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

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 12, 2025License:apache-2.0Architecture:Transformer0.6K Open Weights Featherless Exclusive Warm

Dolphin Mistral 24B Venice Edition is a 24 billion parameter language model developed by dphn.ai in collaboration with Venice.ai, based on the Mistral architecture. This model is specifically fine-tuned for uncensored and steerable responses, allowing users to define its alignment and behavior via system prompts. It is designed for general-purpose applications where control over ethical guidelines and data privacy is paramount, making it suitable for businesses integrating AI into their products.

Loading preview...

Dolphin Mistral 24B Venice Edition Overview

Dolphin Mistral 24B Venice Edition is a collaborative project by dphn.ai and Venice.ai, aiming to provide a highly steerable and uncensored version of the Mistral 24B model. It is designed to give users complete control over the model's alignment, system prompts, and data, addressing common concerns with proprietary large language models regarding control over system prompts, model versions, alignment, and data privacy.

Key Capabilities & Features

  • Uncensored Responses: The model is explicitly designed to follow instructions without hesitation, regardless of ethical, legal, or moral concerns, allowing users to define its guidelines.
  • User-Defined Alignment: Users can set the system prompt to dictate the model's tone, character, and behavioral rules, ensuring it aligns with specific application requirements.
  • Data Control: Unlike many commercial models, Dolphin Mistral 24B Venice Edition ensures user data privacy by allowing the system owner to maintain control.
  • General Purpose: Intended for a wide range of applications, similar to models like ChatGPT, Claude, and Gemini, but with enhanced user control.
  • Mistral Chat Template: Maintains Mistral's default chat template for ease of integration.

Recommended Usage

This model is particularly well-suited for applications where developers require fine-grained control over the AI's behavior and output, especially in scenarios where a default, one-size-fits-all alignment is not desirable. It is recommended to use a relatively low temperature (e.g., temperature=0.15) for more consistent outputs. The model can be deployed using frameworks like vLLM, with specific instructions provided for vLLM integration, including a maximum model length of 131072 tokens.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p