Newest papers

Model-intrinsic, Weight-level, and Training-free Spectral Fingerprinting for Large Language Models

Yanan He, Yanli Chen, Ling Xiong, Gejian Zhao, Hanzhou Wu · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.30626 · cited by 0 works (OpenAlex) · open-access PDF

Indonesian

Seluruh keluarga verifikasi fingerprinting yang ada biasanya membangun bukti dari tanggapan kueri, output yang dihasilkan, atau aktivasi input yang diinduksi. Meskipun efektif dalam pengaturan berbasis representasi, mereka dapat dipengaruhi oleh pemilihan prompt, konfigurasi decoding, lingkungan inferensi, dan implementasi forward-pass, yang dapat membatasi aplikasinya. Di sini kami memperkenalkan kerangka kerja tingkat berat intrinsik model yang menghasilkan sidik jari kompak dari struktur spektral yang dikodekan pada parameter LLM. Secara khusus, kerangka kerja membangun fitur spektral bukti GATE dari distribusi nilai tunggal dari proyeksi GATE, yang membedakan beban dalam jaringan g-feed-forward (FFNs), dan mengagregasi fitur-fitur sampel yang dihasilkan melalui tuning layer-wise yang serupa melalui pelapisan eksperimental untuk membentuk lapisan bebas

Machine translation (Meta M2M100); the English original follows.

Abstract

The increasing release and reuse of Large Language Models (LLMs) as open-weight checkpoints make model fingerprinting important for ownership protection and offline auditing. Existing fingerprinting methods typically construct evidence from query responses, generated outputs, or input-induced activations. Although effective in representation-based settings, they can be affected by prompt selection, decoding configuration, inference environment, and forward-pass implementation, which may limit their applicability. Here we introduce a model-intrinsic weight-level framework that derives compact fingerprints from spectral structures encoded in LLM parameters. Specifically, the framework constructs GATE spectral scalar features from the singular-value distributions of GATE projection weights in gated feed-forward networks (FFNs), and aggregates the resulting layer-wise features through depth-wise segmented pooling to form a model-level fingerprint. Such representation preserves coarse layer-depth organization while reducing dependence on strict layer-to-layer alignment, enabling comparison across LLMs with different depths. The similarity between fingerprints is computed by linear centered kernel alignment. The entire pipeline is training-free and does not require input samples, generated outputs, forward activations, or an auxiliary verifier. Experiments on representative LLM families and post-training variants have demonstrated a clear separation between source-related models and unrelated ones, and stable similarity under instruction tuning, quantization, pruning, and distillation-related modifications, which verify that spectral features are robust evidence for LLM fingerprinting.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Decoding methods, Pipeline (software), Inference, Pooling, Artificial intelligence

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange