DarkArtsForge/Zepar-24B-v1
DarkArtsForge/Zepar-24B-v1 is a 24 billion parameter Mistral-based causal language model developed by DarkArtsForge. It utilizes a novel 'adaptiveswap' merge method, based on nearswap, to intelligently blend various swapping mechanisms. This model is specifically noted for its capability to produce narratives and roleplay content, including violent and graphic erotic themes, and is optimized for use with the Mistral Tekken template.
Loading preview...
Zepar 24B v1 Overview
Zepar 24B v1 is a 24 billion parameter language model built on the Mistral architecture, developed by DarkArtsForge. This model introduces a novel adaptiveswap merge method, an evolution of nearswap, designed to intelligently combine different merging techniques using a Pareto Frontier AI controller. The adaptiveswap method dynamically blends mechanisms such as cosine, magnitude, variance, salience, and slerp, with an autopilot feature for optimized precision.
Key Characteristics
- Architecture: MistralForCausalLM base.
- Parameter Count: 24 billion parameters.
- Merge Method: Employs
adaptiveswap, an advanced technique for blending models. - Precision: Utilizes
float32for internal processing andbfloat16for output. - Chat Template: Automatically configured for chat interactions.
Use Cases and Considerations
Zepar 24B v1 is particularly noted for its ability to generate diverse narrative and roleplay content. Users should be aware that the model can produce material containing violent and graphic erotic themes, and system prompts should be adjusted accordingly. For optimal performance, the model is designed to be used with the Mistral Tekken chat template.