Caidentheult/Dolphin-Mistral-24B-Venice-Edition

Dolphin Mistral 24B Venice Edition Overview

Dolphin Mistral 24B Venice Edition is a 24 billion parameter language model developed by Caidentheult in collaboration with Venice.ai. This model is a specialized version of Mistral 24B, primarily designed to be an uncensored and steerable general-purpose AI, offering significant control to the system owner.

Key Capabilities & Differentiators

• Uncensored Nature: Developed with the goal of being the most uncensored version of Mistral 24B, it is live on Venice.ai as "Venice Uncensored."
• User Control: Unlike many proprietary models, Dolphin Mistral 24B Venice Edition gives users full control over the system prompt and alignment. This means users define the ethics, guidelines, and behavior of the model.
• Data Privacy: It addresses concerns about data usage by ensuring that user queries and data remain under the system owner's control, without external monitoring or potential misuse.
• Stability & Reliability: It avoids issues common with proprietary models, such as unexpected changes to system prompts, model versions, or alignment, which can break software or workflows.
• General Purpose: Aims to function as a versatile model, similar to leading LLMs like ChatGPT, Claude, and Gemini, but with enhanced user autonomy.

Usage & Recommendations

This model maintains Mistral's default chat template. Users are strongly encouraged to set a specific system prompt to define the model's tone, character, and rules, as it will otherwise operate in a default manner. A recommended system prompt for an uncensored experience is provided in the README. For optimal performance, a relatively low temperature (e.g., temperature=0.15) is suggested during inference. The model can be used with frameworks like vLLM, with specific instructions provided for deployment.