analiza aiqbench
This research introduces a novel method for tracing Large Language Model lineage by analyzing intrinsic geometric patterns within model weight spaces, effectively acting as "biometrics" for AI provenance. By leveraging spectral energy and subspace alignment, the study demonstrates how to distinguish between independently trained models and fine-grained variations arising from different datasets or post-training procedures without requiring access to input data. This work significantly advances supply-chain integrity and governance by providing a robust framework to verify model origins and ownership in the open-weight ecosystem.
Who Built This Model? Tracing LLM Lineage via Spectral Fingerprints in Weight Space
arXiv:2608.07786v2 Announce Type: replace-cross Abstract: Open-weight large language models (LLMs) are increasingly developed through complex, multi-stage pipelines, leading to intricate lineage relationships that reflect model origin, ownership, and evolution. Understanding these relationships is important for model provenance, governance, and supply-chain integrity. In this work, we investigate the notion of LLM "biometrics" (analogous to human biometrics) to ask whether LLMs exhibit intrinsic fingerprints in weight space alone, without access to input data, that reveal their origin and lineage. We formulate this as a lineage discrimination problem, distinguishing among independent-origin, same-series, and shared-base models. To characterize these relationships, we propose a unified geometric fingerprinting framework that analyzes weight matrices from two complementary perspectives: (i) spectral energy, captured by singular value distributions to encode global magnitude patterns, and (ii) subspace alignment, quantified via subspace deviations to capture directional geometry. Our analysis uncovers a clear hierarchy of structural similarity in weight space: spectral energy reliably distinguishes independently trained models and different model families, while subspace alignment enables fine-grained discrimination among closely related models, including variations in dataset scale and post-training procedures. Extensive experiments on over 110 diverse open-weight LLM pairs demonstrate that weight-space geometry provides a robust and interpretable signal for model lineage, enabling coarse-grained regime separation and fine-grained discrimination within shared-base models.
Title:Who Built This Model? Tracing LLM Lineage via Spectral Fingerprints in Weight Space
Abstract:Open-weight large language models (LLMs) are increasingly developed through complex, multi-stage pipelines, leading to intricate lineage relationships that reflect model origin, ownership, and evolution. Understanding these relationships is important for model provenance, governance, and supply-chain integrity. In this work, we investigate the notion of LLM "biometrics" (analogous to human biometrics) to ask whether LLMs exhibit intrinsic fingerprints in weight space alone, without access to input data, that reveal their origin and lineage. We formulate this as a lineage discrimination problem, distinguishing among independent-origin, same-series, and shared-base models. To characterize these relationships, we propose a unified geometric fingerprinting framework that analyzes weight matrices from two complementary perspectives: (i) spectral energy, captured by singular value distributions to encode global magnitude patterns, and (ii) subspace alignment, quantified via subspace deviations to capture directional geometry. Our analysis uncovers a clear hierarchy of structural similarity in weight space: spectral energy reliably distinguishes independently trained models and different model families, while subspace alignment enables fine-grained discrimination among closely related models, including variations in dataset scale and post-training procedures. Extensive experiments on over 110 diverse open-weight LLM pairs demonstrate that weight-space geometry provides a robust and interpretable signal for model lineage, enabling coarse-grained regime separation and fine-grained discrimination within shared-base models.
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2608.07786 [cs.AI]
(or arXiv:2608.07786v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2608.07786 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 312 cuvinte
Subiecte conexe
Physics-Based Molecular Fingerprints from Spectral Graph Theory Provide Efficient Geometry-Aware Measures of Chemical Similarity
arXiv · 7 aug. 2026
Distinguishing AI-Generated Music from Edited Audio as a Hard-Negative Robustness Task
arXiv · 19 aug. 2026
Spectral fingerprint recognition of mobile signals via Rydberg atomic sensors
arXiv · 28 oct. 2026
Spectral symmetry fingerprints in equivariant representations
arXiv · 6 nov. 2026