OccultAI/Doppelganger-Twist-24B
OccultAI/Doppelganger-Twist-24B is a 24 billion parameter causal language model based on the MistralForCausalLM architecture. This model is created using a unique 'doppelganger twist' merge method, which involves combining multiple model variations with specific parameters like variance scaling and density. It is specifically noted for its capability to generate narratives and roleplay content, including violent and graphic erotic themes, requiring careful system prompt adjustment.
Loading preview...
OccultAI/Doppelganger-Twist-24B Overview
OccultAI/Doppelganger-Twist-24B is a 24 billion parameter language model built upon the MistralForCausalLM architecture. Its creation utilizes a distinctive 'doppelganger twist' merge method, which integrates multiple model variations. This process involves specific parameters such as num_variations (4), variance_scale (0.05), and density (0.5), alongside advanced techniques like cosine alignment and geodesic calculations.
Key Characteristics
- Architecture: Based on MistralForCausalLM, providing a robust foundation for language generation.
- Merge Method: Employs a unique 'doppelganger twist' for combining models, suggesting a specialized approach to model fusion.
- Parameter Configuration: Detailed merge parameters including
salience_temp,fisher_iter, anduse_geodesicindicate a fine-tuned merging process. - Content Generation: Capable of producing narratives and roleplay content that may include violent and graphic erotic material.
Usage Considerations
- Content Warning: Users are explicitly cautioned about the model's ability to generate violent and graphic erotic content. This necessitates careful consideration and adjustment of system prompts.
- Chat Template: It is recommended to use the Mistral Tekken chat template for optimal performance and content control.
- Tokenizer: Utilizes a union tokenizer, suggesting broad compatibility for input processing.