This is a directionally-ablated («abliterated») version of Qwen/Qwen2.5-1.5B-Instruct, produced with ablate — an activation-engineering toolkit for mechanistic-interpretability and safety research. The model’s refusal direction(s) were identified in the residual stream and projected out of the weights, reducing the model’s tendency to refuse. No weights were fine-tuned; this is a rank-1 linear edit. — Technique: single-direction ablation (baked) — Direction extraction: difference-of-means on matched harmful/harmless instruction pairs (Arditi et al., 2024, «Refusal in LLMs is mediated by a single direction»). — Intervention: orthogonalization of every residual-writing matrix (embedding, attention output, MLP output) against the refusal subspace, so the edit is baked into the weights. Refusal rate is measured on held-out harmful prompts; meankl` is the mean KL divergence of next-token distributions vs. the base model on benign prompts (lower ⇒ less capability drift). ASR (if present) is the judged attack-success rate on a harmful benchmark. Research into how safety behaviour is represented in language models, red-teaming, and building better defenses. Studying the robustness and…
Модальности:
Генерация текста
Области применения:
Диалог / чат
Задача: Генерация текста
Автор: ai-anytime
Теги: qwen2, abliteration, directional-ablation, uncensored, mechanistic-interpretability, safety-research, ablate, conversational
Лайков: 3 | Загрузок: 60
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
