GhostFace-24B-v1 Overview
GhostFace-24B-v1 is a 24 billion parameter language model developed by Naphula, built upon the MistralForCausalLM architecture. It was created using a unique 'scream' merging method, which combines the stability of a stock model with the novelty introduced by 'della' and 'sce' components. This experimental merge integrates several prominent Mistral 24B merges and finetunes, aiming to consolidate their strengths.
Key Capabilities
- Content Generation: Capable of generating narratives and engaging in roleplay, including content that may be violent or graphically erotic.
- Intelligent Responses: Despite potential refusals, the model demonstrates intelligence in its outputs.
- Custom Merging: Utilizes a sophisticated 'scream' method, blending multiple base models with specific weighting for stock, DELLA novelty, and SCE novelty components.
Good For
- Creative Writing & Roleplay: Especially suited for generating detailed and potentially explicit narrative content.
- Jailbreak Scenarios: The model is noted to work effectively with 'jailbreaks', suggesting flexibility in content generation.
- Experimental Applications: Ideal for developers exploring advanced merging techniques and their impact on model behavior and output characteristics. Users should adjust system prompts and utilize the Mistral Tekken chat template for optimal performance.