aiqbench analysis
This research introduces a novel method for verifying the lineage of open-weight language models by analyzing residual structures within model weights, effectively distinguishing fine-tuned descendants from independent or distilled variants without requiring training data. The proposed symmetric lineage score achieves perfect discrimination on benchmarks and remains robust against checkpoint laundering attacks while operating significantly faster than existing baselines. These findings address a critical gap in documenting the provenance of rapidly evolving AI models as they are merged, pruned, and quantized across different ecosystems.
Training Leaves Traces: Centered Residual Signatures for Language Model Lineage Verification
arXiv:2608.14929v1 Announce Type: cross Abstract: Open-weight language models are fine-tuned, quantized, pruned, and merged, yet their provenance is often undocumented. We study data-free white-box lineage verification: can weights alone reveal whether two compatible model checkpoints share ancestry? Residual training produces a shared identity-aligned component in branch products, so this structure alone cannot establish ancestry. We remove it and compare checkpoint-specific structure across residual blocks, yielding a symmetric lineage score calibrated against independent checkpoints. On residual-MLP and GPT-2 benchmarks, the score separates fine-tuned, LoRA-merged, pruned, and quantized descendants from independent and distilled models (AUROC=1.0), distinguishing weight ancestry from behavioral similarity. Under function-preserving checkpoint laundering experiments, weight-space baselines lose margin or fail; our score remains unchanged and runs 76x faster than the nearest robust baseline on GPT-2. The projection-pairing signal appears across six language-model families and beyond, and a case study correctly identifies 3 related and 7 unrelated LLaMA-2 public checkpoints. Collectively, these results establish a passive, data-free provenance signal for compatible open-weight language-model checkpoints
Abstract:Open-weight language models are fine-tuned, quantized, pruned, and merged, yet their provenance is often undocumented. We study data-free white-box lineage verification: can weights alone reveal whether two compatible model checkpoints share ancestry? Residual training produces a shared identity-aligned component in branch products, so this structure alone cannot establish ancestry. We remove it and compare checkpoint-specific structure across residual blocks, yielding a symmetric lineage score calibrated against independent checkpoints. On residual-MLP and GPT-2 benchmarks, the score separates fine-tuned, LoRA-merged, pruned, and quantized descendants from independent and distilled models (AUROC=1.0), distinguishing weight ancestry from behavioral similarity. Under function-preserving checkpoint laundering experiments, weight-space baselines lose margin or fail; our score remains unchanged and runs 76x faster than the nearest robust baseline on GPT-2. The projection-pairing signal appears across six language-model families and beyond, and a case study correctly identifies 3 related and 7 unrelated LLaMA-2 public checkpoints. Collectively, these results establish a passive, data-free provenance signal for compatible open-weight language-model checkpoints
Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2608.14929 [cs.CL]
(or arXiv:2608.14929v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2608.14929 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Full text under the publisher's open licence · 256 words
Related coverage
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026
Low-Background Minds
LessWrong · Oct 3, 2026
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
METR · Aug 26, 2026
Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline
Hugging Face · Oct 16, 2026