CN-121357026-B - Network model configuration method, device and communication system

Abstract

The application provides a network model configuration method, equipment and a communication system. In the method, when the UE requests the network model from the NE, the NE can perform matching according to the first information provided by the UE, and a target model meeting the requirement of the UE is determined. After receiving the network model issued by the NE, the UE may first create a sandboxed environment in which to deploy and run the above network model, determine whether the model is deployable, and whether the output and performance of the model run meet expectations. And when the expectations are met, the UE formally deploys the models in a system environment outside the sandboxes so as to avoid system collapse caused by the fact that the target models encroach on system resources.

Inventors

• WEI JINSHENG
• SUN JIALIANG
• SHAO WENZE
• DING HANYU

Assignees

• 荣耀终端股份有限公司

Dates

Publication Date

20260508

Application Date

20251217

Claims (20)

1. 1. A model configuration method applied to a user equipment UE, the method comprising: sending a first signaling to a Network Equipment (NE), wherein the first signaling carries first information of the UE; Receiving a first model sent by the NE, wherein the first model is determined by the NE according to the first information; Creating a sandboxed environment locally, running the first model in the sandboxed environment; And when the output and the performance of the first model do not meet the expectations, sending a diagnosis report to the NE, and indicating that the first model is not successfully deployed on the UE.
2. 2. The method of claim 1, wherein the NE comprises a core network device to which the first signaling is communicated via NAS messages.
3. 3. The method according to claim 1, wherein a global model list is provided in the NE, wherein model meta information of a plurality of network models is recorded in the global model list, and wherein the first information includes a UE capability report; The first model is one of one or more candidate models determined by the NE from the model meta-information and the UE capability report.
4. 4. The method of claim 3, wherein the UE capability report includes one or more of a model type, a resolvable model format, a UE memory, UE acceleration hardware, an input-output format and dimension, a UE identity; the first model meets one or more of the following conditions that the model type is consistent with the model type requested by the UE in the UE capability report, the model format is in a format range of the UE resolvable model, the peak value is in a loadable range of the UE memory, the input and output format and the dimension are in a format range supported by the UE, the acceleration hardware of the UE meets the minimum acceleration hardware requirement of the first model, and the UE identity is in an authorized range of the first model.
5. 5. The method of claim 3, wherein the first information further comprises UE status information; The first model is the one of the one or more candidate models having the highest score and greater than a first threshold, wherein a model score is determined based on model meta-information of the model and the UE state information.
6. 6. The method of claim 5, wherein the UE status information includes one or more of current CPU utilization, current available memory, current battery power, current chip temperature, current network latency, current power consumption mode, current available acceleration hardware, and model deployment success rate.
7. 7. The method of claim 1, wherein the NE comprises a base station, and wherein prior to the creating the sandboxed environment, the method further comprises: Transmitting configuration related information to the base station; Receiving radio resource scheduling allocation sent by the base station, wherein the radio resource scheduling allocation is determined by the base station according to the configuration related information; the configuration related information includes one or more of a supported frequency band of the UE, RAT support, MIMO capability, modulation order, CA capability, coding scheme, physical layer characteristics.
8. 8. The method of claim 1, wherein the output and performance of the first model do not meet expectations when any of the following conditions are met: the average inference time IT of the first model is greater than a threshold ITt; the peak memory occupation PM of the first model is larger than a threshold PMt; The highest CPU occupancy CU of the first model is greater than a threshold CUt; the average inference energy consumption AE of the first model is greater than a threshold AEt; the performance index PI of the first model is less than a threshold PIt.
9. 9. A model configuration method applied to a communication system comprising a network equipment NE and a user equipment UE, characterized in that the method comprises: The UE sends a first signaling to the NE; The first signaling carries first information of the UE, and the NE determines a first model according to the first information; The NE transmits the first model to the UE; the UE creates a sandbox environment locally, and runs the first model in the sandbox environment; And when the output and the performance of the first model do not meet the expectations, the UE sends a diagnosis report to the NE, and the diagnosis report indicates that the first model is not successfully deployed on the UE.
10. 10. The method of claim 9, wherein the NE comprises a core network device and the first signaling is communicated via NAS messaging.
11. 11. The method according to claim 9, wherein a global model list is provided in the NE, wherein model meta information of a plurality of network models is recorded in the global model list, The NE determines a first model according to the first information, and comprises the following steps: The NE queries the global model list, determines one or more candidate models that match the first information, and determines scores for the one or more candidate models; the NE determines from the one or more candidate models that the first model has the highest score and is greater than a first threshold.
12. 12. The method of claim 11, wherein the first information comprises UE capability reports and UE status information, wherein the determining one or more candidate models that match the first information, and determining a score for the one or more candidate models, comprises: Determining the one or more candidate models from the UE capability report; and determining the scores of the one or more candidate models according to the UE state information.
13. 13. The method of claim 12, wherein the UE capability report includes one or more of a model type, a resolvable model format, a UE memory, UE acceleration hardware, input-output format and dimensions, a UE identity; The candidate model meets one or more of the following conditions that the model type is consistent with the model type requested by the UE in the UE capability report, the model format is in a format range of the UE resolvable model, the peak value is in a loadable range of the UE memory, the input and output format and the dimension are in a format range supported by the UE, the acceleration hardware of the UE meets the minimum acceleration hardware requirement of the candidate model, and the UE identity is in an authorized range of the candidate model.
14. 14. The method of claim 12, wherein the UE status information includes one or more of current CPU utilization, current available memory, current battery power, current chip temperature, current network latency, current power consumption mode, current available acceleration hardware, and model deployment success rate.
15. 15. The method of claim 9, wherein the NE transmitting the first model to the UE comprises: the NE determines a first server for storing the first model according to the model meta-information of the first model; The NE acquires the first model from a first server and sends the first model to the UE.
16. 16. The method of claim 9, wherein the NE comprises a base station, and wherein the method further comprises, prior to the UE creating the sandboxed environment: the UE sends configuration related information to the base station; The base station determines the radio resource scheduling allocation of the UE according to the configuration related information; the configuration related information includes one or more of a supported frequency band of the UE, RAT support, MIMO capability, modulation order, CA capability, coding scheme, physical layer characteristics.
17. 17. The method of claim 9, wherein the output and performance of the first model do not meet expectations when any of the following conditions are met: the average inference time IT of the first model is greater than a threshold ITt; the peak memory occupation PM of the first model is larger than a threshold PMt; The highest CPU occupancy CU of the first model is greater than a threshold CUt; the average inference energy consumption AE of the first model is greater than a threshold AEt; the performance index PI of the first model is less than a threshold PIt.
18. 18. A user equipment comprising one or more processors and one or more memories, wherein the one or more memories are coupled to the one or more processors, the one or more memories for storing a computer program which, when executed by the one or more processors, causes the method of any of claims 1-8 to be performed.
19. 19. A communication system comprising a network device and a user device as claimed in any of the claims 9-17.
20. 20. A computer program product comprising instructions which, when run on a user equipment, cause the user equipment to perform the method of any of claims 1-8.

Description

Network model configuration method, device and communication system Technical Field The present application relates to the field of terminals, and in particular, to a network model configuration method, device, and communication system. Background With the trend of artificial intelligence (ARTIFICIAL INTELLIGENC, AI) and machine learning (MACHINE LEARNING, ML) technologies toward large-scale enterprise applications, a single vendor or technology stack has failed to meet the requirements for AI/ML model diversity, expertise, and iteration speed in complex business scenarios. Thus, multi-vendor collaboration is a necessary trend. In a multi-vendor environment, the AI/ML model is shared, invoked and deployed among network entities through model identification. At present, compatibility detection is used as a core step of model identification, and the cooperation effect of multiple manufacturers is severely restricted. Disclosure of Invention The application provides a network model configuration method, equipment and a communication system. In a first aspect, the application provides a model configuration method applied to User Equipment (UE), the method comprises the steps of sending a first signaling to Network Equipment (NE), wherein the first signaling carries first information of the UE, receiving a first model sent by the NE, determining the first model by the NE according to the first information, creating a sandbox environment, running the first model in the sandbox environment, and running the first model in a system environment outside the sandbox environment when output and performance of the first model meet expectations. Implementing the method provided in the first aspect, after the UE requests a specific network model from the NE and receives the network model issued by the NE, the UE may first create a sandbox environment, deploy and run the network model in the sandbox environment, determine whether the model is deployable, and whether the output and performance of the model run meet expectations. And when the expectations are met, the UE formally deploys the models in a system environment outside the sandboxes so as to avoid system collapse caused by the fact that the target models encroach on system resources. The NE includes a core network device. In some embodiments, the first signaling is communicated to the core network device through NAS messages. The NE is provided with a global model list, and model meta-information of a plurality of network models is recorded in the global model list. In some embodiments, the first information includes a UE capability report, and the first model is one of one or more candidate models determined by the NE from the model meta information and the UE capability report. In some embodiments, the UE capability report includes one or more of a model type, a resolvable model format, a UE memory, UE acceleration hardware, an input output format and a dimension, and a UE identity, wherein the first model satisfies one or more of a model type consistent with a model type requested by the UE in the UE capability report, the model format is within the range of the UE resolvable model format, the peak value is within the range of the UE memory, the input output format and the dimension are within the range of the format supported by the UE, the acceleration hardware of the UE satisfies the minimum acceleration hardware requirement of the first model, and the UE identity is within the authorized range of the first model. In some embodiments, the first information further comprises UE state information, the first model is the one of the one or more candidate models that has a highest score and is greater than a first threshold, wherein the model score is determined from model meta-information of the model and the UE state information. In some embodiments, the UE status information includes one or more of current CPU utilization, current available memory, current battery power, current chip temperature, current network latency, current power consumption mode, current available acceleration hardware, and model deployment success rate. In some embodiments, the NE comprises a base station, and before creating the sandboxed environment, the method further comprises sending configuration related information to the base station, receiving radio resource scheduling assignment sent by the base station, wherein the radio resource scheduling assignment is determined by the base station according to the configuration related information, and the configuration related information comprises one or more of supporting frequency bands, RAT support, MIMO capability, modulation order, CA capability, coding scheme and physical layer characteristics of the UE. In some embodiments, the method further comprises sending a diagnostic report to the NE indicating that the first model was not successfully deployed on the UE when the output and performance of the first model do not meet the expe