Naphula/Slimaki-24B-v1.1-ramplus_tl
Naphula/Slimaki-24B-v1.1-ramplus_tl is a 24 billion parameter experimental merge model with a 32768 token context length, created using the Reinforced Agent Merging Plus (Tensor-Local) method. This model is specifically designed for generating narratives and roleplay content, including potentially violent and graphic erotic material, and may require specific system prompts and chat templates like Mistral Tekken for optimal use. Its primary differentiator is the application of the novel `ramplus_tl` mergekit method to components known for creative and uncensored outputs.
Loading preview...
Slimaki-24B-v1.1-ramplus_tl Overview
Naphula/Slimaki-24B-v1.1-ramplus_tl is an experimental 24 billion parameter language model, leveraging a 32768 token context window. It was created using the novel Reinforced Agent Merging Plus (Tensor-Local) (ramplus_tl) method from arcee-mergekit, drawing inspiration from models like Maginum Cydoms and Slimaki. This merge aims to explore the capabilities of this new merging technique on components known for generating diverse and often unconstrained narrative content.
Key Capabilities
- Experimental Merge Method: Utilizes the
ramplus_tlmethod, a new approach to merging language models, offering insights into its performance and characteristics. - Narrative and Roleplay Generation: Specifically designed and noted for its ability to produce detailed narratives and roleplay scenarios.
- Uncensored Content Potential: Capable of generating violent and graphic erotic content, requiring users to adjust system prompts and utilize specific chat templates (e.g., Mistral Tekken) to manage output.
Good For
- Exploring Novel Merge Techniques: Researchers and developers interested in the
ramplus_tlmerge method and its impact on model behavior. - Creative Writing and Roleplay: Users seeking a model capable of generating detailed and potentially explicit narrative content for creative projects.
- Testing Jailbreaks and Ablations: Developers working on methods to manage or bypass model refusals and censorship, as the model may exhibit such behaviors.