Retrospective Distillation Attribution via Normalized Response Similarity
Abstract
Model distillation transfers capabilities through supervised fine-tuning (SFT) on teacher responses, often collected from commercial APIs, raising questions of model provenance. Existing distillation attribution methods have been largely evaluated on students immediately after the SFT step. However, a distilled model may undergo further SFT, preference optimization, or reinforcement learning before release, while an auditor may lack access to the pre-distillation checkpoint required by reference-based attribution. To close this gap, we propose SCOUT, an output-only method that aggregates recurring *syntactic patterns* into candidate profiles, filters low-contrast patterns, and calibrates student--candidate distances against inter-candidate distances. SCOUT supports attribution and abstention using only current texts, without model weights, token likelihoods, or historical checkpoints. Auditing publicly released descendants of distilled models spanning diverse post-training objectives, SCOUT consistently identifies the distillation source. Furthermore, tracing teacher-associated *syntactic signatures* along training trajectories reveals that they emerge during distillation and persist through subsequent preference optimization and reinforcement learning.
Related
- Retrospective Causal Attribution under Case-Control Sampling
- Normalized solutions to mass supercritical Schrödinger equations with radial potentials
- Normalized Orbit Frames and Essential Spectrum
- Gaussian Normalized Coordinates and Risk-Neutral CDF Deformations
- Diffusion Distillation for Efficient Weather Ensembles
BibTeX
@misc{jang2026retrospective,
title = {{Retrospective Distillation Attribution via Normalized Response Similarity}},
author = {Minwoo Jang and Jaechang Kim and Minhyeon Oh and Jeongyeon Hwang and Jungseul Ok},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.32749},
eprint = {2609.32749},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.32749},
}
RIS
TY - UNPB
TI - Retrospective Distillation Attribution via Normalized Response Similarity
AU - Minwoo Jang
AU - Jaechang Kim
AU - Minhyeon Oh
AU - Jeongyeon Hwang
AU - Jungseul Ok
PY - 2026
DA - 2026/09/26
PB - arXiv
DO - 10.48550/arxiv.2609.32749
UR - https://arxiv.org/abs/2609.32749
AB - Model distillation transfers capabilities through supervised fine-tuning (SFT) on teacher responses, often collected from commercial APIs, raising questions of model provenance. Existing distillation attribution methods have been largely evaluated on students immediately after the SFT step. However, a distilled model may undergo further SFT, preference optimization, or reinforcement learning before release, while an auditor may lack access to the pre-distillation checkpoint required by reference-based attribution. To close this gap, we propose SCOUT, an output-only method that aggregates recurring *syntactic patterns* into candidate profiles, filters low-contrast patterns, and calibrates student--candidate distances against inter-candidate distances. SCOUT supports attribution and abstention using only current texts, without model weights, token likelihoods, or historical checkpoints. Auditing publicly released descendants of distilled models spanning diverse post-training objectives, SCOUT consistently identifies the distillation source. Furthermore, tracing teacher-associated *syntactic signatures* along training trajectories reveals that they emerge during distillation and persist through subsequent preference optimization and reinforcement learning.
ER -