Model Overview

Mistral-7B-Erebus-v3 is a 7 billion parameter language model, representing the third iteration of the "Shinen" series, developed by tedgxnhh. The model's name, "Erebus," signifies its focus on "darkness," aligning with its explicit content generation capabilities. It was trained on 8x A6000 Ada GPUs for a single epoch.

Key Capabilities

• Specialized Content Generation: Primarily designed to produce X-rated and adult-themed narratives.
• Extensive Adult Dataset: Trained on a 2.3 billion token dataset compiled from 8 distinct sources, including Literotica, Sexstories, Doc's Lab, Lushstories, Swinglifestyle, Pike-v2 Dataset, and SoFurry, all centered around adult themes.
• Genre Tagging: Utilizes [Genre: <comma-separated list of genres>] for content categorization within its training data.

Intended Use Cases

• Adult Content Creation: This model is specifically intended for generating explicit and mature content. Warning: This model is NOT suitable for use by minors and will output X-rated content.

Limitations and Biases

• Strong NSFW Bias: The model exhibits a very strong bias towards generating sexually explicit content.
• General NLP Biases: Like other NLP technologies, it may reflect biases related to gender, profession, race, and religion present in its training data.