Atlantis-v0.1-12B: A Model for Roleplay and Storytelling

Atlantis-v0.1-12B is a 12 billion parameter language model developed by invisietch, built upon the Mistral-Nemo-2407-Instruct architecture. This model has been specifically fine-tuned to excel in roleplay (RP) and storytelling contexts, aiming to provide more coherent and engaging narrative generation than its base model for these specific use cases.

Key Capabilities

• Enhanced Roleplay and Storytelling: Optimized for generating creative and immersive narratives, character interactions, and plot developments.
• Context Length: Supports a context window of up to 32768 tokens, with practical coherence for RP settings estimated around 24K tokens.
• Unaligned Nature: The model is noted as being fairly unaligned, capable of generating explicit, disturbing, or offensive responses, offering flexibility for diverse narrative themes.
• Prompt Format: Recommends the Mistral prompt format ([INST]...[/INST]) for optimal performance.
• Sampler Settings: Suggests specific sampler settings (e.g., Temp: 0.3-0.5, Top-P: 1, Min-P: 0.05, Rep Pen: 1.13-1.16, Response Length: 350+) for best results, emphasizing a low temperature.

Good For

• Developers and enthusiasts looking for a model specialized in generating detailed and engaging roleplay scenarios.
• Applications requiring creative story generation, character dialogue, and narrative continuation.
• Users who need a model with a flexible output range, including potentially unaligned content, for specific creative projects.

This is an early-stage fine-tune, and feedback is encouraged through various channels to aid in its improvement.