Model Overview
sh0ck0r/Strawberrylemonade-L3-70B-v1.2-heretic2 is a 70 billion parameter language model, derived from the sophosympatheia/Strawberrylemonade-L3-70B-v1.2 model through a decensoring process using Heretic v1.2.0. This version is designed to be uncensored, providing greater flexibility in content generation.
Key Capabilities
- Decensored Output: Achieves a refusal rate of 10/100, a significant reduction compared to the original model's 99/100, enabling less restricted content generation.
- Roleplaying and Storytelling: Explicitly designed and optimized for immersive roleplaying and creative storytelling tasks, offering enhanced creativity and expressiveness.
- Merge Architecture: Built upon a merge of several models, including zerofata/L3.3-GeneticLemonade-Final-v2-70B, zerofata/L3.3-GeneticLemonade-Unleashed-v3-70B, deepcogito/cogito-v1-preview-llama-70B, and arcee-ai/Arcee-SuperNova-v1, using the NuSLERP merge method.
- Configurable Output: Highly responsive to sampling parameters like
temperature and min-p, allowing users to fine-tune outputs for either reliable, coherent responses or more creative, diverse generations.
Good For
- Creative Writing: Excels in generating imaginative narratives and engaging in detailed storytelling.
- Roleplaying Scenarios: Ideal for interactive roleplaying applications where character consistency and expressive dialogue are crucial.
- Unrestricted Content Generation: Suitable for use cases where content filtering or refusal behaviors are undesirable, given its decensored nature.
Usage Notes
The model's licensing terms are noted as uncertain due to its merged nature, incorporating weights from multiple LLMs. Users are advised to consult legal counsel for commercial applications. Recommended sampler settings are provided for both reliable and creative outputs, along with specific prompting tips for platforms like Silly Tavern to manage character impersonation and content generation.