Research map / Large language models / Instruction and fine-tuning
AWM: Accurate Weight-Matrix Fingerprint for Large Language Models
Boyi Zeng, Lin Chen, Ziwei He, Xinbing Wang, Zhouhan Lin
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 unbiased instead of biased CKA (linear kernel) (ablation) | AWM · CKA variant ablation (SFT) | average absolute Z-score: 18.3099 -> 356.0223 | Improved |
| use unbiased instead of biased CKA (linear kernel) (ablation) | AWM · CKA variant ablation (Continual pretraining) | average absolute Z-score: 11.2954 -> 217.5003 | Improved |
| use unbiased instead of biased CKA (RBF kernel) (ablation) | AWM · CKA variant ablation (SFT) | average absolute Z-score: 11.5843 -> 527.3949 | Improved |
| use RBF instead of linear kernel (unbiased CKA) (ablation) | AWM · CKA variant ablation (SFT) | average absolute Z-score: 356.0223 -> 527.3949 | Improved |
| use RBF instead of linear kernel (unbiased CKA) (ablation) | AWM · CKA variant ablation (Continual pretraining) | average absolute Z-score: 217.5003 -> 320.9173 | Improved |
| use AWM instead of HuRef | HuRef · 150 LLM pairs, all categories | AUC: 0.957 -> 1.000 | Improved |
| use AWM instead of REEF | REEF · 150 LLM pairs, all categories | AUC: 0.778 -> 1.000 | Improved |
| use AWM instead of HuRef under continual pretraining | HuRef · Continual pretraining pairs | TPR@1%FPR: 0.500 -> 1.000 | Improved |
| use AWM instead of REEF under upcycling | REEF · Upcycling pairs | TPR@1%FPR: 0.000 -> 1.000 | Improved |
| use AWM instead of HuRef on unrelated 7B pairs | HuRef · Independent 7B LLM pairs | mean similarity (%): 3.56 -> 0.49 | Improved |
| use AWM instead of REEF on unrelated 7B pairs | REEF · Independent 7B LLM pairs | mean similarity (%): 42.47 -> 0.49 | Improved |
| use AWM instead of HuRef under pruning | HuRef · Pruning pairs | mean absolute Z-score: 13.166 -> 267.233 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- REEF — REEF: Representation Encoding Fingerprints for Large Language Models (ICLR 2025) · better on 7 of 7 shared results
- HuRef — HuRef: HUman-REadable Fingerprint for Large Language Models (NeurIPS 2024) · better on 7 of 7 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 0.90)
- Scalable Fingerprinting of Large Language Models (NeurIPS 2025 · similarity 0.87)
- REEF: Representation Encoding Fingerprints for Large Language Models (ICLR 2025 · similarity 0.87)
- AttnDiff: Attention-based Differential Fingerprinting for Large Language Models (NeurIPS 2026 · similarity 0.87)
- Fingerprinting Inference Systems of Large Language Models (NeurIPS 2026 · similarity 0.86)
- Fail-Closed Alignment for Large Language Models (NeurIPS 2026 · similarity 0.85)
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.
