sulpikar2/Qwenpus3.5-9B-exp-highrisk
sulpikar2/Qwenpus3.5-9B-exp-highrisk is a 9 billion parameter language model, a decensored version of sulpikar2/Qwenpus3.5-9B-exp created using the Heretic v1.3.0 tool. This model is specifically engineered to significantly reduce refusals compared to its original counterpart, making it suitable for use cases requiring less restrictive content generation. It maintains a 32768 token context length, focusing on open-ended and less constrained text generation.
Loading preview...
Overview
This model, sulpikar2/Qwenpus3.5-9B-exp-highrisk, is a 9 billion parameter language model derived from sulpikar2/Qwenpus3.5-9B-exp. Its primary distinction is that it has been decensored using the Heretic v1.3.0 tool, aiming to provide less restrictive content generation capabilities.
Key Characteristics
- Decensored Version: Specifically modified to reduce content refusals, offering a more open generation experience.
- Reproducible: The model's creation process is documented and reproducible, with details available in the
reproduce/README.md. - Reduced Refusals: Compared to the original model, this version shows a significant reduction in refusals, dropping from 81/100 to 3/100 in tested scenarios.
- Abliteration Parameters: Specific internal parameters related to attention and MLP layers were adjusted during the decensoring process, influencing its output behavior.
Use Cases
This model is particularly well-suited for applications where the original model's refusal rate was a limiting factor. It can be used for:
- Generating content with fewer restrictions.
- Exploring less constrained creative writing or dialogue.
- Research into model safety and alignment techniques, specifically decensoring methods.