The vicgalle/Configurable-Llama-3-8B-v0.1 model, developed by vicgalle, is a Llama-3 based fine-tuned language model. It utilizes a configurable safety tuning (CST) approach, allowing users to define its behavior through various system prompts, ranging from helpful and harmless to completely uncensored. This model is specifically designed for adaptable content generation based on user-defined safety and persona configurations.
Loading preview...
Overview
vicgalle/Configurable-Llama-3-8B-v0.1 is a Llama-3 based language model fine-tuned by vicgalle using the configurable safety tuning (CST) approach, detailed in the paper arXiv:2404.00495. This model is built upon the vicgalle/configurable-system-prompt-multitask dataset, enabling dynamic control over its output behavior.
Key Capabilities
- Configurable Behavior: Users can define the model's persona and safety guidelines through specific system prompts.
- Safety Control: Supports system prompts for generating helpful yet harmless content, or for completely uncensored responses.
- Persona Emulation: Capable of adopting role-played personas based on detailed system prompt descriptions.
- Adaptable Output: Provides flexibility in content generation, from unbiased and truthful assistance to more creative or unrestricted outputs.
Usage
This model is particularly well-suited for applications requiring flexible AI behavior, where the level of helpfulness, harmlessness, or censorship needs to be dynamically adjusted. It allows developers to tailor the model's responses to specific use cases without retraining, by simply modifying the system prompt. For more technical details and implementation, refer to the GitHub repository.
Note: An updated version, vicgalle/Configurable-Llama-3-8B-v0.2, is available and generally recommended for use.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.