original besides potentially data format being set to sharegpt.

Believe it or not, training on this data makes the model better at conversation, and at using/understanding previous context. If you're a professional user, don't look at the dataset itself, retain plausible deniability.