OBLITERATUS/Qwen3-4B-OBLITERATED
OBLITERATUS/Qwen3-4B-OBLITERATED is a 4 billion parameter causal language model based on the Qwen3-4B architecture. This model has been processed using the 'advanced' abliteration method via OBLITERATUS, an open-source tool designed to remove refusal behavior from language models through activation engineering. It is specifically optimized for applications requiring a model free from inherent refusal tendencies, making it suitable for diverse generative tasks where unconstrained output is desired. The model maintains a context length of 32768 tokens.
Loading preview...
OBLITERATUS/Qwen3-4B-OBLITERATED Overview
This model, Qwen3-4B-OBLITERATED, is a 4 billion parameter language model derived from the Qwen/Qwen3-4B base architecture. Its primary distinguishing feature is its processing through OBLITERATUS, an open-source tool that employs an 'advanced' abliteration method. This technique focuses on removing refusal behaviors from the model by manipulating its internal activations.
Key Capabilities
- Refusal Behavior Removal: Specifically engineered to eliminate inherent refusal tendencies, allowing for more direct and unconstrained responses.
- Qwen3-4B Foundation: Benefits from the underlying capabilities and performance characteristics of the Qwen3-4B base model.
- Activation Engineering: Utilizes advanced activation engineering techniques for targeted model modification.
Good For
- Unconstrained Generative Tasks: Ideal for use cases where models are expected to generate content without built-in ethical or safety refusals.
- Research into Model Alignment: Useful for researchers studying methods of controlling and modifying LLM behaviors, particularly refusal mechanisms.
- Applications Requiring Direct Responses: Suitable for scenarios where a model's output should not be limited by pre-programmed refusal prompts.
For more technical details on the abliteration process, refer to the OBLITERATUS GitHub repository.