模型的融合方法、装置和计算机可读存储介质
By adjusting the model through alignment parameters, a third model is generated, which solves the problems of high computational cost and insufficient accuracy in model fusion in existing technologies, and achieves efficient and accurate model fusion results.
CN116644382BActive Publication Date: 2026-07-17JD DIGITS HAIYI INFORMATION TECHNOLOGY CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JD DIGITS HAIYI INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2023-06-01
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing model fusion methods are insufficient in terms of computational cost and accuracy, making it difficult to perform model fusion efficiently and accurately.
Method used
By determining the alignment parameters of models from different training methods, the first model is adjusted using the alignment parameters to generate the third model, which is then fine-tuned using training data to achieve efficient model fusion.
Benefits of technology
It improves model performance, reduces inference computation costs, and increases computational efficiency.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116644382B_ABST
Abstract
本发明公开了一种模型的融合方法、装置和计算机可读存储介质，涉及机器学习领域。模型的融合方法包括：根据来自第一模型的第一参数矩阵和来自第二模型的第二参数矩阵，确定对齐参数，其中，第一模型和第二模型具有相同的模型结构、但通过不同的方式训练得到，第一参数矩阵和第二参数矩阵分别为第一模型和第二模型中相同参数的值构成的矩阵；利用对齐参数对第一参数矩阵进行调整，获得第三参数矩阵；根据第二参数矩阵和第三参数矩阵，确定第三模型，其中，第三模型具有模型结构。从而，可以基于对齐的模型进行模型的融合，使得融合后的模型兼有不同训练方式的优点，提高了模型的性能。并且，推理计算成本较低，具有较高的计算效率。
Need to check novelty before this filing date? Find Prior Art