Power-Softmax: Towards Secure LLM Inference over Encrypted Data
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910191093547008 |
|---|---|
| author | Zimerman, Itamar Adir, Allon Aharoni, Ehud Avitan, Matan Baruch, Moran Drucker, Nir Lerner, Jenny Masalha, Ramy Meiri, Reut Soceanu, Omri |
| author_facet | Zimerman, Itamar Adir, Allon Aharoni, Ehud Avitan, Matan Baruch, Moran Drucker, Nir Lerner, Jenny Masalha, Ramy Meiri, Reut Soceanu, Omri |
| contents | Modern cryptographic methods for implementing privacy-preserving LLMs such as \gls{HE} require the LLMs to have a polynomial form. Forming such a representation is challenging because transformers include non-polynomial components, such as \Softmax and layer normalization. Previous approaches have either directly approximated pre-trained models with large-degree polynomials, which are less efficient over HE, or replaced non-polynomial components with easier-to-approximate primitives before training, e.g., \Softmax with pointwise attention. The latter approach might introduce scalability challenges. We present a new HE-friendly variant of self-attention that offers a stable form for training and is easy to approximate with polynomials for secure inference. Our work introduces the first polynomial LLMs over a billion parameters, exceeding the size of previous models by more than tenfold. The resulting models demonstrate reasoning and in-context learning (ICL) capabilities comparable to standard transformers of the same size, representing a breakthrough in the field. Finally, we provide a detailed latency breakdown for each computation over encrypted data, paving the way for further optimization, and explore the differences in inductive bias between models relying on our HE-friendly variant and standard transformers. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_09457 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Power-Softmax: Towards Secure LLM Inference over Encrypted Data Zimerman, Itamar Adir, Allon Aharoni, Ehud Avitan, Matan Baruch, Moran Drucker, Nir Lerner, Jenny Masalha, Ramy Meiri, Reut Soceanu, Omri Machine Learning Cryptography and Security F.2.2; I.2.7 Modern cryptographic methods for implementing privacy-preserving LLMs such as \gls{HE} require the LLMs to have a polynomial form. Forming such a representation is challenging because transformers include non-polynomial components, such as \Softmax and layer normalization. Previous approaches have either directly approximated pre-trained models with large-degree polynomials, which are less efficient over HE, or replaced non-polynomial components with easier-to-approximate primitives before training, e.g., \Softmax with pointwise attention. The latter approach might introduce scalability challenges. We present a new HE-friendly variant of self-attention that offers a stable form for training and is easy to approximate with polynomials for secure inference. Our work introduces the first polynomial LLMs over a billion parameters, exceeding the size of previous models by more than tenfold. The resulting models demonstrate reasoning and in-context learning (ICL) capabilities comparable to standard transformers of the same size, representing a breakthrough in the field. Finally, we provide a detailed latency breakdown for each computation over encrypted data, paving the way for further optimization, and explore the differences in inductive bias between models relying on our HE-friendly variant and standard transformers. |
| title | Power-Softmax: Towards Secure LLM Inference over Encrypted Data |
| topic | Machine Learning Cryptography and Security F.2.2; I.2.7 |
| url | https://arxiv.org/abs/2410.09457 |