A model training method, device, equipment, medium and program product

Abstract

The application provides a model training method and device, equipment, medium and program product, and relates to the technical field of wireless communication. The method comprises the following steps: a first request sent by a first network element is acquired, the first request is used for requesting a model training service producer to establish a training task; a second request is sent to a second network element or a third network element according to the first request, the second request is used for querying a target model training service producer; an identifier of the target model training service producer sent by the second network element or the third network element is acquired; and a third request is sent to the target model training service producer, the third request is used for requesting the target model training service producer to establish a training task. According to the scheme, the model can be cooperatively trained between network elements.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, specifically to a model training method, apparatus, device, medium, and program product. Background Technology

[0002] In the O-RAN system, the main modules that can embody intelligent functions are: Non-RT RIC (Non Real Time Radio Intelligent Controller), also known as non-real-time RIC, and Near-RT RIC (Near Real Time Radio Intelligent Controller), also known as near-real-time RIC.

[0003] Non-RT RICs are primarily responsible for non-real-time intelligent radio resource management, high-level program optimization, and policy optimization, while Near-RT RICs are responsible for near-real-time radio resource management. In non-real-time RICs, AI / ML workflow functions and rApps (Non-RT RIC Applications) can provide internal training services, with other rApps serving as consumers. In near-real-time RICs, AI / ML support functions can provide internal training services, with xApps (Near-RT RIC Applications) serving as consumers. The A1 interface acts as a communication bridge between non-real-time and near-real-time RICs, supporting policy-based near-real-time RIC policy / use case guidance and transmitting enrichment information (EI) from the non-real-time RIC to the near-real-time RIC.

[0004] Currently, the producers and consumers of intelligent functions in non-real-time RICs and near-real-time RICs in O-RAN are all internal logical network elements or applications, and their training services or training capabilities cannot be exposed to the outside world. In other words, there is currently no effective collaborative training method between RICs. Summary of the Invention

[0005] At least one embodiment of this application provides a model training method, apparatus, device, medium, and program product to address the problem of the lack of collaborative training methods among RICs in the prior art.

[0006] To solve the above-mentioned technical problems, this application is implemented as follows:

[0007] In a first aspect, embodiments of this application provide a model training method, including:

[0008] Obtain the first request sent by the first network element, the first request being used to request the model training service producer to establish a training task;

[0009] Based on the first request, a second request is sent to the second or third network element, the second request being used to query the target model training service producer;

[0010] Obtain the identifier of the target model training service producer sent by the second network element or the third network element;

[0011] A third request is sent to the target model training service producer, the third request being used to request the target model training service producer to establish a training task.

[0012] Optionally, the model training method further includes:

[0013] If a fourth request is received from the first network element, and there are multiple target model training service producers that have completed the training task, then the models of the multiple target model training service producers that have completed the training task are aggregated. The fourth request is used to request the target model training service producers to cancel the training task.

[0014] Optionally, the model training method, wherein model aggregation is performed on the models of multiple target model training service producers that have completed the training task, includes:

[0015] If the model requirement identifier carried in the fourth request is used to indicate that a model is needed, or if all the first notifications sent by the target model training service producers are currently obtained, then the models of the multiple target model training service producers that have completed the training task are aggregated, and the first notification is used to instruct the target model training service producers to complete the training task.

[0016] Optionally, the model training method further includes:

[0017] Obtain the first feedback sent by the target model training service producer, wherein the first feedback carries a training task identifier;

[0018] Based on the first feedback, a second feedback is sent to the first network element, the second feedback carrying the training task identifier corresponding to the target model training service producer;

[0019] When the target model training service producer completes the training task, the training status of the target model training service producer corresponding to the training task identifier is sent to the first network element.

[0020] Optionally, in the model training method, sending the training status of the target model training service producer corresponding to the training task identifier to the first network element includes:

[0021] Obtain the fifth request sent by the first network element, the fifth request carrying the training task identifier, the fifth request being used to request the training status of the target model training service producer corresponding to the training task identifier;

[0022] Send the training status of the target model training service producer corresponding to the training task identifier to the first network element.

[0023] Optionally, in the model training method, before obtaining the first request sent by the first network element, the method further includes:

[0024] Obtain the training capability registration information and / or training service registration information of the model training service producer. The training capability registration information carries the identifier and model training capability information of the model training service producer, and the training service registration information carries the identifier and service configuration file information of the model training service producer.

[0025] Register the training capability of the model training service producer based on the model training service producer's identifier and model training capability information, and / or register the training service of the model training service producer based on the model training service producer's identifier and service configuration file information.

[0026] Optionally, the model training method further includes:

[0027] Obtain the sixth request sent by the third network element, the sixth request being used to request subscription to training capabilities;

[0028] A third feedback is sent to the third network element, and a seventh request is sent to the second network element. The third feedback carries the identifier of the model training service producer and the model training capability information. The seventh request is used to request service registration, and the seventh request carries the identifier of the model training service producer and service configuration file information.

[0029] Upon receiving a second notification from the model training service producer, a third notification is sent to the third network element, and / or, upon receiving an eighth request from the model training service producer, a ninth request is sent to the third network element. The second notification and the third notification are used to indicate changes in the training capabilities of the model training service producer, and the eighth request and the ninth request are used to request changes to the training service.

[0030] Optionally, the model training method further includes:

[0031] Obtain the tenth request sent by the third network element, the tenth request being used to request cancellation of the training capability subscription;

[0032] A fourth feedback is sent to the third network element, indicating that the training capability subscription request has been cancelled, and an eleventh request is sent to the second network element, requesting the cancellation of some or all model training services of the target model training service producers.

[0033] Secondly, embodiments of this application provide a model training method, including:

[0034] Obtain the second request sent by the fourth network element, the second request being used to query the target model training service producer;

[0035] Based on the second request, a fourth request is sent to the third network element, the fourth request being used to query the target model training service producer;

[0036] Obtain the identifier of the candidate model training service producer sent by the third network element;

[0037] Based on the identifier of the candidate model training service producer, the identifier of the target model training service producer is sent to the fourth network element.

[0038] Thirdly, embodiments of this application provide a model training method, including:

[0039] Obtain the second request sent by the fourth network element, or obtain the fourth request sent by the second network element; both the second request and the fourth request are used to query the target model training service producer;

[0040] The target model training service producer is determined based on the second request or the fourth request;

[0041] Send the identifier of the target model training service producer to the fourth network element or the second network element.

[0042] Optionally, in the model training method, the fourth request carries the identifier of the model training service producer;

[0043] Based on the fourth request, the target model training service producer is determined, including:

[0044] The target model training service producer is determined based on the model training capability information corresponding to the identifier of the model training service producer.

[0045] Optionally, the model training method further includes:

[0046] Send a sixth request to the fourth network element, the sixth request being used to request subscription to training capabilities;

[0047] Obtain the third feedback sent by the fourth network element, wherein the third feedback carries the identifier of the model training service producer and model training capability information;

[0048] Obtain the third notification sent by the fourth network element, and / or obtain the ninth request sent by the fourth network element, wherein the third notification is used to indicate the change of training capability of the model training service producer, and the ninth request is used to request a change of training service.

[0049] Optionally, the model training method further includes:

[0050] Send a tenth request to the fourth network element, the tenth request being used to request cancellation of the training capability subscription;

[0051] Obtain the fourth feedback sent by the fourth network element, which indicates that the training capability subscription request has been cancelled.

[0052] Fourthly, embodiments of this application provide a model training apparatus, comprising:

[0053] The first acquisition module is used to acquire the first request sent by the first network element, the first request being used to request the model training service producer to establish a training task.

[0054] The first sending module is used to send a second request to a second network element or a third network element according to the first request, wherein the second request is used to query the target model training service producer;

[0055] The second acquisition module is used to acquire the identifier of the target model training service producer sent by the second network element or the third network element;

[0056] The second sending module is used to send a third request to the target model training service producer, the third request being used to request the target model training service producer to establish a training task.

[0057] Fifthly, embodiments of this application provide a model training apparatus, comprising:

[0058] The third acquisition module is used to acquire the second request sent by the fourth network element. The second request is used to query the target model training service producer.

[0059] The third sending module is used to send a fourth request to the third network element according to the second request, wherein the fourth request is used to query the target model training service producer;

[0060] The fourth acquisition module is used to acquire the identifier of the candidate model training service producer sent by the third network element;

[0061] The fourth sending module is used to send the identifier of the target model training service producer to the fourth network element according to the identifier of the candidate model training service producer.

[0062] Sixthly, embodiments of this application provide a model training apparatus, comprising:

[0063] The fifth acquisition module is used to acquire the second request sent by the fourth network element, or to acquire the fourth request sent by the second network element; both the second request and the fourth request are used to query the target model training service producer.

[0064] The determination module is used to determine the target model training service producer based on the second request or the fourth request;

[0065] The fifth sending module is used to send the identifier of the target model training service producer to the fourth network element or the second network element.

[0066] In a seventh aspect, embodiments of this application provide a model training device, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the model training method as described in the first aspect, or the model training method as described in the second aspect, or the model training method as described in the third aspect.

[0067] Eighthly, embodiments of this application provide a readable storage medium storing a program that, when executed by a processor, implements the model training method as described in the first aspect, or the model training method as described in the second aspect, or the model training method as described in the third aspect.

[0068] Ninthly, embodiments of this application provide a computer program product, including computer instructions, which, when executed by a processor, implement the model training method as described in the first aspect, or the model training method as described in the second aspect, or the model training method as described in the third aspect.

[0069] Compared with existing technologies, the model training method, apparatus, device, medium, and program product provided in this application embodiment, by acquiring a first request sent by a first network element, the first request being used to request a model training service producer to establish a training task; based on the first request, sending a second request to a second or third network element, the second request being used to query a target model training service producer; acquiring the identifier of the target model training service producer sent by the second or third network element; and sending a third request to the target model training service producer, the third request being used to request the target model training service producer to establish a training task. In this way, collaborative training between network elements can be achieved, thereby expanding the wireless resource management range of network elements and improving the overall performance and intelligence level of the network. Attached Figure Description

[0070] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0071] Figure 1 This is a schematic flowchart of the model training method according to one embodiment of this application;

[0072] Figure 2 This is a flowchart illustrating one implementation method of the model training method described in the application embodiment;

[0073] Figure 3 This is a flowchart illustrating a second implementation of the model training method described in the application embodiment;

[0074] Figure 4 This is a flowchart illustrating implementation method three of the model training method described in the application embodiment;

[0075] Figure 5 This is a flowchart illustrating implementation four of the model training method described in the application embodiment;

[0076] Figure 6 This is a flowchart illustrating implementation five of the model training method described in the application embodiment;

[0077] Figure 7 This is a schematic flowchart of a model training method according to another embodiment of this application;

[0078] Figure 8 This is a schematic flowchart of a model training method according to another embodiment of this application;

[0079] Figure 9 This is a schematic diagram of the structure of the model training device according to one embodiment of this application;

[0080] Figure 10 This is a schematic diagram of the structure of the model training device according to another embodiment of this application;

[0081] Figure 11 This is a schematic diagram of the structure of the model training device according to another embodiment of this application;

[0082] Figure 12 This is a hardware block diagram of the model training device described in the embodiments of this application. Detailed Implementation

[0083] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, without limiting the number of objects; for example, the first object can be one or more. Furthermore, "or" in this application indicates at least one of the connected objects. For example, "A or B" covers three scenarios: Scenario 1: including A but not B; Scenario 2: including B but not A; Scenario 3: including both A and B. The character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0084] Please refer to Figure 1 This application provides a model training method, which is applied to a fourth network element, including A1-ML (A1 ML Model Management Service), which can be set in a non-real-time RIC.

[0085] Furthermore, the method includes the following steps:

[0086] 101. Obtain the first request sent by the first network element. The first request is used to request the model training service producer to establish a training task.

[0087] It should be noted that the first network element, as a model training service consumer, includes at least one of the following:

[0088] Near real-time RIC; rApp in non-real-time RIC; AI / ML workflow function; third-party network equipment; third-party applications; third-party platforms.

[0089] Optionally, the first request includes at least one of the following:

[0090] The identifier of the first network element; training information; notification URI (Uniform Resource Identifier); model retrieval URI.

[0091] The training information includes at least one of the following:

[0092] Model training service identifier; Model training service registration identifier; Model identifier; Model information (including application scenario and data information).

[0093] The data information mentioned herein is different from the response to the input request; it is the training data information of an existing model, including at least one of the following:

[0094] Data collection information; data duration; data source; model accuracy; data scale.

[0095] Optionally, the model training service producer includes at least one of the following:

[0096] Near real-time RIC; rApp in near real-time RIC.

[0097] 102. Based on the first request, send a second request to the second network element or the third network element. The second request is used to query the target model training service producer.

[0098] It should be noted that the second network element includes: SME (Service Management and Exposure). The third network element includes: AI / ML workflow function.

[0099] It should also be noted that the first network element, the second network element, and the third network element may belong to the same non-real-time RIC.

[0100] Optionally, the second request includes at least one of the following:

[0101] Training information; notification URI; model retrieval URI.

[0102] In one implementation, the fourth network element sends a second request to the second network element based on the first request; the second network element sends a fourth request to the third network element based on the second request, the fourth request being used to query the target model training service producer; the third network element determines a candidate model training service producer based on the fourth request; the third network element sends the identifier of the candidate model training service producer to the second network element; the second network element determines the target model training service producer based on the candidate model training service producer; and the second network element sends the identifier of the target model training service producer to the fourth network element.

[0103] Optionally, the fourth request carries the identifier of the model training service producer; the third network element determines the target model training service producer based on the model training capability information and / or service configuration file information corresponding to the identifier of the model training service producer.

[0104] Optionally, the fourth request carries the selection criteria and the identifier of the model training service producer; the third network element determines the target model training service producer based on the selection criteria and the model training capability information and / or service configuration file information corresponding to the identifier of the model training service producer.

[0105] Optionally, the selection criteria include at least one of the following:

[0106] Consumer identifier; Model training service identifier; Resource requirements; Computing power requirements; Algorithm requirements; Input requirements; Output requirements; Training framework requirements; Software library requirements; Scenario requirements; Model identifier; Requirement priority; Expected number of feedback producers.

[0107] Specifically, the algorithm requirements are described in Table 1 below. In Table 1, M indicates necessary and O indicates optional.

[0108]

[0109] Table 1: Description of Algorithm Requirements

[0110] Specifically, the input requirements are described in Table 2 below. In Table 2, M indicates necessary and O indicates optional.

[0111]

[0112]

[0113] Table 2: Description of Input Requirements

[0114] Optionally, the model training capability information includes at least one of the following:

[0115] Resource information (computing power information); algorithm information; input information; output information; training framework information; software library information; model identifier; model information (including application scenarios and data information).

[0116] The algorithm information and input information are respectively responses to algorithm requirements and input requirements.

[0117] The data information is different from the response to the input request; it is the training data information of an existing model, including at least one of the following:

[0118] Data collection information; data duration; data source; model accuracy; data scale.

[0119] Figure 2 This is a flowchart illustrating an implementation of the model training method described in this application, where Near-RT RIC 1 represents the first network element; the target model training service producers include Near-RT RIC 2 and rApp; A1-ML represents the fourth network element; SME represents the second network element; and the AI / ML workflow function represents the third network element. (Refer to...) Figure 2 In this first embodiment, the method includes the following steps:

[0120] 201. Near-RT RIC 1 sends the first request (Request training service request) to A1-ML.

[0121] Query model training service producers:

[0122] 202, A1-ML sends a second request (Request training service request) to SME.

[0123] 203. SME sends a fourth request (Query training capability request) to the AI / ML workflow function.

[0124] 204. The AI / ML workflow function determines candidate model training service producers (Trainer selection) based on selection criteria and model training capability information of the model training service producers.

[0125] 205. The AI / ML workflow function sends a fifth feedback (Query training capability response) to the SME. This fifth feedback is used to instruct the candidate model training service producer and carries the identifier of the candidate model training service producer and model training capability information.

[0126] 206. SME determines the target model training service producer based on the candidate model training service producers and sends a sixth feedback (Request training service response) to A1-ML. This sixth feedback is used to instruct the target model training service producer and carries the identifier of the target model training service producer.

[0127] In another implementation, the fourth network element sends a second request to the third network element based on the first request; the third network element determines the target model training service producer based on the second request; and the third network element sends the identifier of the target model training service producer to the fourth network element.

[0128] 103. Obtain the identifier of the target model training service producer sent by the second network element or the third network element.

[0129] 104. Send a third request to the target model training service producer, the third request being used to request the target model training service producer to establish a training task.

[0130] Optionally, the third request shown includes at least one of the following:

[0131] The identifier of the first network element; training information; notification URI; model retrieval URI.

[0132] It should be noted that the fourth network element can obtain model information based on the model retrieval URI provided by the first network element, and store it in the Model repository function of the fourth network element or the non-real-time RIC, and provide the model retrieval URI as a new model retrieval URI to the target model training service producer.

[0133] In one embodiment, optionally, the above method further includes:

[0134] Obtain the first feedback sent by the target model training service producer, wherein the first feedback carries a training task identifier;

[0135] Based on the first feedback, a second feedback is sent to the first network element, the second feedback carrying the training task identifier corresponding to the target model training service producer;

[0136] When the target model training service producer completes the training task, the training status of the target model training service producer corresponding to the training task identifier is sent to the first network element.

[0137] Continue to refer to Figure 2 In this first embodiment, the method further includes the following steps:

[0138] Request training service:

[0139] 207. A1-ML sends a third request (Request training service request) to Near-RT RIC 2, which instructs Near-RT RIC 2 to set up and execute a training task.

[0140] 208. Near-RT RIC 2: Create training job.

[0141] 209. Near-RT RIC 2 sends the first feedback (Request training service response) to A1-ML. This first feedback indicates that the training task has been established and carries the training task identifier corresponding to Near-RT RIC 2.

[0142] 210. A1-ML sends a third request (Request training service request) to rApp, which instructs rApp to set up and execute the training task.

[0143] 211, rApp creates a training task.

[0144] 212. rApp sends a first feedback (Request training service response) to A1-ML. This first feedback indicates that the training task has been established and carries the training task identifier corresponding to rApp.

[0145] 213. A1-ML sends a second feedback (Request training service response) to Near-RT RIC 1. This second feedback is used to indicate that the training task has been established, and the second feedback carries a list of training task identifiers, which includes the training task identifiers corresponding to Near-RT RIC 2 and the training task identifiers corresponding to rApp.

[0146] In one implementation, optionally, sending the training status of the target model training service producer corresponding to the training task identifier to the first network element includes:

[0147] Obtain the fifth request sent by the first network element, the fifth request carrying the training task identifier, the fifth request being used to request the training status of the target model training service producer corresponding to the training task identifier;

[0148] Send the training status of the target model training service producer corresponding to the training task identifier to the first network element.

[0149] The fifth request includes at least one of the following:

[0150] The identifier of the first network element; at least one training task identifier.

[0151] It should be noted that the training task identifier can be the training task identifier corresponding to all target model training service producers, or it can be the training task identifier corresponding to some target model training service producers.

[0152] In this embodiment, after receiving the fifth request sent by the first network element, the fourth network element sends a query request to the target model training service producer that needs to query the training status. The query request includes at least one of the following: the identifier of the first network element; and the training task identifier. Then, the target model training service producer that needs to query the training status sends the current training status to the first network element; the fourth network element sends feedback on the fifth request to the first network element, including the training task identifier and the training status.

[0153] Optionally, the training state of the target model training service producer includes at least one of the following:

[0154] Waiting for training; Training in progress; Updating; Training interrupted; Training resumed; Training stopped; System crashed; Training completed.

[0155] It should be noted that when the target model training service producer completes the training task or the training task status changes, the target model training service producer sends a training task status change notification to the fourth network element. The training task status change notification includes at least one of the following: training task identifier; training status; model retrieval URI. Then, the fourth network element forwards the training status change notification to the first network element.

[0156] In one embodiment, optionally, the above method further includes:

[0157] If a fourth request is received from the first network element, and there are multiple target model training service producers that have completed the training task, then the models of the multiple target model training service producers that have completed the training task are aggregated. The fourth request is used to request the target model training service producers to cancel the training task.

[0158] In this embodiment of the application, when the first network element decides to cancel the training task, a fourth request is sent to the fourth network element. The fourth request includes at least one of the following: training task identifier; model requirement identifier (used to indicate whether a model is needed); and model aggregation algorithm.

[0159] Optionally, the first network element can determine whether to cancel the training task based on the training status of the target model training service producer.

[0160] Optionally, the model aggregation algorithm includes, but is not limited to:

[0161] FedAvg; FedProx; FedNova; FedCos; FedALA; SCAFFOLD.

[0162] In one implementation, optionally, model aggregation is performed on the models of multiple target model training service producers that have completed the training task, including:

[0163] If the model requirement identifier carried in the fourth request is used to indicate that a model is needed, or if all the first notifications sent by the target model training service producers are currently obtained, then the models of the multiple target model training service producers that have completed the training task are aggregated, and the first notification is used to instruct the target model training service producers to complete the training task.

[0164] In this embodiment, if the model requirement identifier carried in the fourth request indicates a model is needed, or if all target model training service producers have completed their training tasks, then the fourth network element performs model aggregation on the models of the target model training service producers that have completed their training tasks. Simultaneously, the fourth network element sends a training task cancellation notification to the target model training service producers that have not yet completed their training tasks. Next, the fourth network element sends a training task status change notification to the first network element, the training task status change notification including at least one of the following: training task identifier; training status; model retrieval URI.

[0165] In one embodiment, optionally, the above method further includes:

[0166] If a fourth request is received from the first network element, and the model requirement identifier carried in the fourth request indicates that a model is not needed, then a training task status change notification is sent directly to the first network element.

[0167] Figure 3 This is a flowchart illustrating a second implementation of the model training method described in this application, where Near-RT RIC 1 represents the first network element; the target model training service producers include Near-RT RIC 2 and rApp; and A1-ML represents the fourth network element. (Refer to...) Figure 3 In this second embodiment, the method includes the following steps:

[0168] Query training status:

[0169] 301. Near-RT RIC 1 sends the twelfth request (Query training status request) to A1-ML. This twelfth request is used to query the training status of the target model training service producer.

[0170] 302, A1-ML sends a thirteenth request (Query training status request) to Near-RT RIC 2, which is used to query the training status of the target model training service producer.

[0171] 303. Near-RT RIC 2 sends the ninth feedback (Query training status response) to A1-ML, which carries the training status information of the target model training service producer.

[0172] 304. A1-ML sends a 13th request (Query training status request) to rApp, which is used to query the training status of the target model training service producer.

[0173] 305, rApp sends the ninth feedback to A1-ML (Query training status response), which carries the training status information of the target model training service producer.

[0174] 306. A1-ML sends the tenth feedback (Query training status response) to Near-RT RIC 1, which carries the training status information of the target model training service producer.

[0175] Training status change:

[0176] 307. rApp sends a first notification (Notify training job status change) to A1-ML, which indicates a change in the training job status.

[0177] 308. Near-RT RIC 2 sends the first notification (Notify training job status change) to A1-ML, which indicates a change in the training job status.

[0178] 309. A1-ML sends a fourth notification (Notify training job status change) to Near-RT RIC 1. This fourth notification indicates a change in the training job status and carries a list of training job identifiers and the training status.

[0179] Training ended or was cancelled:

[0180] 310. Near-RT RIC 1 sends a fourth request (Cancel training job request) to A1-ML. This fourth request is used to request the target model training service producer to cancel the training task.

[0181] 311, A1-ML performs model aggregation.

[0182] 312, A1-ML sends a fifth notification (Notify training job status change) to Near-RT RIC 1, which is used to indicate the change in the training job status of the target model training service producer.

[0183] In one embodiment, optionally, before obtaining the first request sent by the first network element, the above method further includes:

[0184] Obtain the training capability registration information and / or training service registration information of the model training service producer. The training capability registration information carries the identifier and model training capability information of the model training service producer, and the training service registration information carries the identifier and service configuration file information of the model training service producer.

[0185] Register the training capability of the model training service producer based on the model training service producer's identifier and model training capability information, and / or register the training service of the model training service producer based on the model training service producer's identifier and service configuration file information.

[0186] In this embodiment, the model training service producer can request the fourth network element to register model training capabilities and model training services. Thus, after receiving the first request, the fourth network element can determine the target model training service producer based on the model training capability information and / or service configuration file information of the model training service producer.

[0187] Optionally, the service configuration file information includes at least one of the following:

[0188] Service name; service type; communication mechanism; interface endpoint details; version number; data format.

[0189] The interface endpoint details include at least one of the following:

[0190] IP address; port number; URI; protocol.

[0191] Figure 4This is a flowchart illustrating a second implementation of the model training method described in this application, where Near-RT RIC represents the model training service producer; A1-ML represents the fourth network element. (Refer to...) Figure 4 In this third embodiment, the method includes the following steps:

[0192] Register AI / ML model training capabilities:

[0193] 401. Near-RT RIC sends training capability registration information to A1-ML (Register ML model training capability), which includes at least one of the following: the identifier of the model training service producer; and model training capability information.

[0194] 402, A1-ML registers the AI / ML model training capability for Near-RT RIC.

[0195] 403. A1-ML sends the tenth feedback (Register ML model trainingcapability response) to Near-RT RIC. This tenth feedback indicates that the training capability has been registered and carries a registration identifier.

[0196] Subscribe to changes in AI / ML model training capabilities:

[0197] 404. A1-ML sends a fourteenth request (Subscribe training capability request) to Near-RT RIC. This fourteenth request is used to subscribe to changes in training capabilities, and the fourteenth request includes at least one of the following: model training capability type; model identifier; and the purpose of the model training capability.

[0198] 405, Near-RT RIC Authorization: Create subscription.

[0199] 406. Near-RT RIC sends an eleventh feedback (Subscribe training capability request response) to A1-ML. This eleventh feedback indicates that the training capability change has been subscribed to, and it carries a subscription identifier.

[0200] 407. When the model training capability changes, Near-RT RIC sends a sixth notification (Notify training capability changes) to A1-ML. This sixth notification indicates that the training capability has changed and includes at least one of the following: subscription identifier; new model training capability information; and information on changes to the old capability.

[0201] Cancel AI / ML model training capabilities:

[0202] 408. Near-RT RIC sends a message to A1-ML deregistering the ML model training capability, carrying the model producer's identifier and / or registration identifier.

[0203] 409, A1-ML deregisters Near-RT RIC model training capability.

[0204] 410. A1-ML sends the twelfth feedback (Deregister ML model training capability response) to Near-RT RIC, which indicates that the model training capability information has been deregistered.

[0205] Figure 5 This is a flowchart illustrating a second implementation of the model training method described in this application, where Near-RT RIC represents the model training service producer; A1-ML represents the fourth network element. (Refer to...) Figure 5 In this fourth embodiment, the method includes the following steps:

[0206] Register for AI / ML model training services:

[0207] 501. Near-RT RIC sends training service registration information (Register ML model trainingservice) to A1-ML. The training service registration information includes at least one of the following: the identifier of the model training service producer; and service configuration file information.

[0208] 502, A1-ML registration for Near-RT RIC model training service.

[0209] 503, A1-ML sends a thirteenth response (Register ML model trainingservice response) to Near-RT RIC. This thirteenth response indicates that the training service has been registered and carries a service identifier.

[0210] Changes to AI / ML model training services:

[0211] 504, Near-RT RIC sends a fifteenth request (Update training service request) to A1-ML. This fifteenth request is used to request changes to the training service, and the fifteenth request includes at least one of the following: the identifier of the model training service producer; the service identifier; the new service configuration file information; and the change information of the old service.

[0212] 505, A1-ML updates Near-RT RIC service configuration file information (Update ML model trainingservice).

[0213] 506, A1-ML sends the fourteenth feedback (Update training service response) to Near-RT RIC, which indicates that the model training service has changed.

[0214] Cancel AI / ML model training service:

[0215] 507. Near-RT RIC sends a message to A1-ML to deregister the ML modeltraining service, carrying the identifier of the model training service producer and / or the service identifier.

[0216] 508, A1-ML Remove ML model training service from Near-RT RIC.

[0217] 509, A1-ML sends its fifteenth response (Deregister ML model training service response) to Near-RT RIC, which indicates that the model training service information has been deregistered.

[0218] In one embodiment, optionally, the above method further includes:

[0219] Obtain the sixth request sent by the third network element, the sixth request being used to request subscription to training capabilities;

[0220] A third feedback is sent to the third network element, and a seventh request is sent to the second network element. The third feedback carries the identifier of the model training service producer and the model training capability information. The seventh request is used to request service registration, and the seventh request carries the identifier of the model training service producer and service configuration file information.

[0221] Upon receiving a second notification from the model training service producer, a third notification is sent to the third network element, and / or, upon receiving an eighth request from the model training service producer, a ninth request is sent to the third network element. The second notification and the third notification are used to indicate changes in the training capabilities of the model training service producer, and the eighth request and the ninth request are used to request changes to the training service.

[0222] It should be noted that by obtaining the sixth request sent by the third network element through the fourth network element, the fourth network element can enable the fourth network element to control the training resources of the target model training service producer. Of course, it can also enable the non-real-time RIC to control the training resources of the near real-time RIC.

[0223] After the fourth network element enables control over the training resources of the target model training service producer, the fourth network element can send the new model training capability information and / or the change information of the old capability in the training capability change notification to the third network element, and send the new service configuration file information and / or the change information of the old service in the training service change request to the second network element.

[0224] Figure 6 This is a flowchart illustrating one implementation of the model training method described in this application, where Near-RT RIC represents the model training service producer; A1-ML represents the fourth network element; SME represents the second network element; and AI / MLworkflow function represents the third network element. (Refer to...) Figure 6 In this fifth embodiment, the method includes the following steps:

[0225] Non-RT RIC enables control over training resources for Near-RT RIC:

[0226] 601, the AI / ML workflow function sends a sixth request (Subscribe trainingcapability request) to A1-ML.

[0227] 602, A1-ML Create subscription.

[0228] 603. A1-ML sends a third feedback (Subscribe trainingcapability request response) to the AI / ML workflow function, wherein the third feedback includes at least one of the following: subscription identifier; identifier of the model training service producer; and model training capability information.

[0229] 604, A1-ML sends a seventh request (Register A1-ML service request) to SME.

[0230] 605, SME Registration Service.

[0231] 606. SME sends a seventh response (Register A1-ML service response) to A1-ML, which carries a service identifier.

[0232] Changes to model training capabilities or services of Near-RT RIC:

[0233] 607, Near-RT RIC training capability change (AI / ML model training capability change).

[0234] 608. Near-RT RIC sends a second notification to A1-ML (Notify training capability changes).

[0235] 609. A1-ML sends a third notification (notify training capability changes) to the AI / ML workflow function.

[0236] 610, Near-RT RIC sends the eighth request (Update training service request) to A1-ML.

[0237] 611, A1-ML sends the ninth request (Update A1-ML service request) to SME.

[0238] 612, SME Update A1-ML service.

[0239] 613, SME sends the eighth feedback (Update A1-ML service response) to A1-ML.

[0240] In one embodiment, optionally, the above method further includes:

[0241] Obtain the tenth request sent by the third network element, the tenth request being used to request cancellation of the training capability subscription;

[0242] A fourth feedback is sent to the third network element, indicating that the training capability subscription request has been cancelled, and an eleventh request is sent to the second network element, requesting the cancellation of some or all model training services of the target model training service producers.

[0243] It should be noted that by obtaining the tenth request sent by the third network element through the fourth network element, the fourth network element can close the control of training resources of the target model training service producer. Of course, it can also close the control of training resources of the near real-time RIC by the non-real-time RIC, thereby canceling the model training services of some or all target model training service producers, that is, removing the A1-ML service.

[0244] In this embodiment, after receiving the tenth request from the third network element, the fourth network element deletes its subscription to the model training capabilities of the model training service producer and sends a fourth feedback to the third network element. Simultaneously, the fourth network element may send an eleventh request to the second network element, or the third network element may also send an eleventh request to the second network element, the eleventh request carrying a service identifier; the second network element cancels some or all of the target model training service producer's model training services, and the second network element sends a ninth feedback to the fourth network element and / or the third network element.

[0245] Continue to refer to Figure 6 In this fifth embodiment, the method further includes the following steps:

[0246] Non-RT RIC disables training resource control for Near-RT RIC:

[0247] 614. The AI / ML workflow function sends a tenth request (Unsubscribe trainingcapability request) to A1-ML, which carries the subscription identifier.

[0248] 615, A1-ML Cancel subscription.

[0249] 616, A1-ML sends a fourth feedback (Unsubscribe trainingcapability request response) to the AI / ML workflow function.

[0250] 617. A1-ML sends the eleventh request (Deregister A1-ML service request) to SME.

[0251] 618, SME Remove A1-ML service.

[0252] 619. SME sends the ninth feedback to A1-ML (Deregister A1-ML service response).

[0253] Therefore, it is understood that the fourth network element (including A1-ML) in the embodiments of this application can support cross-RIC collaborative training, provide A1-ML services, and also serve as a near real-time model training management function, providing A1-ML services as a near real-time RIC training service set internally or externally.

[0254] The A1-ML service supports near real-time RICs to register their AI / ML model training services and / or AI / ML model training capabilities on the non-real-time RIC side; the A1-ML service enables non-real-time RICs to register or discover near real-time RICs' model training capabilities through their subscriptions or requests for model training services; the A1-ML service also provides model aggregation services.

[0255] Please refer to Figure 7 This application provides a model training method, which is applied to a second network element.

[0256] Furthermore, the method includes the following steps:

[0257] 701, Obtain the second request sent by the fourth network element. The second request is used to query the target model training service producer.

[0258] 702. Based on the second request, a fourth request is sent to the third network element, the fourth request being used to query the target model training service producer.

[0259] 703, Obtain the identifier of the candidate model training service producer sent by the third network element.

[0260] 704. Based on the identifier of the candidate model training service producer, send the identifier of the target model training service producer to the fourth network element.

[0261] It should be noted that the second network element can select the target model training service producer from the candidate model training service producers.

[0262] In this embodiment, the second network element can also perform the steps performed by the second network element in the model training method applied to the fourth network element. To avoid repetition, these steps will not be repeated here.

[0263] Please refer to Figure 8 This application provides a model training method, which is applied to a third network element.

[0264] Furthermore, the method includes the following steps:

[0265] 801, obtain the second request sent by the fourth network element, or obtain the fourth request sent by the second network element; both the second request and the fourth request are used to query the target model training service producer.

[0266] 802, Based on the second request or the fourth request, determine the target model training service producer.

[0267] 803, send the identifier of the target model training service producer to the fourth network element or the second network element.

[0268] In this embodiment, the third network element can also perform the steps performed by the third network element in the model training method applied to the fourth network element. To avoid repetition, these steps will not be repeated here.

[0269] In one implementation, optionally, the fourth request carries the identifier of the model training service producer;

[0270] Based on the fourth request, the target model training service producer is determined, including:

[0271] The target model training service producer is determined based on the model training capability information corresponding to the identifier of the model training service producer.

[0272] In one embodiment, optionally, the above method further includes:

[0273] Send a sixth request to the fourth network element, the sixth request being used to request subscription to training capabilities;

[0274] Obtain the third feedback sent by the fourth network element, wherein the third feedback carries the identifier of the model training service producer and model training capability information;

[0275] Obtain the third notification sent by the fourth network element, and / or obtain the ninth request sent by the fourth network element, wherein the third notification is used to indicate the change of training capability of the model training service producer, and the ninth request is used to request a change of training service.

[0276] In one embodiment, optionally, the above method further includes:

[0277] Send a tenth request to the fourth network element, the tenth request being used to request cancellation of the training capability subscription;

[0278] Obtain the fourth feedback sent by the fourth network element, which indicates that the training capability subscription request has been cancelled.

[0279] In summary, the model training method described in this application addresses the current lack of a specific solution for cross-RIC collaborative training in O-RAN systems. Simultaneously, it expands the scope of training resource management for non-real-time RICs; otherwise, near-real-time RICs or other potential model training service consumers can only train themselves, unable to receive assistance from other near-real-time RICs, let alone non-real-time RICs or other applications. By using the first function (i.e., A1-ML function) in the non-real-time RIC as a training service set, training services are provided to both inside and outside the non-real-time RIC. Furthermore, the non-real-time RIC can enable or disable its management of training resources for the near-real-time RIC.

[0280] The various methods of the embodiments of this application have been described above. Apparatus for implementing the above methods will now be provided.

[0281] Please refer to Figure 9 This application also provides a model training apparatus, comprising:

[0282] The first acquisition module 901 is used to acquire the first request sent by the first network element, the first request being used to request the model training service producer to establish a training task.

[0283] The first sending module 902 is used to send a second request to a second network element or a third network element according to the first request, wherein the second request is used to query the target model training service producer;

[0284] The second acquisition module 903 is used to acquire the identifier of the target model training service producer sent by the second network element or the third network element;

[0285] The second sending module 904 is used to send a third request to the target model training service producer, the third request being used to request the target model training service producer to establish a training task.

[0286] Optionally, the model training apparatus further includes:

[0287] The aggregation module is used to aggregate the models of multiple target model training service producers that have completed the training task if the number of target model training service producers that have completed the training task is multiple, upon receiving a fourth request sent by the first network element. The fourth request is used to request the target model training service producers to cancel the training task.

[0288] Optionally, in the aforementioned model training apparatus, the aggregation module is specifically used for:

[0289] If the model requirement identifier carried in the fourth request is used to indicate that a model is needed, or if all the first notifications sent by the target model training service producers are currently obtained, then the models of the multiple target model training service producers that have completed the training task are aggregated, and the first notification is used to instruct the target model training service producers to complete the training task.

[0290] Optionally, the model training apparatus further includes:

[0291] The sixth acquisition module is used to acquire the first feedback sent by the target model training service producer, wherein the first feedback carries a training task identifier;

[0292] The sixth sending module is used to send a second feedback to the first network element based on the first feedback, wherein the second feedback carries the training task identifier corresponding to the target model training service producer;

[0293] The seventh sending module is used to send the training status of the target model training service producer corresponding to the training task identifier to the first network element when the target model training service producer completes the training task.

[0294] Optionally, in the aforementioned model training apparatus, the seventh sending module is specifically used for:

[0295] Obtain the fifth request sent by the first network element, the fifth request carrying the training task identifier, the fifth request being used to request the training status of the target model training service producer corresponding to the training task identifier;

[0296] Send the training status of the target model training service producer corresponding to the training task identifier to the first network element.

[0297] Optionally, the model training apparatus further includes:

[0298] The seventh acquisition module is used to acquire the training capability registration information and / or training service registration information of the model training service producer. The training capability registration information carries the identifier and model training capability information of the model training service producer, and the training service registration information carries the identifier and service configuration file information of the model training service producer.

[0299] The registration module is used to register the training capabilities of the model training service producer based on the producer's identifier and model training capability information, and / or to register the producer's training services based on the producer's identifier and service configuration file information.

[0300] Optionally, the model training apparatus further includes:

[0301] The eighth acquisition module is used to acquire the sixth request sent by the third network element, the sixth request being used to request subscription to training capabilities;

[0302] The eighth sending module is used to send a third feedback to the third network element and a seventh request to the second network element. The third feedback carries the identifier of the model training service producer and model training capability information. The seventh request is used to request service registration and carries the identifier and service configuration file information of the model training service producer.

[0303] The ninth sending module is used to send a third notification to the third network element when it receives a second notification sent by the model training service producer, and / or to send a ninth request to the third network element when it receives an eighth request sent by the model training service producer. The second notification and the third notification are used to indicate changes in the training capabilities of the model training service producer, and the eighth request and the ninth request are used to request changes in the training service.

[0304] Optionally, the model training apparatus further includes:

[0305] The ninth acquisition module is used to acquire the tenth request sent by the third network element, the tenth request being used to request cancellation of training capability subscription;

[0306] The tenth sending module is used to send a fourth feedback to the third network element, the fourth feedback being used to indicate that the training capability subscription request has been cancelled, and to send an eleventh request to the second network element, the eleventh request being used to request the cancellation of some or all model training service producers' model training services.

[0307] It should be noted that the device in this embodiment is the same as the model training method described above. The implementation methods in the above embodiments are also applicable to the embodiments of this device and can achieve the same technical effect.

[0308] Please refer to Figure 10 This application also provides a model training apparatus, comprising:

[0309] The third acquisition module 1001 is used to acquire the second request sent by the fourth network element, the second request being used to query the target model training service producer;

[0310] The third sending module 1002 is used to send a fourth request to the third network element according to the second request, wherein the fourth request is used to query the target model training service producer;

[0311] The fourth acquisition module 1003 is used to acquire the identifier of the candidate model training service producer sent by the third network element;

[0312] The fourth sending module 1004 is used to send the identifier of the target model training service producer to the fourth network element according to the identifier of the candidate model training service producer.

[0313] It should be noted that the device in this embodiment is the same as the model training method described above. The implementation methods in the above embodiments are also applicable to the embodiments of this device and can achieve the same technical effect.

[0314] Please refer to Figure 11 This application also provides a model training apparatus, comprising:

[0315] The fifth acquisition module 1101 is used to acquire the second request sent by the fourth network element, or to acquire the fourth request sent by the second network element; both the second request and the fourth request are used to query the target model training service producer.

[0316] The determining module 1102 is used to determine the target model training service producer based on the second request or the fourth request;

[0317] The fifth sending module 1103 is used to send the identifier of the target model training service producer to the fourth network element or the second network element.

[0318] Optionally, in the model training apparatus, the fourth request carries the identifier of the model training service producer;

[0319] The determining module 1102 is specifically used for:

[0320] The target model training service producer is determined based on the model training capability information corresponding to the identifier of the model training service producer.

[0321] Optionally, the model training apparatus further includes:

[0322] The eleventh sending module is used to send a sixth request to the fourth network element, the sixth request being used to request subscription to training capabilities;

[0323] The tenth acquisition module is used to acquire the third feedback sent by the fourth network element, wherein the third feedback carries the identifier of the model training service producer and model training capability information;

[0324] The eleventh acquisition module is used to acquire the third notification sent by the fourth network element, and / or acquire the ninth request sent by the fourth network element. The third notification is used to indicate the change of training capability of the model training service producer, and the ninth request is used to request a change of training service.

[0325] Optionally, the model training apparatus further includes:

[0326] The twelfth sending module is used to send a tenth request to the fourth network element, the tenth request being used to request cancellation of the training capability subscription;

[0327] The twelfth acquisition module is used to acquire the fourth feedback sent by the fourth network element, which indicates that the training capability subscription request has been cancelled.

[0328] It should be noted that the device in this embodiment is the same as the model training method described above. The implementation methods in the above embodiments are also applicable to the embodiments of this device and can achieve the same technical effect.

[0329] This application also provides a model training device, such as... Figure 12 As shown, it includes: a processor 1201, a memory 1202, a transceiver 1203, and a program or instructions stored in the memory 1202 and executable on the processor 1201; when the processor 1201 executes the program or instructions, it implements the various processes of the above-described model training method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0330] The transceiver 1203 is used to receive and send data under the control of the processor 1201.

[0331] Among them, Figure 12In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 1201 and memory represented by memory 1202 together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 1203 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. For different user equipment, the user interface 1204 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0332] The processor 1201 is responsible for managing the bus architecture and general processing, while the memory 1202 can store the data used by the processor 1201 when performing operations.

[0333] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described model training method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0334] This application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various processes of the above-described model training method embodiments and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0335] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0336] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0337] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A model training method, characterized in that, include: Obtain the first request sent by the first network element, the first request being used to request the model training service producer to establish a training task; Based on the first request, a second request is sent to the second or third network element, the second request being used to query the target model training service producer; Obtain the identifier of the target model training service producer sent by the second network element or the third network element; A third request is sent to the target model training service producer, the third request being used to request the target model training service producer to establish a training task.

2. The model training method according to claim 1, characterized in that, The method further includes: If a fourth request is received from the first network element, and if there are multiple target model training service producers that have completed the training task, then the models of the multiple target model training service producers that have completed the training task are aggregated. The fourth request is used to request the target model training service producers to cancel the training task.

3. The model training method according to claim 2, characterized in that, Model aggregation is performed on the models of multiple target model training service producers that have completed the training task, including: If the model requirement identifier carried in the fourth request is used to indicate that a model is needed, or if all the first notifications sent by the target model training service producers are currently obtained, then the models of the multiple target model training service producers that have completed the training task are aggregated, and the first notification is used to instruct the target model training service producers to complete the training task.

4. The model training method according to claim 1, characterized in that, The method further includes: Obtain the first feedback sent by the target model training service producer, wherein the first feedback carries a training task identifier; Based on the first feedback, a second feedback is sent to the first network element, the second feedback carrying the training task identifier corresponding to the target model training service producer; When the target model training service producer completes the training task, the training status of the target model training service producer corresponding to the training task identifier is sent to the first network element.

5. The model training method according to claim 4, characterized in that, Sending the training status of the target model training service producer corresponding to the training task identifier to the first network element includes: Obtain the fifth request sent by the first network element, the fifth request carrying the training task identifier, the fifth request being used to request the training status of the target model training service producer corresponding to the training task identifier; Send the training status of the target model training service producer corresponding to the training task identifier to the first network element.

6. The model training method according to claim 1, characterized in that, Before obtaining the first request sent by the first network element, the method further includes: Obtain the training capability registration information and / or training service registration information of the model training service producer. The training capability registration information carries the identifier and model training capability information of the model training service producer, and the training service registration information carries the identifier and service configuration file information of the model training service producer. Register the training capability of the model training service producer based on the model training service producer's identifier and model training capability information, and / or register the training service of the model training service producer based on the model training service producer's identifier and service configuration file information.

7. The model training method according to claim 6, characterized in that, The method further includes: Obtain the sixth request sent by the third network element, the sixth request being used to request subscription to training capabilities; A third feedback is sent to the third network element, and a seventh request is sent to the second network element. The third feedback carries the identifier of the model training service producer and the model training capability information. The seventh request is used to request service registration, and the seventh request carries the identifier of the model training service producer and service configuration file information. Upon receiving a second notification from the model training service producer, a third notification is sent to the third network element, and / or, upon receiving an eighth request from the model training service producer, a ninth request is sent to the third network element. The second notification and the third notification are used to indicate changes in the training capabilities of the model training service producer, and the eighth request and the ninth request are used to request changes to the training service.

8. The model training method according to claim 7, characterized in that, The method further includes: Obtain the tenth request sent by the third network element, the tenth request being used to request cancellation of the training capability subscription; A fourth feedback is sent to the third network element, indicating that the training capability subscription request has been cancelled, and an eleventh request is sent to the second network element, requesting the cancellation of some or all model training services of the target model training service producers.

9. A model training method, characterized in that, include: Obtain the second request sent by the fourth network element, the second request being used to query the target model training service producer; Based on the second request, a fourth request is sent to the third network element, the fourth request being used to query the target model training service producer; Obtain the identifier of the candidate model training service producer sent by the third network element; Based on the identifier of the candidate model training service producer, the identifier of the target model training service producer is sent to the fourth network element.

10. A model training method, characterized in that, include: Get the second request sent by the fourth network element, or get the fourth request sent by the second network element; Both the second request and the fourth request are used to query the target model training service producer; The target model training service producer is determined based on the second request or the fourth request; Send the identifier of the target model training service producer to the fourth network element or the second network element.

11. The model training method according to claim 10, characterized in that, The fourth request carries the identifier of the model training service producer; Based on the fourth request, the target model training service producer is determined, including: The target model training service producer is determined based on the model training capability information corresponding to the identifier of the model training service producer.

12. The model training method according to claim 10, characterized in that, The method further includes: Send a sixth request to the fourth network element, the sixth request being used to request subscription to training capabilities; Obtain the third feedback sent by the fourth network element, wherein the third feedback carries the identifier of the model training service producer and model training capability information; Obtain the third notification sent by the fourth network element, and / or obtain the ninth request sent by the fourth network element, wherein the third notification is used to indicate the change of training capability of the model training service producer, and the ninth request is used to request a change of training service.

13. The model training method according to claim 12, characterized in that, The method further includes: Send a tenth request to the fourth network element, the tenth request being used to request cancellation of the training capability subscription; Obtain the fourth feedback sent by the fourth network element, which indicates that the training capability subscription request has been cancelled.

14. A model training device, characterized in that, include: The first acquisition module is used to acquire the first request sent by the first network element, the first request being used to request the model training service producer to establish a training task. The first sending module is used to send a second request to a second network element or a third network element according to the first request, wherein the second request is used to query the target model training service producer; The second acquisition module is used to acquire the identifier of the target model training service producer sent by the second network element or the third network element; The second sending module is used to send a third request to the target model training service producer, the third request being used to request the target model training service producer to establish a training task.

15. A model training device, characterized in that, include: The third acquisition module is used to acquire the second request sent by the fourth network element. The second request is used to query the target model training service producer. The third sending module is used to send a fourth request to the third network element according to the second request, the fourth request being used to query the target model training service producer; The fourth acquisition module is used to acquire the identifier of the candidate model training service producer sent by the third network element; The fourth sending module is used to send the identifier of the target model training service producer to the fourth network element according to the identifier of the candidate model training service producer.

16. A model training device, characterized in that, include: The fifth acquisition module is used to acquire the second request sent by the fourth network element, or to acquire the fourth request sent by the second network element; Both the second request and the fourth request are used to query the target model training service producer; The determination module is used to determine the target model training service producer based on the second request or the fourth request; The fifth sending module is used to send the identifier of the target model training service producer to the fourth network element or the second network element.

17. A model training device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the model training method as described in any one of claims 1 to 8, or the model training method as described in claim 9, or the model training method as described in claims 10 to 13.

18. A readable storage medium, characterized in that, The readable storage medium stores a program that, when executed by a processor, implements the model training method as described in any one of claims 1 to 8, or the model training method as described in claim 9, or the model training method as described in claims 10 to 13.

19. A computer program product, characterized in that, It includes computer instructions, which, when executed by a processor, implement the model training method as described in any one of claims 1 to 8, or the model training method as described in claim 9, or the model training method as described in claims 10 to 13.

Images