Research map / Large language models / Language, safety and interpretability
SeedPrints: Fingerprints Can Even Tell Which Seed Your Large Language Model Was Trained From
Yao Tong, Haonan Wang, Siquan Li, Kenji Kawaguchi, Tianyang Hu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use initialization-bias SeedPrints test instead of post-hoc fingerprints for early pretraining | lineage verification · seed vs trained model, 12-layer LLaMA-style | lineage detection: baselines 0.580 PCS / 0.375 REEF below 0.8 threshold vs p=3.37e{-}26 | Improved |
| test SeedPrints across different init seeds with identical data and order (ablation) | SeedPrints · cross-seed pairs on OpenWebText | p-value: p-values 0.109 to 0.883 (>0.01, distinguishes seeds) | Improved |
| continual pretraining on code data (The Stack) distribution shift (ablation) | baselines (Intrinsic, REEF, PCS, ICS) · continual training from seed-1000 base | lineage detection: Intrinsic 0.489, REEF 0.557, PCS 0.585, ICS 0.123 on true descendant | Hurt |
| continual pretraining on The Stack for SeedPrints (ablation) | SeedPrints · continual training from seed-1000 base | lineage detection: p approximately 0 for true descendant | No clear effect |
| evaluate on early OLMo-2-7B checkpoints (5B tokens onward) | baselines vs SeedPrints · OLMo-2-7B Stage 1 vs final checkpoint | lineage detection: baselines ICS~0, PCS<0.3, REEF<0.6 in first trillion tokens; SeedPrints p<<0.01 | Improved |
| large-scale finetuning up to 500B-700B tokens | PCS and ICS baselines · Llama-2-7B to CodeLlama-7B / Llemma-7B | similarity score: ICS 0.3369 (CodeLlama), 0.2905 (Llemma) | Hurt |
| weight-space merging with unrelated models | SeedPrints · LeaFBench Merge | AUC: 1.000 (Instruct) -> 0.959 (Merge) | Hurt |
| increase number of random input sequences n (ablation) | SeedPrints · seed identification | accuracy: 0.7368 -> 0.9375 | Improved |
| increase random sequence length l (ablation) | SeedPrints · seed identification | accuracy: 0.9048 -> 1.0000 | Improved |
| increase fingerprint size m to 800 (ablation) | SeedPrints · seed identification | empirical FPR: 0 -> 0.1463 | Hurt |
| use final-layer hidden states instead of logits as model output (ablation) | SeedPrints · random-input bias estimation | number of random inputs needed | Improved |
| use mean response vector instead of per-sample argmin frequency (ablation) | SeedPrints · identity dimension extraction | stability | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- REEF — REEF: Representation Encoding Fingerprints for Large Language Models (ICLR 2025) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- HuRef: HUman-REadable Fingerprint for Large Language Models (NeurIPS 2024 · similarity 0.88)
- Fingerprinting Inference Systems of Large Language Models (NeurIPS 2026 · similarity 0.88)
- Scalable Fingerprinting of Large Language Models (NeurIPS 2025 · similarity 0.88)
- Measuring Fingerprints of Web-filtered Text Datasets and Fingerprint Propagation Through Training (NeurIPS 2025 · similarity 0.87)
- AttnDiff: Attention-based Differential Fingerprinting for Large Language Models (NeurIPS 2026 · similarity 0.87)
- REEF: Representation Encoding Fingerprints for Large Language Models (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
