agentlans/Llama3.1-vodka is an 8 billion parameter language model based on the Llama 3.1 Instruct architecture, developed by agentlans. This model is a merge of several 'abliterated' Llama 3.1 Instruct variants, specifically designed to be uncensored while retaining the core capabilities and writing style of the original Llama 3.1 Instruct. It offers a potent and versatile foundation for applications requiring less restrictive content generation, with a context length of 32768 tokens.
Loading preview...
Llama3.1-vodka: An Uncensored Llama 3.1 Instruct Variant
Llama3.1-vodka is an 8 billion parameter language model developed by agentlans, built upon the Llama 3.1 Instruct architecture. Its primary goal is to provide a "pure, potent, and versatile" uncensored alternative to the standard Llama 3.1 Instruct model, while largely preserving its original capabilities and writing style. The model achieves this by merging several 'abliterated' Llama 3.1 Instruct variants, including mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated and adapted versions of grimjim/Llama-3.1-8B-Instruct-abliterated_via_adapter and agentlans/Llama3-vodka.
Key Characteristics
- Uncensored Output: Designed to produce less restricted content, aiming to bypass typical safety filters.
- Llama 3.1 Instruct Foundation: Retains the core capabilities and conversational style of the base Llama 3.1 Instruct model.
- Merged Architecture: Created using the "task arithmetic" merge method on
NousResearch/Meta-Llama-3.1-8B-Instructwithmergekit. - Versatile Use: Suitable for direct use, mixing with other models, or further fine-tuning for specific applications.
Important Considerations
- Weakened Safeguards: The model has intentionally weakened safeguards and may generate harmful, offensive, or inappropriate content.
- User Responsibility: Users are responsible for all outputs produced by the model.
- Controlled Environments: Recommended for use in controlled environments where risks can be managed.
- Potential for Residual Censorship: While designed to be uncensored, it may still exhibit some censorship at times; users can prompt the AI to be more uninhibited if desired.