arXiv · cs.AI · Alessandro Zirilli, Davide Marincione, Evgenios M. Kornaropoulos, Giuseppe Ateniese, Emanuele Rodol\`a · 5 小时前模型评分40
Perseus：面向同态模型推理的全栈开源系统
Perseus: A Full-Stack System for Homomorphic Model Inference
导读
Perseus 是一个免费开源的 Python 软件库，支持从密码学后端、数据打包到加密非线性与端到端模型执行的加密模型推理全栈。它在 Llama-3-8B 上单 token 解码比任何其他实现快 3.96 倍，精度与困惑度与明文模型相差在 3% 以内，并首次发起加密 GPT-2 的公开竞速。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
事件类型：研究发布
证据摘录
- “We present Perseus, a free and open Python software library that supports the full stack of encrypted model inference, from the cryptographic backend and data packing, to encrypted nonlinearities and end-to-end model execution.”
- “Perseus achieves state-of-the-art latency across multiple models and modalities, within $3\%$ of the plaintext models' accuracy and perplexity.”
- “Remarkably, it is $3.96\times$ faster at single-token decoding than any other implementation on Llama-3-8B.”
来源：arXiv · cs.AI · arxiv.org