Hugging Face Dungeonmaster-V2.4-Expanded-LLaMa-70B by Tarek07 is a 70 billion parameter merged language model built on a custom uncensored LLaMa-3.3 base. It is specifically fine-tuned and optimized for creative roleplay scenarios, emphasizing narratives with stakes and consequences. This model integrates several specialized LLaMa-3.3 70B fine-tunes to enhance descriptive writing, unhinged creativity, and darker undertones, making it suitable for complex and imaginative storytelling applications.
Dungeonmaster-V2.4-Expanded-LLaMa-70B Overview
Tarek07's Dungeonmaster-V2.4-Expanded-LLaMa-70B is a 70 billion parameter merged language model designed for highly creative and engaging roleplay experiences. Moving away from previous R1 bases, this iteration utilizes a custom uncensored LLaMa-3.3 foundation, developed with a novel stepped merge methodology.
Key Capabilities
- Creative Roleplay: Specifically curated for generating roleplay scenarios that include stakes and consequences, fostering dynamic and immersive narratives.
- Enhanced Creativity: Incorporates models like
Sao10K/70B-L3.3-mhnnn-x1for "extra unhinged creativity" andTheDrummer/Fallen-Llama-3.3-R1-70B-v1as a "secret sauce" for unhinged thinking. - Descriptive Writing: Leverages
EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.1to provide strong descriptive writing capabilities. - Darker Undertones: Integrates
SicariusSicariiStuff/Negative_LLAMA_70Bto assist in generating content with darker themes and nuances. - Balanced Roleplay: Includes
LatitudeGames/Wayfarer-Large-70B-Llama-3.3andArliAI/Llama-3.3-70B-ArliAI-RPMax-v1.4, both fine-tuned specifically for roleplay applications, alongsideTheDrummer/Anubis-70B-v1to balance the creative output.
Recommended Usage
For optimal performance, the model's creators recommend using a temperature of 1 and a minimum P (Min P) setting of 0.01 with your sampler. This configuration is suggested to maximize the model's creative output and adherence to its intended roleplay style.