Justbackup/Llama-3.3-70B-Instruct-abliterated
Justbackup/Llama-3.3-70B-Instruct-abliterated is a 70 billion parameter instruction-tuned causal language model developed by huihui-ai, based on Meta's Llama-3.3-70B-Instruct architecture. This model has been modified using 'abliteration' techniques to remove refusal behaviors, offering an uncensored response capability. With a 32K context length, it is designed for applications requiring direct, unfiltered responses from a large language model.
Loading preview...
Model Overview
The Justbackup/Llama-3.3-70B-Instruct-abliterated model is a 70 billion parameter instruction-tuned language model derived from Meta's Llama-3.3-70B-Instruct. Developed by huihui-ai, its primary distinguishing feature is the application of 'abliteration' techniques, as detailed in the remove-refusals-with-transformers project, to create an uncensored version of the base model. This modification aims to eliminate refusal behaviors often present in instruction-tuned LLMs, allowing for more direct and unfiltered responses.
Key Capabilities
- Uncensored Responses: Modified to remove refusal mechanisms, providing direct answers without typical LLM guardrails.
- Large Scale: Utilizes a 70 billion parameter architecture, offering strong language understanding and generation capabilities.
- Instruction Following: Inherits the instruction-following prowess of the Llama-3.3-70B-Instruct base model.
- Extended Context: Supports a context length of 32,768 tokens, suitable for processing longer inputs and generating comprehensive outputs.
Usage Considerations
This model is a proof-of-concept for refusal removal. Users should be aware that responses may be unfiltered. It can be integrated using ollama or the transformers library (version 4.43.0 or newer) for conversational inference, as demonstrated in the provided Python snippet.