| license: apache-2.0 | |
| datasets: | |
| - openerotica/Lamia | |
| tags: | |
| - NSFW | |
| - Porn | |
| - Ecommerce | |
| - Roleplay | |
| - Summarization | |
| This is a combination of the pruned erotica-analysis data, freedom-rp, and a subest of Airoboros. | |
| The following Categories are what was taken out of the Airoborus datset and added to my own Lamia dataset: | |
| "roleplay", "unalignment", "editor", "writing", "detailed_writing", "stylized_response", "unalign", "cot", "song" | |
| I'm hoping that this can improve the models narrative/storywriting ability, logic, and intelligence, while reducing any potential inherent ethical "alignment" that may be present in the base mistral model from pretaining on Chat-GPT generated data. | |
| The format is Chatml, and the base model is Yarn Mistral which increases the context size to a true 16k+ rather than rellying on the sliding attention window. |