Evathene-v1.0: A Roleplaying and Storytelling LLM
Evathene-v1.0 is a 72.7 billion parameter language model developed by sophosympatheia, created through a merge of Nexusflow/Athene-V2-Chat and EVA-UNIT-01/EVA-Qwen2.5-72B-v0.1 using the TIES merge method. This model is uncensored and primarily optimized for generating high-quality content in roleplaying and storytelling scenarios.
Key Capabilities
- Specialized for Roleplaying and Storytelling: Designed to excel in generating engaging and consistent narratives, making it ideal for interactive fiction and character-driven experiences.
- Uncensored Output: Provides unrestricted content generation, allowing for diverse and mature themes in creative writing.
- Intelligent Responses: Preserves much of Athene's intelligence, responding competently to out-of-character (OOC) prompts and rewrite requests.
- Extensive Context Window: Features a 131072-token context length, supporting long and complex narrative arcs.
Use Cases
- Roleplaying: Excellent for interactive roleplaying sessions, generating character dialogue and actions with personality.
- Story Generation: Suitable for creative writing, generating plotlines, and developing narrative content.
- Experimental Content Creation: Its uncensored nature allows for exploration of a wide range of themes and styles in creative projects.
Important Considerations
- Licensing Uncertainty: As a merged model, its licensing terms are complex due to the combination of multiple base models. Users are advised to consult legal counsel for commercial applications.
- Sampler and Prompting Tips: The developer provides specific recommendations for sampler settings (e.g., Min-P, DRY repetition penalty) and instruct templates to optimize performance, particularly within platforms like Silly Tavern.