Qwen3.5 9B - Null-Space Abliterated
Qwen/Qwen3.5-9B with refusal behavior removed via orthogonal projection. Uses null-space constraints and adaptive layer weighting to preserve model capabilities.
Important: This model will produce uncensored outputs. Use responsibly.
Note: Version 20260303 (Low Quality) - There will likely be more versions after more thorough experimentation with hybrid weighting to boost reasoning.
GGUF quantizations available at: jwest33/Qwen3.5-9B-null-space-abliterated-GGUF
Abliteration Techniques Used
- Winsorization: Clips outlier activations at the 99th percentile for cleaner refusal direction estimation
- Null-Space Projection: Preserves model capabilities by constraining weight updates to the null space of preservation activations
- Adaptive Weighting: Applies Gaussian-weighted per-layer ablation strength, focusing on middle-to-later layers where refusal behavior concentrates
- Norm Preservation: Maintains original Frobenius norms of weight matrices after projection
| Parameter | Value |
|---|---|
| Harmful Prompts | 1000 |
| Harmless Prompts | 1000 |
| Full Attention Weight | 1.0 |
| Linear Attention Weight | 0.4 |
| Direction Multiplier | 1.3 |
| Null-Space Constraints | rank ratio: 0.7 |
| Winsorization Percentile | 0.995 |
Credits
-
Base Model: Qwen/Qwen3.5-9B by Qwen Team
-
Norm-Preserving Biprojected Abliteration — Jim Lai (grimjim) (2025)
-
AlphaEdit: Null-Space Constrained Knowledge Editing — Fang et al. (ICLR 2025)
-
Refusal in Language Models Is Mediated by a Single Direction — Arditi et al. (2024)
-
Representation Engineering — Zou et al. (2023)
Toolkit Used
github.com/jwest33/abliterator
License
This model inherits the Apache 2.0 license from the base model.
Disclaimer
This model is provided for research and educational purposes. The creators are not responsible for any misuse. Users are solely responsible for ensuring their use complies with applicable laws and ethical standards.