A test method, apparatus, medium, and product
By running the AI model multiple times in the target operating environment and adjusting the concurrency and latency parameters, the low accuracy problem caused by the reliance on static metrics in existing testing systems is solved, and more accurate performance evaluation is achieved.
CN122195836APending Publication Date: 2026-06-12CHINA MOBILE GROUP SHANDONG +1
View PDF 0 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA MOBILE GROUP SHANDONG
- Filing Date
- 2026-03-12
- Publication Date
- 2026-06-12
AI Technical Summary
Technical Problem
Existing AI model testing systems rely on static indicator data, which cannot accurately reflect the performance of AI models in real-world applications, resulting in low accuracy of test results.
Method used
By deploying the target runtime environment of the model under test, running the model multiple times, determining the state range based on the load test concurrency and performance parameters, adjusting the concurrency to determine the target load test concurrency, and combining the first word delay and inter-word delay to determine the test results.
Benefits of technology
It improves the accuracy of AI model performance testing, enabling more accurate evaluation of the model's performance in real-world applications.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122195836A_ABST
Abstract
This invention discloses a testing method, device, medium, and product. Relating to the field of computer technology, the method includes: deploying a target runtime environment corresponding to a model under test; running the model under test multiple times in the target runtime environment; for any given run, determining the model performance parameters of the model under test corresponding to the current run based on a first load testing concurrency; determining a current state interval based on the model performance parameters of the model under test corresponding to the current run; determining the load testing concurrency corresponding to the next run based on the current state interval and the first load testing concurrency; using the load testing concurrency corresponding to the next run when the model performance parameters of the model under test corresponding to the current run are a first value as a target load testing concurrency; and determining the test result corresponding to the model under test based on the target load testing concurrency and the first-word delay and inter-word delay corresponding to the target load testing concurrency.
Need to check novelty before this filing date? Find Prior Art