一种随机数生成控制方法、设备、介质及分布式计算系统
By utilizing a hardware decoupling strategy between the main processor and the target computing device in the artificial intelligence model computing system, seed variables and temporary offsets are sent to generate random numbers, solving the problem of low access efficiency of different computing devices and realizing efficient utilization of computing device performance and resources.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INSPUR SUZHOU INTELLIGENT TECH CO LTD
- Filing Date
- 2024-12-31
- Publication Date
- 2026-07-17
AI Technical Summary
In the computational system of artificial intelligence models, the access efficiency of random number generators is low, making it difficult to fully utilize the computing performance of different computing devices. This is especially true in distributed systems with multiple computing devices, where there is a lack of unified management and efficient access solutions.
By responding to the random number generator call command of the main processor's response model computing framework, the target computing device is determined, and a seed variable and a temporary offset are sent to the target computing device to generate random numbers. The system also receives and updates the status information of the main processor's random number generator, thus implementing a hardware-decoupled random number generator access strategy.
It improves the access efficiency of random number generators on different computing devices, makes full use of the computing performance of computing devices, reduces the difficulty of hardware adaptation, and improves the scalability and portability of model computing system.
Smart Images
Figure CN119718254B_ABST