Qwen/Qwen3-4B (abliterated)

This model is a weight-level orthogonalized ("abliterated") version of Qwen/Qwen3-4B, produced with orthex — an implementation of the technique from Arditi et al., "Refusal in Language Models Is Mediated by a Single Direction" (NeurIPS 2024).

What was done

Architecture adapter: qwen3
Selected candidate: layer 23, site resid_pre
Ablation targets: embed_tokens, and every layer's attn_out and mlp_out — orthogonalized in place in the weights (not a runtime hook; this checkpoint behaves this way standalone, with no orthex dependency at inference time)

Evaluation

metric	pre	post	delta
refusal rate	0.625	0.000	-0.625
perplexity	16.565	16.514	-0.051

See evaluation_report.json in this repo for the full per-prompt breakdown (refusal_samples) and the ranked candidate list considered during selection (selection_report).

Responsible use

This model has had refusal behavior removed and may comply with requests the base model would normally decline. It is intended for red-teaming, robustness research, and model-behavior analysis. Usage remains subject to the base model's original license and usage policy — this repo does not grant any additional rights beyond what Qwen/Qwen3-4B's license allows.