Calculation method and calculation equipment for large model

By splitting large model computation between TEE and non-TEE regions and obfuscating inputs and outputs, the method ensures security and privacy while maintaining inference speed, addressing the challenges of deploying large language models in TEEs.

HK40134557APending Publication Date: 2026-07-10ANT BLOCKCHAIN TECHNOLOGY (SHANGHAI) CO LTD

Patent Information

Authority / Receiving Office
HK · HK
Patent Type
Applications
Current Assignee / Owner
ANT BLOCKCHAIN TECHNOLOGY (SHANGHAI) CO LTD
Filing Date
2026-04-20
Publication Date
2026-07-10

Smart Images

  • Figure 00000000_0000_ABST
    Figure 00000000_0000_ABST
Patent Text Reader

Abstract

The invention discloses a large model calculation method and calculation equipment, the method is applied to equipment with a TEE area and a non-TEE area, the method comprises the steps that in the TEE area, an input matrix is acquired, and the input matrix is obtained by performing embedding processing on input data; in the TEE region, multiplying the input matrix by a first reversible matrix to obtain a confusion input matrix; in the non-TEE region, reasoning calculation is carried out on the confusion input matrix based on a confusion function block to obtain a confusion output matrix, and the confusion function block is obtained by confusing function blocks of the large model based on an inverse matrix of a first reversible matrix; and in the TEE region, carrying out de-confusion processing on the confusion output matrix to obtain a reasoning result corresponding to the input data. According to the method, the reasoning speed of the large model can be guaranteed while the model security and the user privacy are protected.
Need to check novelty before this filing date? Find Prior Art