CrucibleLab/L3.3-Dark-Ages-70b-v0.1

L3.3-Dark-Ages-70b: A Multi-Stage Merged Model for Dark Fantasy Roleplaying

CrucibleLab/L3.3-Dark-Ages-70b-v0.1 is a 70 billion parameter language model developed by Crucible Lab, a collaboration between SteelSkull and Tarek. This model is the result of an intricate multi-stage merging process, built upon the Llama 3.3 architecture, and is specifically designed to excel in creative roleplaying scenarios with a distinct dark fantasy flavor.

Model Creation Process

The development of L3.3-Dark-Ages-70b involved three key stages:

• Stage 1: Cognitive Foundation (Cogmo-70b): The process began with a model_stock merge to create "Cogmo-70b," a base model focused on cognitive abilities and tool-use, leveraging models known for reasoning and instruction-following, built on a Nemotron-lorablated base.
• Stage 2: Narrative Weaving (Dark-Prose & Negative-RP): Two distinct models were then derived from Cogmo-70b. "Dark-Prose-70b" was created using an SCE merge to enhance creative writing and prose style, incorporating models like ArliAI's RPMax and Anubis. Concurrently, "Negative-RP-70b" was developed via a della_linear merge, integrating various roleplaying models (e.g., Wayfarer, Vulpecula) and utilizing Negative_LLAMA to balance its output.
• Stage 3: Final Merge (Dark-Ages-70b): The final model, "Dark-Ages-70b," was produced by merging Dark-Prose and Negative-RP using the nuslerp method. This step combines the rich, creative prose with robust, negative-prompt-influenced roleplaying capabilities.

Key Capabilities & Use Cases

• Creative Roleplaying: Optimized for generating detailed and immersive roleplaying narratives, particularly within dark fantasy settings.
• Rich Prose Generation: Excels at producing high-quality, descriptive text, enhancing storytelling and character development.
• Negative Prompt Influence: Designed to effectively interpret and utilize negative prompts, allowing for greater control over generated content and avoiding undesired elements.
• Complex Character Interactions: Capable of handling intricate character dynamics and maintaining consistency in their knowledge and learning.

Recommended Sampler Settings

For optimal performance, the following sampler settings are recommended:

• Static Temperature: 1.0 - 1.5
• Min P: 0.02
• DRY: Multiplier: 0.8, Base: 1.74, Length: 4-6

Future Development

This initial release (v0.1) is an un-finetuned version of the final merge. Crucible Lab plans further finetuning to enhance its performance and solidify its unique characteristics based on user feedback.