Have You Merged My Model? On The Robustness of Large Language Model IP Protection Methods Against Model Merging
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866909374913445888 |
|---|---|
| author | Cong, Tianshuo Ran, Delong Liu, Zesen He, Xinlei Liu, Jinyuan Gong, Yichen Li, Qi Wang, Anyu Wang, Xiaoyun |
| author_facet | Cong, Tianshuo Ran, Delong Liu, Zesen He, Xinlei Liu, Jinyuan Gong, Yichen Li, Qi Wang, Anyu Wang, Xiaoyun |
| contents | Model merging is a promising lightweight model empowerment technique that does not rely on expensive computing devices (e.g., GPUs) or require the collection of specific training data. Instead, it involves editing different upstream model parameters to absorb their downstream task capabilities. However, uncertified model merging can infringe upon the Intellectual Property (IP) rights of the original upstream models. In this paper, we conduct the first study on the robustness of IP protection methods under model merging scenarios. Specifically, we investigate two state-of-the-art IP protection techniques: Quantization Watermarking and Instructional Fingerprint, along with various advanced model merging technologies, such as Task Arithmetic, TIES-MERGING, and so on. Experimental results indicate that current Large Language Model (LLM) watermarking techniques cannot survive in the merged models, whereas model fingerprinting techniques can. Our research aims to highlight that model merging should be an indispensable consideration in the robustness assessment of model IP protection techniques, thereby promoting the healthy development of the open-source LLM community. Our code is available at https://github.com/ThuCCSLab/MergeGuard. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2404_05188 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Have You Merged My Model? On The Robustness of Large Language Model IP Protection Methods Against Model Merging Cong, Tianshuo Ran, Delong Liu, Zesen He, Xinlei Liu, Jinyuan Gong, Yichen Li, Qi Wang, Anyu Wang, Xiaoyun Cryptography and Security Artificial Intelligence Computation and Language Model merging is a promising lightweight model empowerment technique that does not rely on expensive computing devices (e.g., GPUs) or require the collection of specific training data. Instead, it involves editing different upstream model parameters to absorb their downstream task capabilities. However, uncertified model merging can infringe upon the Intellectual Property (IP) rights of the original upstream models. In this paper, we conduct the first study on the robustness of IP protection methods under model merging scenarios. Specifically, we investigate two state-of-the-art IP protection techniques: Quantization Watermarking and Instructional Fingerprint, along with various advanced model merging technologies, such as Task Arithmetic, TIES-MERGING, and so on. Experimental results indicate that current Large Language Model (LLM) watermarking techniques cannot survive in the merged models, whereas model fingerprinting techniques can. Our research aims to highlight that model merging should be an indispensable consideration in the robustness assessment of model IP protection techniques, thereby promoting the healthy development of the open-source LLM community. Our code is available at https://github.com/ThuCCSLab/MergeGuard. |
| title | Have You Merged My Model? On The Robustness of Large Language Model IP Protection Methods Against Model Merging |
| topic | Cryptography and Security Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2404.05188 |