skilledu/Dolphin3.0-Llama3.1-8B-abliterated
The skilledu/Dolphin3.0-Llama3.1-8B-abliterated is an 8 billion parameter language model, based on the Llama 3.1 architecture, specifically modified to be an uncensored version of cognitivecomputations/Dolphin3.0-Llama3.1-8B. This model was created using an 'abliteration' technique to remove refusals, serving as a proof-of-concept for uncensoring LLMs without TransformerLens. It is primarily intended for use cases requiring a less restrictive language model.
Loading preview...
Model Overview
The skilledu/Dolphin3.0-Llama3.1-8B-abliterated is an 8 billion parameter language model derived from cognitivecomputations/Dolphin3.0-Llama3.1-8B. Its core differentiator is the application of an "abliteration" technique, as detailed in the remove-refusals-with-transformers project, to create an uncensored version of the base model.
Key Characteristics
- Base Model: Built upon the Llama 3.1 architecture.
- Parameter Count: Features 8 billion parameters, offering a balance between performance and computational efficiency.
- Uncensored Nature: The primary modification involves the removal of refusal behaviors, making it suitable for applications where less restrictive outputs are desired.
- Proof-of-Concept: This model serves as a practical demonstration of a method to modify LLM behavior without relying on tools like TransformerLens.
Use Cases
This model is particularly suited for:
- Research into LLM censorship and modification techniques.
- Applications requiring a language model with fewer built-in content restrictions.
- Experimentation with uncensored language generation.
Users can easily deploy this model via Ollama using ollama run huihui_ai/dolphin3-abliterated.