Hugging Face Overview
Dungeonmaster-V2.2-Expanded-LLaMa-70B Overview
Dungeonmaster-V2.2-Expanded-LLaMa-70B is a 70 billion parameter merged language model developed by Tarek07. It moves away from previous R1 bases, utilizing a custom uncensored base and the Linear DELLA merge method. The model is specifically designed for creative roleplay, aiming to generate narratives with significant stakes and consequences.
Key Capabilities
- Enhanced Roleplay: Fine-tuned using a selection of models specifically designed for roleplaying, including LatitudeGames/Wayfarer-Large-70B-Llama-3.3 and ArliAI/Llama-3.3-70B-ArliAI-RPMax-v1.4.
- Uncensored and Unhinged Creativity: Incorporates models like Sao10K/70B-L3.3-mhnnn-x1 and TheDrummer/Fallen-Llama-3.3-R1-70B-v1 to foster highly creative and unrestricted outputs.
- Strong Descriptive Writing: Leverages EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.1 for rich and detailed narrative generation.
- Darker Undertones: Includes SicariusSicariiStuff/Negative_LLAMA_70B to assist in generating content with darker themes when appropriate for the roleplay.
- Prompt Following: Optimized to follow user prompts effectively, ensuring relevant and engaging responses.
Recommended Usage
This model is particularly well-suited for applications requiring highly creative, detailed, and potentially unhinged roleplay scenarios. Optimal sampler settings are suggested around Temp: 0.8 and Min P: 0.02 for best results. It is built using mergekit and uses TareksLab/L3.3-TRP-BASE-80-70B as its base model.