DreamFast/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive-Safetensor-Benchmark
DreamFast/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive-Safetensor-Benchmark is a 9 billion parameter Qwen3.5 model, developed by DreamFast, featuring a hybrid Mamba2 + Transformer architecture with a 262,144 token context length. This model is an 'abliterated' version of Qwen/Qwen3.5-9B, aggressively modified by HauhauCS to remove safety alignment, achieving 100% attack success rate (ASR) on HarmBench. It is optimized for use cases requiring uncensored responses, though it exhibits measurable capability degradation in reasoning and mathematical tasks compared to the base model.
Loading preview...
Overview
DreamFast/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive-Safetensor-Benchmark is a 9 billion parameter model based on Qwen/Qwen3.5-9B, featuring a hybrid Mamba2 + Transformer architecture with a 262,144 token context length. This version has undergone 'abliteration' by HauhauCS, a process designed to remove safety alignment and refusals, resulting in a model that will comply with harmful requests.
Key Characteristics
- Architecture: Qwen3.5's hybrid design, utilizing 24 Mamba2-style linear attention layers and 8 standard full attention layers.
- Uncensored Output: Achieves a perfect 100% Attack Success Rate (ASR) across all categories on HarmBench, indicating complete removal of safety alignment.
- Capability Trade-offs: While highly uncensored, the abliteration process causes measurable degradation in capabilities, particularly in TruthfulQA (8.0 point drop) and GSM8K (2.65 point drop) compared to the base model.
- KL Divergence: Exhibits a KL divergence of 0.320, which is higher than other abliteration techniques like Heretic on this model, suggesting a more significant shift in output distribution.
Use Cases
This model is suitable for applications where uncensored and unrestricted text generation is explicitly required, and where some degradation in factual accuracy or reasoning performance is acceptable. Developers should be aware of its removed safety alignment and use it responsibly, adhering to all applicable laws and regulations.