Comparative Analysis of LLM Abliteration Methods: A Cross-Architecture Evaluation

TL;DR

This paper evaluates four abliteration tools across sixteen instruction-tuned large language models, analyzing their effectiveness and impact on capabilities, especially in mathematical reasoning, to guide researchers in tool selection.

Contribution

It provides a comprehensive cross-architecture evaluation of abliteration methods, highlighting their effects on model capabilities and offering evidence-based selection criteria.

Findings

Single-pass methods better preserve capabilities on benchmark tasks.

Bayesian-optimized abliteration causes variable distribution shifts.

Mathematical reasoning is highly sensitive to abliteration interventions.

Abstract

Safety alignment mechanisms in large language models prevent responses to harmful queries through learned refusal behavior, yet these same mechanisms impede legitimate research applications including cognitive modeling, adversarial testing, and security analysis. While abliteration techniques enable surgical removal of refusal representations through directional orthogonalization, the relative effectiveness of available implementations remains uncharacterized. This study evaluates four abliteration tools (Heretic, DECCP, ErisForge, FailSpy) across sixteen instruction-tuned models (7B-14B parameters), reporting tool compatibility on all 16 models and quantitative metrics on subsets dictated by tool support. Single-pass methods demonstrated superior capability preservation on the benchmarked subset (avg GSM8K change across three models: ErisForge -0.28 pp; DECCP -0.13 pp), while Bayesian-optimized abliteration produced variable distribution shift (KL divergence: 0.043-1.646) with model-dependent capability impact. These findings provide researchers with evidence-based selection criteria for abliteration tool deployment across diverse model architectures. The principal finding indicates that mathematical reasoning capabilities exhibit the highest sensitivity to abliteration interventions, with GSM8K change ranging from +1.51 pp to -18.81 pp (-26.5% relative) depending on tool selection and model architecture.