DOGe: Defensive Output Generation for LLM Protection Against Knowledge Distillation

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Li, Pingzhi, Tan, Zhen, Zhang, Mohan, Qu, Huaizhi, Liu, Huan, Chen, Tianlong
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915562571956224
author Li, Pingzhi
Tan, Zhen
Zhang, Mohan
Qu, Huaizhi
Liu, Huan
Chen, Tianlong
author_facet Li, Pingzhi
Tan, Zhen
Zhang, Mohan
Qu, Huaizhi
Liu, Huan
Chen, Tianlong
contents Large Language Models (LLMs) represent substantial intellectual and economic investments, yet their effectiveness can inadvertently facilitate model imitation via knowledge distillation (KD). In practical scenarios, competitors can distill proprietary LLM capabilities by simply observing publicly accessible outputs, akin to reverse-engineering a complex performance by observation alone. Existing protective methods like watermarking only identify imitation post-hoc, while other defenses assume the student model mimics the teacher's internal logits, rendering them ineffective against distillation purely from observed output text. This paper confronts the challenge of actively protecting LLMs within the realistic constraints of API-based access. We introduce an effective and efficient Defensive Output Generation (DOGe) strategy that subtly modifies the output behavior of an LLM. Its outputs are accurate and useful for legitimate users, yet are designed to be misleading for distillation, significantly undermining imitation attempts. We achieve this by fine-tuning only the final linear layer of the teacher LLM with an adversarial loss. This targeted training approach anticipates and disrupts distillation attempts during inference time. Our experiments show that, while preserving the performance of the teacher model, student models distilled from the defensively generated outputs demonstrate catastrophically reduced performance, demonstrating DOGe as a practical safeguard against KD-based model imitation.
format Preprint
id arxiv_https___arxiv_org_abs_2505_19504
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DOGe: Defensive Output Generation for LLM Protection Against Knowledge Distillation
Li, Pingzhi
Tan, Zhen
Zhang, Mohan
Qu, Huaizhi
Liu, Huan
Chen, Tianlong
Machine Learning
Artificial Intelligence
Computation and Language
Large Language Models (LLMs) represent substantial intellectual and economic investments, yet their effectiveness can inadvertently facilitate model imitation via knowledge distillation (KD). In practical scenarios, competitors can distill proprietary LLM capabilities by simply observing publicly accessible outputs, akin to reverse-engineering a complex performance by observation alone. Existing protective methods like watermarking only identify imitation post-hoc, while other defenses assume the student model mimics the teacher's internal logits, rendering them ineffective against distillation purely from observed output text. This paper confronts the challenge of actively protecting LLMs within the realistic constraints of API-based access. We introduce an effective and efficient Defensive Output Generation (DOGe) strategy that subtly modifies the output behavior of an LLM. Its outputs are accurate and useful for legitimate users, yet are designed to be misleading for distillation, significantly undermining imitation attempts. We achieve this by fine-tuning only the final linear layer of the teacher LLM with an adversarial loss. This targeted training approach anticipates and disrupts distillation attempts during inference time. Our experiments show that, while preserving the performance of the teacher model, student models distilled from the defensively generated outputs demonstrate catastrophically reduced performance, demonstrating DOGe as a practical safeguard against KD-based model imitation.
title DOGe: Defensive Output Generation for LLM Protection Against Knowledge Distillation
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.19504