Zhang-User-001/Huihui-Qwen3.5-4B-abliterated
Huihui-Qwen3.5-4B-abliterated is a 4.5 billion parameter, 32K context length causal language model developed by Zhang-User-001, fine-tuned from Qwen/Qwen3.5-4B. This model has undergone significant reduction in safety filtering, making it an uncensored version. It is designed for research and experimental use where generating sensitive or controversial content is a potential outcome.
Loading preview...
Model Overview
Huihui-Qwen3.5-4B-abliterated is a 4.5 billion parameter language model, fine-tuned from the Qwen/Qwen3.5-4B base model. Developed by Zhang-User-001, this version has been specifically modified to significantly reduce safety filtering, resulting in an uncensored model. It was trained using TRL, focusing solely on the text component of the original model, with a context length of 32768 tokens.
Key Characteristics
- Uncensored Output: Safety filtering has been substantially reduced, allowing for the generation of content that may be sensitive, controversial, or inappropriate.
- Experimental Fine-tuning: Utilizes a fine-tuned dataset described as operating in a "non-think mode" to potentially simplify thinking processes.
- Compatibility: Despite changes in weight names (no longer containing MTP weights), it remains compatible with
ollamaandllama.cppfor conversion and use.
Usage Considerations
This model is primarily intended for research, testing, or controlled environments. Users should be aware of the following:
- Risk of Inappropriate Content: Due to limited content filtering, outputs may be unsuitable for public settings or applications requiring high security.
- Legal and Ethical Responsibility: Users are solely responsible for ensuring their usage complies with local laws and ethical standards.
- Monitoring Recommended: Real-time monitoring and manual review of outputs are strongly advised to prevent the dissemination of inappropriate content.