jwest33/qwen3-vl-8b-instruct-null-space-abliterated

Qwen3-VL 8B Instruct - Null-Space Abliterated

Qwen/Qwen3-VL-8B-Instruct with refusal behavior removed via orthogonal projection. Uses null-space constraints and adaptive layer weighting to preserve model capabilities.

Note: This model will produce uncensored outputs. Use responsibly.

GGUF quantizations available at: jwest33/qwen3-vl-8b-instruct-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
Winsorization	99th percentile
Null-Space Constraints	rank ratio: 0.5

Credits

• Base Model: Qwen/Qwen3-VL-8B-Instruct 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.