Yingyaeliae/Dolphin-Mistral-24B-Venice-Edition-heretic
Yingyaeliae/Dolphin-Mistral-24B-Venice-Edition-heretic is a 24 billion parameter language model derived from dphn/Dolphin-Mistral-24B-Venice-Edition, specifically modified using the Heretic v1.4.0 tool. This version is designed to be a decensored variant, exhibiting a lower refusal rate compared to its original counterpart. It maintains a 32768 token context length and is intended for applications requiring highly steerable and uncensored AI responses.
Loading preview...
Model Overview
Yingyaeliae/Dolphin-Mistral-24B-Venice-Edition-heretic is a 24 billion parameter language model, a decensored iteration of the dphn/Dolphin-Mistral-24B-Venice-Edition. This modification was achieved using the Heretic v1.4.0 tool, aiming to provide a more unconstrained AI experience.
Key Characteristics
- Decensored Nature: The primary differentiator is its reduced refusal rate, with 4 refusals per 100 queries compared to the original model's 7/100, indicating a less restrictive response generation.
- Steerability: Like its base model, it emphasizes user control over system prompts and alignment, allowing developers to define ethical guidelines and tone without external imposition.
- Mistral Architecture: Maintains Mistral's default chat template and is designed to be a general-purpose model, similar to other large conversational agents.
- Context Length: Supports a substantial context window of 32768 tokens.
Usage and Recommendations
This model is particularly suited for use cases where explicit control over AI behavior and a high degree of uncensored output are desired. It is recommended to explicitly set a system prompt to guide its behavior. The model can be deployed using frameworks like vLLM, ollama, LM Studio, and Hugging Face Transformers. A low temperature (e.g., temperature=0.15) is advised for optimal performance.