Communication method and apparatus, related device, and storage medium

Abstract

The application discloses a communication method, device, first model training logical function (MTLF), network repository function (NRF), analysis logical function (AnLF) and storage medium. The method comprises the following steps: a first MTLF obtains a first model from a first function body, the first model comprising at least one model pre-trained by a second MTLF; and the first model is fine-tuned to obtain a second model, the second model being used for inference analysis performed by an AnLF.

Description

Technical Field

[0001] This application relates to the field of core networks, and more particularly to a communication method, apparatus, related equipment, and storage medium. Background Technology

[0002] In the network intelligence architecture defined by relevant technologies, the Network Data Analytics Function (NWDAF) serves as the big data analysis engine within the network. It is responsible for centrally collecting data from various layers, including network elements, network management systems, and applications; training data models to obtain relevant data models; and performing inference processes based on real-time class data. The NWDAF's functionality is provided by splitting it into a Model Training Logical Function (MTLF) and an Analytics Logical Function (AnLF). MTLF and AnLF can share models; that is, MTLF can perform model training and send the trained model to AnLF for inference and analysis.

[0003] However, in related technologies, model sharing between MTLF and AnLF poses a risk of privacy data leakage. Summary of the Invention

[0004] To address the related technical problems, embodiments of this application provide a communication method, apparatus, related devices, and storage medium.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] This application provides a communication method applied to a first MTLF, including:

[0007] A first model is obtained from a first functional body, wherein the first model contains at least one model pre-trained by a second MTLF;

[0008] The first model is fine-tuned to obtain the second model, which is used by AnLF to perform inference analysis.

[0009] In the above scheme, the first functional unit includes the second MTLF or Analytics Data Repository Function (ADRF).

[0010] In the above scheme, obtaining the first model from the first functional body includes:

[0011] Send a first message to the Network Repository Function (NRF), the first message being used to request the discovery of a first model;

[0012] Receive second information sent by the NRF, the second information indicating that the first model is obtained from the first functional body;

[0013] The first model is obtained from the first functional body.

[0014] In the above scheme, the first information includes at least the identifier of the first model.

[0015] In the above scheme, the first information further includes at least one of the following:

[0016] The type of the first model;

[0017] The first model supports an artificial intelligence (AI) platform;

[0018] The AI ​​framework supported by the first model;

[0019] The first model supports the following device manufacturers;

[0020] Does the first model support model sharing?

[0021] Does the first model support cross-platform model format conversion?

[0022] Does the first model support cross-language model format conversion?

[0023] The conversion formats supported by the first model.

[0024] In the above scheme, the second information includes at least one of the following:

[0025] The address of the first functional unit;

[0026] The identifier of the first functional unit.

[0027] The method in the above scheme further includes:

[0028] Send the second model to the AnLF;

[0029] or,

[0030] Send the second model to the ADRF so that the AnLF can obtain the second model from the ADRF;

[0031] or,

[0032] A third message is sent to the NRF so that when the NRF receives the fourth message sent by the AnLF, it can send a fifth message to the AnLF; the third message is used to register the second model for service; the fourth message is used to request the discovery of the second model; and the fifth message indicates that the second model is obtained from the second functional body.

[0033] In the above scheme, the second functional body includes the first MTLF or ADRF.

[0034] In the above scheme, the third information includes at least the identifier of the second model.

[0035] In the above scheme, the third information also includes at least one of the following:

[0036] The address of the second functional unit;

[0037] The identifier of the second functional entity;

[0038] The type of the second model;

[0039] The second model supports AI platforms;

[0040] The second model supports the following AI frameworks;

[0041] The second model supports the following equipment manufacturers;

[0042] Does the second model support model sharing?

[0043] Does the second model support cross-platform model format conversion?

[0044] Does the second model support cross-language model format conversion?

[0045] The conversion formats supported by the second model.

[0046] In the above scheme, the fourth information includes at least the identifier of the second model.

[0047] In the above scheme, the fourth information also includes at least one of the following:

[0048] The type of the second model;

[0049] The second model supports AI platforms;

[0050] The second model supports the following AI frameworks;

[0051] The second model supports the following equipment manufacturers;

[0052] Does the second model support model sharing?

[0053] Does the second model support cross-platform model format conversion?

[0054] Does the second model support cross-language model format conversion?

[0055] The conversion formats supported by the second model.

[0056] In the above scheme, the fifth piece of information includes at least one of the following:

[0057] The address of the second functional unit;

[0058] The identifier of the second functional body.

[0059] In the above scheme, obtaining the first model from the first functional body includes:

[0060] Send a sixth message to the first functional body, the sixth message being used to request or subscribe to the first model;

[0061] Receive the seventh information sent by the first functional unit, wherein the seventh information includes one of the following:

[0062] The first model;

[0063] The Uniform Resource Locator (URL) corresponding to the first model.

[0064] The method in the above scheme further includes:

[0065] Receive the eighth message sent by AnLF, the eighth message being used to request or subscribe to the second model;

[0066] Send a ninth message to the AnLF, the ninth message including one of the following:

[0067] The second model;

[0068] The URL corresponding to the second model.

[0069] This application also provides a communication method applied to NRF, including:

[0070] Receive the tenth information sent by the first functional body, the tenth information being used for service registration of at least one model pre-trained by the second MTLF;

[0071] The tenth information is used to register services for at least one model pre-trained by the second MTLF;

[0072] Receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF;

[0073] Send a second message to the first MTLF, the second message indicating that the first model is obtained from the first functional body, so that the first MTLF can perform model fine-tuning on the first model.

[0074] In the above scheme, the first functional body includes the second MTLF or ADRF.

[0075] In the above scheme, the tenth information includes at least the identifier of the at least one model.

[0076] In the above scheme, the tenth information also includes at least one of the following:

[0077] The address of the first functional unit;

[0078] The identifier of the first functional unit;

[0079] The type of at least one model;

[0080] The AI ​​platform supported by at least one model;

[0081] The AI ​​framework supported by at least one model;

[0082] The at least one model supports the equipment manufacturers;

[0083] Does the at least one model support model sharing?

[0084] Does the at least one model support cross-platform model format conversion?

[0085] Does the at least one model support cross-language model format conversion?

[0086] The conversion format supported by at least one model.

[0087] In the above scheme, the first information includes at least the identifier of the first model.

[0088] In the above scheme, the first information further includes at least one of the following:

[0089] The type of the first model;

[0090] The AI ​​platforms supported by the first model;

[0091] The AI ​​framework supported by the first model;

[0092] The first model supports the following device manufacturers;

[0093] Does the first model support model sharing?

[0094] Does the first model support cross-platform model format conversion?

[0095] Does the first model support cross-language model format conversion?

[0096] The conversion formats supported by the first model.

[0097] In the above scheme, the second information includes at least one of the following:

[0098] The address of the first functional unit;

[0099] The identifier of the first functional unit.

[0100] The method in the above scheme further includes:

[0101] Receive third information sent by the second functional body, the third information being used to register the second model as a service, the second model being obtained by the first MTLF by fine-tuning the first model;

[0102] The second model is registered for service using the third information;

[0103] Receive the fourth message sent by AnLF, the fourth message being used to request the discovery of the second model;

[0104] Send a fifth message to the AnLF, the fifth message indicating that the second model is obtained from the second functional body.

[0105] In the above scheme, the second functional body includes the first MTLF or ADRF.

[0106] In the above scheme, the third information includes at least the identifier of the second model.

[0107] In the above scheme, the third information also includes at least one of the following:

[0108] The address of the second functional unit;

[0109] The identifier of the second functional entity;

[0110] The type of the second model;

[0111] The second model supports AI platforms;

[0112] The second model supports the following AI frameworks;

[0113] The second model supports the following equipment manufacturers;

[0114] Does the second model support model sharing?

[0115] Does the second model support cross-platform model format conversion?

[0116] Does the second model support cross-language model format conversion?

[0117] The conversion formats supported by the second model.

[0118] In the above scheme, the fourth information includes at least the identifier of the second model.

[0119] In the above scheme, the fourth information also includes at least one of the following:

[0120] The type of the second model;

[0121] The second model supports AI platforms;

[0122] The second model supports the following AI frameworks;

[0123] The second model supports the following equipment manufacturers;

[0124] Does the second model support model sharing?

[0125] Does the second model support cross-platform model format conversion?

[0126] Does the second model support cross-language model format conversion?

[0127] The conversion formats supported by the second model.

[0128] In the above scheme, the fifth piece of information includes at least one of the following:

[0129] The address of the second functional unit;

[0130] The identifier of the second functional body.

[0131] This application also provides a communication method applied to AnLF, including:

[0132] Send a fourth message to the NRF, the fourth message being used to request the discovery of a second model, the second model being obtained by fine-tuning the first model by the first MTLF, the first model including at least one model pre-trained by the second MTLF;

[0133] Receive the fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body;

[0134] The second model is obtained from the second functional body to perform inference analysis.

[0135] In the above scheme, the second functional body includes the first MTLF or ADRF.

[0136] In the above scheme, the fourth information includes at least the identifier of the second model.

[0137] In the above scheme, the fourth information also includes at least one of the following:

[0138] The type of the second model;

[0139] The second model supports AI platforms;

[0140] The second model supports the following AI frameworks;

[0141] The second model supports the following equipment manufacturers;

[0142] Does the second model support model sharing?

[0143] Does the second model support cross-platform model format conversion?

[0144] Does the second model support cross-language model format conversion?

[0145] The conversion formats supported by the second model.

[0146] In the above scheme, the fifth piece of information includes at least one of the following:

[0147] The address of the second functional unit;

[0148] The identifier of the second functional body.

[0149] In the above scheme, obtaining the second model from the second functional body includes:

[0150] Send an eighth message to the second functional body, the eighth message being used to request or subscribe to the second model;

[0151] Receive the ninth message sent by the second functional body, the ninth message including one of the following:

[0152] The second model;

[0153] The URL corresponding to the second model.

[0154] This application also provides a communication device, disposed on a first MTLF, including:

[0155] The first acquisition unit is configured to acquire a first model from a first functional body, wherein the first model includes at least one model pre-trained by a second MTLF.

[0156] The first processing unit is used to fine-tune the first model to obtain a second model, which is used by AnLF to perform inference analysis.

[0157] This application also provides a communication device, disposed on an NRF, including:

[0158] The first receiving unit is used to receive the tenth information sent by the first functional body, the tenth information being used to perform service registration for at least one model pre-trained by the second MTLF.

[0159] The second processing unit is used to perform service registration on at least one model pre-trained by the second MTLF using the tenth information;

[0160] The second receiving unit is used to receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF;

[0161] The first sending unit is configured to send second information to the first MTLF, the second information indicating that the first model is obtained from the first functional body, so that the first MTLF can perform model fine-tuning on the first model.

[0162] This application also provides a communication device, disposed on AnLF, including:

[0163] The second sending unit is used to send fourth information to the NRF, the fourth information being used to request the discovery of a second model, the second model being obtained by the first MTLF by fine-tuning the first model, and the first model containing at least one model pre-trained by the second MTLF.

[0164] The third receiving unit is used to receive the fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body;

[0165] The second acquisition unit is used to acquire the second model from the second functional body in order to perform inference analysis.

[0166] This application embodiment also provides a first MTLF, including: a first communication interface and a first processor; wherein,

[0167] The first processor is configured to obtain a first model from a first functional unit through the first communication interface, the first model including at least one model pre-trained by a second MTLF; and to fine-tune the first model to obtain a second model, the second model being used for inference analysis by AnLF.

[0168] This application also provides an NRF, including:

[0169] The second communication interface is used to receive the tenth information sent by the first functional body, the tenth information being used to register services for at least one model pre-trained by the second MTLF.

[0170] The second processor is used to perform service registration on at least one model pre-trained by the second MTLF using the tenth information;

[0171] The second communication interface is further configured to receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF; and to send second information to the first MTLF, the second information indicating that the first model is obtained from the first functional body, so that the first MTLF can fine-tune the first model.

[0172] This application also provides an AnLF, including:

[0173] The third communication interface is used to send fourth information to the NRF, the fourth information being used to request the discovery of a second model, the second model being obtained by the first MTLF by fine-tuning the first model, the first model including at least one model pre-trained by the second MTLF; and to receive fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body;

[0174] A third processor is configured to obtain the second model from the second functional body via the third communication interface in order to perform inference analysis.

[0175] This application also provides a first MTLF, including: a first processor and a first memory for storing a computer program capable of running on the processor.

[0176] Wherein, when the first processor is used to run the computer program, it executes the steps of any of the methods of the first MTLF side described above.

[0177] This application also provides an NRF, including: a second processor and a second memory for storing a computer program capable of running on the processor.

[0178] Wherein, when the second processor is used to run the computer program, it executes the steps of any of the above-described NRF-side methods.

[0179] This application also provides an AnLF, including: a third processor and a third memory for storing a computer program capable of running on the processor.

[0180] The third processor is used to execute any of the steps of the AnLF side method when running the computer program.

[0181] This application embodiment also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any of the above-described first MTLF side methods, or implements the steps of any of the above-described NRF side methods, or implements the steps of any of the above-described AnLF side methods.

[0182] The communication method, apparatus, related devices, and storage medium provided in this application embodiment involve a first MTLF obtaining a first model from a first functional unit. The first model includes at least one model pre-trained by a second MTLF. The first model is then fine-tuned to obtain a second model, which is used by the AnLF to perform inference analysis. The solution provided in this application embodiment divides the function of the MTLF into a first MTLF supporting model fine-tuning and a second MTLF providing pre-trained models. The first MTLF obtains the pre-trained model (i.e., the first model) from the first functional unit and performs model fine-tuning to obtain a model (i.e., the second model) that can adapt to the application requirements, data types, platform types, computing power, performance, and other requirements of the AnLF. Since the first MTLF is responsible for obtaining the target model (i.e., the second model) that can adapt to the requirements of AnLF, the second MTLF can be pre-trained in an open-source manner. Furthermore, the model pre-trained by the second MTLF obtained by the first MTLF can be free of privacy data related to the second MTLF (such as model privacy data and / or training privacy data). In other words, cross-platform and / or cross-vendor model sharing between the first MTLF and the second MTLF can be achieved without leaking privacy data. Therefore, when there is a need for cross-platform and / or cross-vendor model sharing between MTLF and AnLF, by deploying the first MTLF locally on the AnLF side (i.e., deploying the first MTLF and AnLF on the same platform and / or the same vendor), cross-platform and / or cross-vendor model sharing between AnLF and MTLF (i.e., the second MTLF) can be achieved, effectively protecting model privacy and intellectual property rights.

[0183] Meanwhile, since the first MTLF only needs to fine-tune the model and does not need to perform the entire model training process, it can save a lot of computing resources and computing time, and improve computing efficiency (i.e., the local model training efficiency of the first MTLF).

[0184] In addition, for certain specific situations (such as when the first MTLF has limited local sample data), the first MTLF can optimize the model pre-trained by the second MTLF (i.e., fine-tune the model), which can improve the accuracy of the first MTLF's local model training. Attached Figure Description

[0185] Figure 1 This is a flowchart illustrating a communication method according to an embodiment of this application;

[0186] Figure 2 This is a flowchart illustrating another communication method according to an embodiment of this application;

[0187] Figure 3 This is a flowchart illustrating the third communication method according to an embodiment of this application;

[0188] Figure 4 This application provides a schematic diagram of a model sharing process as an example.

[0189] Figure 5 This is another schematic diagram illustrating the model sharing process used in this application.

[0190] Figure 6 This is a schematic diagram of a communication device structure according to an embodiment of this application;

[0191] Figure 7 This is a schematic diagram of another communication device structure according to an embodiment of this application;

[0192] Figure 8 This is a schematic diagram of the third type of communication device structure according to an embodiment of this application;

[0193] Figure 9 This is a schematic diagram of the first MTLF structure according to an embodiment of this application;

[0194] Figure 10 This is a schematic diagram of the NRF structure in an embodiment of this application;

[0195] Figure 11 This is a schematic diagram of the AnLF structure according to an embodiment of this application;

[0196] Figure 12 This is a schematic diagram of the communication system structure according to an embodiment of this application. Detailed Implementation

[0197] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0198] In related technologies, there are two ways to achieve cross-platform and / or cross-vendor model sharing between MTLF and AnLF. One approach is to define a common model description language between AnLF and MTLF to break down AI framework barriers, meaning that AnLF and MTLF use the same model format, such as Open Neural Network Exchange (ONNX) or Predictive Model Markup Language (PMML). The other approach is that AnLF and MTLF can achieve interoperability through interoperability support identifiers (such as vendor identifiers), that is, the ability to share models is determined by verifying the interoperability support identifiers.

[0199] However, while the first approach enables cross-platform and / or cross-vendor model sharing between MTLF and AnLF based on model consistency description (i.e., a universal description language for models), it still requires the transfer of MTLF-trained model data between MTLF and AnLF. In other words, the risk of privacy data (such as model privacy data and / or training privacy data) leakage remains. As for the second approach, which achieves interoperability between AnLF and MTLF based on interoperability support identifiers, in practical applications and deployments, considering factors such as user privacy and intellectual property rights related to model training, it essentially only promotes interoperability between AnLF and MTLF from the same vendor, and still cannot promote interoperability between AnLF and MTLF from different vendors and / or different platforms.

[0200] As can be seen from the above description, the relevant technologies cannot effectively avoid the leakage of privacy data when realizing cross-platform and / or cross-vendor model sharing between MTLF and AnLF.

[0201] Based on this, in various embodiments of this application, the function of MTLF is divided into a first MTLF that supports model fine-tuning and a second MTLF that provides pre-trained models. The first MTLF obtains the pre-trained model (i.e., the first model) from the first functional body and performs model fine-tuning to obtain a model (i.e., the second model) that can adapt to the application requirements, data types, platform types, computing power, performance and other requirements of AnLF. Since the first MTLF is responsible for obtaining the target model (i.e., the second model) that can adapt to the requirements of AnLF, the second MTLF can be pre-trained in an open-source manner. Furthermore, the model pre-trained by the second MTLF obtained by the first MTLF can be free of privacy data related to the second MTLF (such as model privacy data and / or training privacy data). In other words, cross-platform and / or cross-vendor model sharing between the first MTLF and the second MTLF can be achieved without leaking privacy data. Therefore, when there is a need for cross-platform and / or cross-vendor model sharing between MTLF and AnLF, by deploying the first MTLF locally on the AnLF side (i.e., deploying the first MTLF and AnLF on the same platform and / or the same vendor), cross-platform and / or cross-vendor model sharing between AnLF and MTLF (i.e., the second MTLF) can be achieved, effectively protecting model privacy and intellectual property rights.

[0202] Meanwhile, since the first MTLF only needs to fine-tune the model and does not need to perform the entire model training process, it can save a lot of computing resources and computing time, and improve computing efficiency (i.e., the local model training efficiency of the first MTLF).

[0203] In addition, for certain specific situations (such as when the first MTLF has limited local sample data), the first MTLF can optimize the model pre-trained by the second MTLF (i.e., fine-tune the model), which can improve the accuracy of the first MTLF's local model training.

[0204] This application provides a communication method applied to a first MTLF, such as... Figure 1 As shown, the method includes:

[0205] Step 101: Obtain a first model from the first functional body, wherein the first model contains at least one model pre-trained by the second MTLF;

[0206] Step 102: Fine-tune the first model to obtain the second model, which is used by AnLF to perform inference analysis.

[0207] The second model is used for inference analysis by AnLF, which can be understood as AnLF using the second model for inference analysis; the fine-tuning of the first model can be understood as optimizing or updating the first model so that the obtained second model meets AnLF's application requirements, data types, platform types, computing power, performance and other requirements. In other words, fine-tuning the first model means using specific data stored locally (such as local training sample data and application requirements, data types, platform types, computing power, performance and other requirements information obtained and saved from AnLF) to further train the first model.

[0208] In practical applications, the first MTLF can be represented as MTLF (Finetune); the second MTLF can be represented as MTLF (Pre-train).

[0209] In practical applications, the second MTLF can be pre-trained using an open-source approach. This avoids the problem of incompatibility between AI frameworks, model formats, and model description languages, which prevents cross-platform and / or cross-vendor model sharing. Furthermore, it ensures that the pre-trained model does not contain privacy data related to the second MTLF, such as model privacy data (e.g., methods for model parameter tuning) and / or training privacy data (i.e., user privacy information potentially involved in the sample data used to train the model). Here, "vendor" can be understood as a company that produces electronic devices that support network functions (i.e., MTLF, AnLF), and "platform" can be understood as the AI ​​platform or framework supported by the model.

[0210] In practical applications, since the second MLTF pre-trained model needs optimization (i.e. fine-tuning, or update) of the first MTLF to obtain the target model (i.e. the second model) that meets the AnLF requirements, the second MTLF pre-trained model can be understood as the model to be optimized, that is, an incomplete model that has not reached the ideal training objective (also known as a model fragment); in other words, at least one model pre-trained by the second MTLF can also be understood as at least one model fragment pre-trained by the second MTLF.

[0211] In practical applications, the second MTLF can update the local pre-trained model according to a pre-set period. In other words, the second MTLF can periodically pre-train the model.

[0212] In practical applications, the first functional unit may include the second MTLF, meaning the first MTLF can directly obtain the first model from the second MTLF. For example, the first MTLF may send a request to the second MTLF to obtain the first model, and upon receiving the request, the second MTLF may send the first model to the first MTLF; alternatively, the second MTLF may directly send the first model to the first MTLF after completing its pre-training.

[0213] In practical applications, the first functional unit may also include ADRF, that is, the second MTLF can store all pre-trained models in ADRF, and the first MTLF can obtain the first model from the ADRF.

[0214] In practical applications, the specific implementation of the first MTLF obtaining the first model from the first functional entity can be achieved through the NRF service registration and service discovery process. Specifically, after the second MTLF completes the pre-training of at least one model, it can initiate service registration for the pre-trained at least one model with the NRF. Alternatively, the second MTLF can send the pre-trained at least one model to the ADRF, which will then initiate service registration for the pre-trained at least one model with the NRF. That is, the first functional entity (the second MTLF or the ADRF) can send tenth information to the NRF, which is used for service registration of the pre-trained at least one model of the second MTLF. After receiving the tenth information, the NRF can use it to register the pre-trained at least one model of the second MTLF for service.

[0215] The tenth piece of information may include at least the identifier of the at least one model. In practical applications, the identifier of the at least one model may be represented by the identifier (e.g., ID) of the service (which may be expressed as Analytics) corresponding to the at least one model.

[0216] In practical applications, the tenth piece of information may also include at least one of the following:

[0217] The address of the first functional entity (such as an Internet Protocol (IP) address, URL, etc.);

[0218] The identifier (e.g., ID) of the first functional entity;

[0219] The type of at least one model (e.g., whether it is a pre-trained model, etc.);

[0220] The at least one model supports an AI platform (such as the Google AI platform);

[0221] The at least one model supports an AI framework (such as MindSpore);

[0222] The at least one model supports equipment manufacturers (such as ZTE);

[0223] Does the at least one model support model sharing?

[0224] Does the at least one model support cross-platform model format conversion (i.e., model format conversion when the AI ​​platforms supported by the second MTLF and the AnLF are different)?

[0225] Does the at least one model support cross-language model format conversion (i.e., model format conversion when the second MTLF and the AnLF support different model languages)?

[0226] The at least one model supports conversion formats (such as ONNX, PMML, etc.).

[0227] In practical applications, when the first functional entity initiates service registration with the NRF, the relationship between services and models can be one-to-one, meaning one pre-trained model is used to provide one service; or, the relationship between services and models can be many-to-one, meaning one pre-trained model is used to provide multiple services (i.e., at least two services); or, the relationship between services and models can be one-to-many, meaning at least two pre-trained models are used to provide one service. The specific correspondence between services and models can be determined according to requirements, and this application embodiment does not limit this.

[0228] In practical applications, when the first functional body includes ADRF, the tenth information may include the address and / or identifier (e.g., ADRF ID) of the first functional body; when the first functional body does not include ADRF, the tenth information may not include the address and / or identifier of the first functional body.

[0229] In practical applications, the first MTLF can initiate a service discovery request corresponding to the first model to the NRF as needed. After receiving the request, the NRF can instruct the first MTLF on how to obtain the first model (i.e., instruct the storage location of the first model).

[0230] Based on this, in one embodiment, obtaining the first model from the first functional body may include:

[0231] Send a first message to the NRF, the first message being used to request the discovery of a first model;

[0232] Receive second information sent by the NRF, the second information indicating that the first model is obtained from the first functional body;

[0233] The first model is obtained from the first functional body; in other words, after receiving the second information, the first model is obtained from the first functional body based on the second information.

[0234] The first information may include at least the identifier of the first model. In practical applications, the identifier of the first model may be the identifier of the service corresponding to the first model.

[0235] In practical applications, the first information may also include at least one of the following:

[0236] The type of the first model (e.g., whether it is a pre-trained model, etc.);

[0237] The AI ​​platforms supported by the first model;

[0238] The AI ​​framework supported by the first model;

[0239] The first model supports the following device manufacturers;

[0240] Does the first model support model sharing?

[0241] Does the first model support cross-platform model format conversion?

[0242] Does the first model support cross-language model format conversion?

[0243] The conversion formats supported by the first model.

[0244] In practical applications, the second information may include at least one of the following:

[0245] The address of the first functional unit (such as IP address, URL, etc.);

[0246] The identifier of the first functional unit.

[0247] In practical applications, the first MTLF can request or subscribe to the first model from the first functional body based on the second information.

[0248] Based on this, in one embodiment, obtaining the first model from the first functional body may include:

[0249] Send a sixth message to the first functional body, the sixth message being used to request or subscribe to the first model;

[0250] Receive the seventh information sent by the first functional unit, wherein the seventh information includes one of the following:

[0251] The first model;

[0252] The URL corresponding to the first model.

[0253] In practical applications, the sixth piece of information may include at least the identifier of the first model.

[0254] In practical applications, the sixth piece of information may also include the type of the first model.

[0255] In practical applications, the specific method by which the first MTLF fine-tunes the first model can be determined according to the requirements of the AnLF, so that the second model can adapt to the application requirements, data types, platform types, computing power, performance, and other requirements of the AnLF. For example, the AnLF can pre-send the application requirements, data types, platform types, computing power, performance, and other requirement information to the first MTLF, so that after the first MTLF obtains the first model from the first functional unit, it can use this requirement information and local training sample data to fine-tune the first model to obtain the second model.

[0256] In practical applications, after the first MTLF obtains the second model, it can directly send the second model to the AnLF.

[0257] Based on this, in one embodiment, the method may further include:

[0258] Send the second model to the AnLF.

[0259] In practical applications, after the first MTLF obtains the second model, it can store the second model in the ADRF, and the AnLF can obtain the second model from the ADRF.

[0260] Based on this, in one embodiment, the method may further include:

[0261] The second model is sent to the ADRF so that the AnLF can obtain the second model from the ADRF.

[0262] In practical applications, after the first MTLF obtains the second model, it can provide the second model to the AnLF through the NRF's service registration and service discovery processes. Specifically, the first MTLF can initiate service registration for the second model with the NRF, or the first MTLF can send the second model to the ADRF, which will then initiate service registration for the second model with the NRF. In other words, the second functional entity (i.e., the first MTLF or ADRF) can send third information to the NRF, which is used for service registration of the second model. After receiving the third information, the NRF can use it to register the second model for service. Subsequently, the AnLF can initiate a service discovery request corresponding to the second model to the NRF as needed. After receiving the request, the NRF can instruct the AnLF on how to obtain the second model (i.e., indicate the storage location of the second model).

[0263] Based on this, in one embodiment, the method may further include:

[0264] A third message is sent to the NRF so that when the NRF receives the fourth message sent by the AnLF, it can send a fifth message to the AnLF; the third message is used to register the second model for service; the fourth message is used to request the discovery of the second model; and the fifth message indicates that the second model is obtained from the second functional body.

[0265] The second functional unit may include the first MTLF or ADRF.

[0266] In practical applications, the third information may include at least the identifier of the second model; the identifier of the second model may be reflected as the identifier of the service corresponding to the second model.

[0267] In practical applications, the third information may also include at least one of the following:

[0268] The address of the second functional unit (such as IP address, URL, etc.);

[0269] The identifier of the second functional entity;

[0270] The type of the second model (e.g., whether it is the target model (i.e., a model that can adapt to AnLF requirements));

[0271] The second model supports AI platforms;

[0272] The second model supports the following AI frameworks;

[0273] The second model supports the following equipment manufacturers;

[0274] Does the second model support model sharing?

[0275] Does the second model support cross-platform model format conversion?

[0276] Does the second model support cross-language model format conversion?

[0277] The conversion formats supported by the second model.

[0278] In practical applications, when the second functional body includes ADRF, the third information may include the address and / or identifier (e.g., ADRF ID) of the second functional body; when the second functional body does not include ADRF, the third information may not include the address and / or identifier of the second functional body.

[0279] In practical applications, the fourth information may include at least the identifier of the second model; the identifier of the second model may be reflected as the identifier of the service corresponding to the second model.

[0280] In practical applications, the fourth information may also include at least one of the following:

[0281] The type of the second model;

[0282] The second model supports AI platforms;

[0283] The second model supports the following AI frameworks;

[0284] The second model supports the following equipment manufacturers;

[0285] Does the second model support model sharing?

[0286] Does the second model support cross-platform model format conversion?

[0287] Does the second model support cross-language model format conversion?

[0288] The conversion formats supported by the second model.

[0289] In practical applications, the fifth piece of information may include at least one of the following:

[0290] The address of the second functional entity, such as IP address, URL, etc.;

[0291] The identifier of the second functional body.

[0292] As can be seen from the above description, in various embodiments of this application, the information used to initiate the NRF service registration process (the tenth and third information) and the information used to initiate the NRF service discovery process (the first and fourth information) both include the model identifier (which can be represented as the identifier of the service corresponding to the model). Therefore, in practical applications, in order to ensure the consistency of the model identifier when the various functional entities (i.e., the first MTLF, the second MTLF, and the AnLF) interact, the first MTLF, the second MTLF, and the AnLF can adopt the same strategy or rules to determine the model identifier. For example, the first MTLF, the second MTLF, and the AnLF can pre-agree to use the model's function to determine the model identifier.

[0293] Here, it can be understood that for the information used to initiate the NRF service discovery process, when the NRF determines that no corresponding service is registered locally, it can return a message refusing to provide the service. That is, after receiving the first information, the NRF can determine whether it has registered at least one model pre-trained by the second MTLF for service based on the identifier of the first model contained in the first information. If it determines that at least one model pre-trained by the second MTLF has been registered for service, it sends the second information to the first MTLF; if it determines that at least one model pre-trained by the second MTLF has not been registered for service, it sends a message refusing to provide the service corresponding to the first model to the first MTLF. Similarly, after receiving the fourth information, the NRF can determine whether it has registered the second model for service based on the identifier of the second model contained in the fourth information. If it determines that the second model has been registered for service, it sends the fifth information to the AnLF; if it determines that the second model has not been registered for service, it sends a message refusing to provide the service corresponding to the second model to the AnLF.

[0294] In practical applications, after receiving the fifth information, the AnLF can request or subscribe to the second model from the second functional entity based on the fifth information. Specifically, the AnLF can send an eighth information to the second functional entity, which is used to request or subscribe to the second model; after receiving the eighth information, the second functional entity can send a ninth information to the AnLF, which may include one of the following:

[0295] The second model;

[0296] The URL corresponding to the second model.

[0297] Based on this, when the second functional body includes the first MTLF, the method may further include:

[0298] Receive the eighth message sent by AnLF, the eighth message being used to request or subscribe to the second model;

[0299] Send a ninth message to the AnLF, the ninth message including one of the following:

[0300] The second model;

[0301] The URL corresponding to the second model.

[0302] In practical applications, the eighth piece of information may include at least the identifier of the second model; the identifier of the second model may be reflected as the identifier of the service corresponding to the second model.

[0303] In practical applications, the eighth piece of information may also include at least one of the following:

[0304] The second model supports AI platforms;

[0305] The second model supports the following AI frameworks;

[0306] The second model supports the following equipment manufacturers;

[0307] Does the second model support model sharing?

[0308] Does the second model support cross-platform model format conversion?

[0309] Does the second model support cross-language model format conversion?

[0310] The conversion formats supported by the second model.

[0311] In practical applications, the first MTLF can communicate with the NRF via the Nnrf interface (e.g., transmitting the first, second, and third information); the first MTLF can communicate with the second MTLF via the Nnwdaf interface (e.g., transmitting the sixth and seventh information); the first MTLF can communicate with the AnLF via the Nnwdaf interface (e.g., transmitting the eighth and ninth information); the first MTLF can communicate with the ADRF via the Nadrf interface (e.g., transmitting the sixth and seventh information); the second MTLF can communicate with the NRF via the Nnrf interface (e.g., transmitting the tenth information); the second MTLF can communicate with the ADRF via the Nadrf interface (e.g., transmitting at least one model pre-trained by the second MTLF); the NRF can communicate with the AnLF via the Nnrf interface (e.g., transmitting the fourth and fifth information); the NRF can communicate with the ADRF via the Nnrf interface (e.g., transmitting the tenth information); and the AnLF can communicate with the ADRF via the Nadrf interface (e.g., transmitting the eighth and ninth information).

[0312] Accordingly, embodiments of this application also provide a communication method applied to NRF, such as... Figure 2 As shown, the method includes:

[0313] Step 201: Receive the tenth information sent by the first functional body, the tenth information being used for service registration of at least one model pre-trained by the second MTLF;

[0314] Step 202: Use the tenth information to perform service registration on at least one model pre-trained by the second MTLF;

[0315] Step 203: Receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF;

[0316] Step 204: Send a second message to the first MTLF, the second message indicating that the first model is obtained from the first functional body, so that the first MTLF can perform model fine-tuning on the first model.

[0317] The first functional unit may include the second MTLF or ADRF.

[0318] In one embodiment, the method may further include:

[0319] Receive third information sent by the second functional body, the third information being used to register the second model as a service, the second model being obtained by the first MTLF by fine-tuning the first model;

[0320] The second model is registered for service using the third information;

[0321] Receive the fourth message sent by AnLF, the fourth message being used to request the discovery of the second model;

[0322] Send a fifth message to the AnLF, the fifth message indicating that the second model is obtained from the second functional body.

[0323] The second functional unit may include the first MTLF or ADRF.

[0324] Accordingly, embodiments of this application also provide a communication method applied to AnLF, such as... Figure 3 As shown, the method includes:

[0325] Step 301: Send a fourth message to the NRF, the fourth message being used to request the discovery of a second model, the second model being obtained by the first MTLF by fine-tuning the first model, the first model containing at least one model pre-trained by the second MTLF;

[0326] Step 302: Receive the fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body;

[0327] Step 303: Obtain the second model from the second functional body to perform inference analysis.

[0328] The second functional unit may include the first MTLF or ADRF.

[0329] In one embodiment, obtaining the second model from the second functional body may include:

[0330] Send an eighth message to the second functional body, the eighth message being used to request or subscribe to the second model;

[0331] Receive the ninth message sent by the second functional body, the ninth message including one of the following:

[0332] The second model;

[0333] The URL corresponding to the second model.

[0334] The communication method provided in this application embodiment involves a first MTLF obtaining a first model from a first functional entity. The first model includes at least one model pre-trained by a second MTLF. The first model is then fine-tuned to obtain a second model, which is used by the AnLF to perform inference analysis. The solution provided in this application embodiment divides the function of the MTLF into a first MTLF supporting model fine-tuning and a second MTLF providing pre-trained models. The first MTLF obtains the pre-trained model (i.e., the first model) from the first functional entity and performs model fine-tuning to obtain a model (i.e., the second model) that can adapt to the application requirements, data types, platform types, computing power, performance, and other requirements of the AnLF. Since the first MTLF is responsible for obtaining the target model (i.e., the second model) that can adapt to the requirements of AnLF, the second MTLF can be pre-trained in an open-source manner. Furthermore, the model pre-trained by the second MTLF obtained by the first MTLF can be free of privacy data related to the second MTLF (such as model privacy data and / or training privacy data). In other words, cross-platform and / or cross-vendor model sharing between the first MTLF and the second MTLF can be achieved without leaking privacy data. Therefore, when there is a need for cross-platform and / or cross-vendor model sharing between MTLF and AnLF, by deploying the first MTLF locally on the AnLF side (i.e., deploying the first MTLF and AnLF on the same platform and / or the same vendor), cross-platform and / or cross-vendor model sharing between AnLF and MTLF (i.e., the second MTLF) can be achieved, effectively protecting model privacy and intellectual property rights.

[0335] Meanwhile, since the first MTLF only needs to fine-tune the model and does not need to perform the entire model training process, it can save a lot of computing resources and computing time, and improve computing efficiency (i.e., the local model training efficiency of the first MTLF).

[0336] In addition, for certain specific situations (such as when the first MTLF has limited local sample data), the first MTLF can optimize the model pre-trained by the second MTLF (i.e., fine-tune the model), which can improve the accuracy of the first MTLF's local model training.

[0337] The following section provides a more detailed description of this application with reference to application examples.

[0338] This application example provides a new model sharing method, dividing MTLF into two types: one is MTLF (Pre-train) which provides pre-trained models (i.e., the second MTLF), and the other is MTLF (Finetune) which supports model fine-tuning (i.e., the first MTLF). In other words, NWDAF supports two training services: MTLF (Pre-train) model pre-training and MTLF (Finetune) model fine-tuning. MTLF (Pre-train) provides MTLF (Finetune) with a model pre-trained in an open-source manner, including the basic architecture, parameters, algorithm, or model fragments (the model fragments refer to the pre-trained model itself, i.e., the model to be optimized that does not meet AnLF requirements). MTLF (Finetune) then fine-tunes the pre-trained model (i.e., the first model) to form a target model (i.e., the second model) that meets AnLF's application requirements, data types, platform types, computing power, performance, etc. MTLF (Finetune) then further provides the fine-tuned model to AnLF.

[0339] The following is combined with Figure 4 This describes a model sharing process for this application example.

[0340] like Figure 4 As shown in this application example, model sharing can be achieved based on service registration of a pre-trained model. Specifically, the model sharing process may include the following steps:

[0341] Step 401: MTLF (Pre-train) performs service registration for the pre-trained model, and then proceeds to step 402;

[0342] Step 402: MTLF (Finetune) and NRF perform service discovery and response using the pre-trained model, then proceed to step 403;

[0343] Step 403: MTLF (Finetune) requests or subscribes to the pre-trained model from MTLF (Pre-train), and then proceeds to step 404;

[0344] Step 404: MTLF (Pre-train) response pre-trained model, then proceed to step 405;

[0345] Step 405: MTLF (Finetune) performs model fine-tuning to generate a target model adapted to local inference needs, and then proceeds to step 406;

[0346] Step 406: MTLF (Finetune) performs service registration for the target model, and then proceeds to step 407;

[0347] Step 407: AnLF and NRF perform service discovery and response for the target model, and then proceed to step 408;

[0348] Step 408: AnLF requests or subscribes to the target model from MTLF (Finetune), and then proceeds to step 409;

[0349] Step 409: MTLF (Finetune) response target model.

[0350] In step 401, the MTLF (Pre-train) can initiate service registration of the pre-trained model with the NRF, that is, send service registration information (i.e., the tenth information mentioned above) to the NRF. The service registration information can include relevant information about the pre-trained model, and can at least include the Analytics ID. In practical applications, the service registration information can also include one or more of the following information: model type (e.g., whether the model is a pre-trained model or a target model), AI platform information, AI framework information, device manufacturer information, whether the model supports model sharing, whether it supports cross-platform model format conversion, whether it supports cross-language model format conversion, and supported conversion formats (e.g., ONNX, PMML, etc.).

[0351] In step 402, MTLF (Finetune) can initiate (i.e. send) a service discovery request for the pre-trained model (i.e., the first information) to NRF; this request can include at least the Analytics ID. In practical applications, the request can also include one or more of the following information: model type (e.g., whether the model is a pre-trained model or a target model), AI platform information, AI framework information, device manufacturer information, whether the model supports model sharing, whether it supports cross-platform model format conversion, whether it supports cross-language model format conversion, and supported conversion formats (e.g., ONNX, PMML, etc.). After receiving the request, NRF can respond to MTLF (Finetune) by feeding back the storage location information of the pre-trained model that meets MTLF's (Finetune's) requirements (i.e., the second information), such as the address and / or identifier of the MTLF (Pre-train).

[0352] In step 403, MTLF (Finetune) can send information to MTLF (Pre-train) for requesting or subscribing to a pre-trained model (i.e., the sixth information), which may include at least the Analytics ID. In practical applications, this information may also include the model type (e.g., whether the model is a pre-trained model or the target model).

[0353] In step 404, MTLF (Pre-train) responds to MTLF (Finetune) with the pre-trained model (i.e., sends the seventh information), such as directly feeding back the pre-trained model (i.e., a model fragment) or feeding back a URL storing the pre-trained model; MTLF (Finetune) obtains the pre-trained model.

[0354] In step 405, MTLF (Finetune) can fine-tune the pre-trained model based on the application requirements, data types, platform types, computing power, performance requirements, and local training sample data of the pre-obtained AnLF.

[0355] In step 406, after MTLF (Finetune) completes the fine-tuning of the pre-trained model and obtains the target model, it can initiate service registration of the target model with the NRF, that is, send service registration information (i.e., the third information) to the NRF. The service registration information may contain relevant information about the target model, and may at least include the Analytics ID. In practical applications, the service registration information may also include one or more of the following: model type (e.g., whether the model is a pre-trained model or the target model), AI platform information, AI framework information, device manufacturer information, whether the model supports model sharing, whether it supports cross-platform model format conversion, whether it supports cross-language model format conversion, and supported conversion formats (e.g., ONNX, PMML, etc.).

[0356] In step 407, AnLF can initiate a service discovery request for the target model to NRF (i.e., the fourth information). This request can include at least the Analytics ID. In practical applications, the request can also include one or more of the following: model type (e.g., whether the model is a pre-trained model or a target model), AI platform information, AI framework information, device manufacturer information, whether the model supports model sharing, whether it supports cross-platform model format conversion, whether it supports cross-language model format conversion, and supported conversion formats (e.g., ONNX, PMML, etc.). After receiving the request, NRF can respond to AnLF by feeding back relevant information about the storage location of the target model that meets AnLF's requirements (i.e., the fifth information), such as the address and / or identifier of the MTLF (Finetune).

[0357] In step 408, AnLF can send information to MTLF (Finetune) for requesting or subscribing to the target model (i.e., the eighth information), which may include at least the Analytics ID. In practical applications, this information may also include one or more of the following: AI platform information, AI framework information, device manufacturer information, whether the model supports model sharing, whether it supports cross-platform model format conversion, whether it supports cross-language model format conversion, and supported conversion formats (such as ONNX, PMML, etc.).

[0358] In step 409, MTLF (Finetune) responds to AnLF with the target model (i.e., sends the ninth information), such as directly feeding back the target model or feeding back a URL storing the target model; AnLF obtains the target model and can use the target model to perform inference analysis.

[0359] The following is combined with Figure 5 This describes another model-sharing process for this application example.

[0360] like Figure 5 As shown in this application example, model sharing can be achieved based on the storage of pre-trained models in the ADRF. That is, one or more MTLFs (Pre-trained Modules) can store pre-trained models in the ADRF, and different MTLFs (Finetune Modules) can retrieve pre-trained models from the ADRF. Specifically, this model sharing process can include the following steps:

[0361] Step 501: MTLF (Pre-train) stores the pre-trained model into ADRF, and then proceeds to step 502;

[0362] Step 502: Perform service registration for the pre-trained model using MTLF (Pre-train) or ADRF, and then proceed to step 503;

[0363] Step 503: MTLF (Finetune) and NRF perform service discovery and response of the pre-trained model, and then proceed to step 504;

[0364] Step 504: MTLF (Finetune) requests or subscribes to the pre-trained model from ADRF, and then proceeds to step 505;

[0365] Step 505: Response the ADRF to the pre-trained model, then proceed with steps 405 through 409.

[0366] In step 501, one or more MTLFs (Pre-train) can store the pre-trained model (i.e. model fragments) into the ADRF.

[0367] In step 502, when the MTLF (Pre-train) initiates the service registration of the pre-trained model, the MTLF (Pre-train) can send the service registration information of the pre-trained model (i.e., the tenth information mentioned above) to the NRF. This service registration information may include relevant information about the pre-trained model, and at least the Analytics ID. In practical applications, the service registration information may also include one or more of the following: ADRF ID, model type (e.g., whether the model is a pre-trained model or a target model), AI platform information, AI framework information, device manufacturer information, whether the model supports model sharing, whether it supports cross-platform model format conversion, whether it supports cross-language model format conversion, and supported conversion formats (e.g., ONNX, PMML, etc.).

[0368] When ADRF initiates service registration of a pre-trained model, ADRF can send the service registration information of the pre-trained model (i.e., the tenth information mentioned above) to NRF. This service registration information can include relevant information about the pre-trained model, and at least the Analytics ID. In practical applications, the service registration information may also include one or more of the following: ADRF ID, model type (e.g., whether the model is a pre-trained model or a target model), AI platform information, AI framework information, device manufacturer information, whether the model supports model sharing, whether it supports cross-platform model format conversion, whether it supports cross-language model format conversion, and supported conversion formats (e.g., ONNX, PMML, etc.).

[0369] In step 503, MTLF (Finetune) can initiate (i.e. send) a service discovery request for the pre-trained model (i.e., the first information) to NRF; this request can include at least the Analytics ID. In practical applications, the request can also include one or more of the following information: model type (e.g., whether the model is a pre-trained model or a target model), AI platform information, AI framework information, device manufacturer information, whether the model supports model sharing, whether it supports cross-platform model format conversion, whether it supports cross-language model format conversion, and supported conversion formats (e.g., ONNX, PMML, etc.). After receiving the request, NRF can respond to MTLF (Finetune) by feeding back the storage location information of the pre-trained model that meets MTLF's (Finetune's) requirements (i.e., the second information), such as the ADRF's address and / or identifier.

[0370] In step 504, MTLF (Finetune) can send information to ADRF for requesting or subscribing to a pre-trained model (i.e., the sixth information), which may include at least the Analytics ID. In practical applications, this information may also include the model type (e.g., whether the model is a pre-trained model or the target model).

[0371] In step 505, ADRF can respond to MTLF (Finetune) with a pre-trained model (i.e., send the seventh information), such as directly feeding back the pre-trained model (i.e., a model fragment) or feeding back a URL storing the pre-trained model; MTLF (Finetune) obtains the pre-trained model.

[0372] The solution provided in this application example has the following advantages:

[0373] 1) By utilizing open-source AI (i.e., MTLF (Pre-train) uses an open-source approach for model pre-training) and local fine-tuning (i.e., MTLF (Finetune) fine-tunes the pre-trained model locally), the problem of cross-vendor and cross-platform model sharing can be effectively solved. It supports interoperability between MTLF (Pre-train) and MTLF (Finetune) from different vendors. Compared with the model sharing method of direct interaction between MTLF and AnLF in related technologies, the interaction of pre-trained models can effectively promote the interconnection and interoperability of cross-vendor and cross-enterprise model sharing while protecting data privacy and intellectual property rights.

[0374] 2) MTLF (Finetune) leverages the pre-trained machine learning model from MTLF (Pre-train) to further fine-tune the model by combining it with local specific tasks (such as local samples and AnLF requirements), which can save a lot of computing resources and computing time and improve the efficiency of local model training.

[0375] 3) For model training scenarios such as MTLF (Finetune) local small data samples, it can also improve the accuracy of MTLF (Finetune) local model training.

[0376] To implement the method on the first MTLF side of this application embodiment, this application embodiment also provides a communication device, disposed on the first MTLF, such as... Figure 6 As shown, the device includes:

[0377] The first acquisition unit 601 is used to acquire a first model from the first functional body, wherein the first model includes at least one model pre-trained by the second MTLF.

[0378] The first processing unit 602 is used to fine-tune the first model to obtain a second model, which is used for AnLF to perform inference analysis.

[0379] In one embodiment, such as Figure 6 As shown, the device may further include:

[0380] The third sending unit 603 is used to send first information to the NRF, the first information being used to request the discovery of the first model;

[0381] The fourth receiving unit 604 is used to receive the second information sent by the NRF, the second information indicating that the first model is obtained from the first functional body;

[0382] Accordingly, the first acquisition unit 601 is specifically used to acquire the first model from the first functional body based on the second information.

[0383] In one embodiment, the third transmitting unit 603 is further configured to:

[0384] Send the second model to the AnLF;

[0385] or,

[0386] Send the second model to the ADRF so that the AnLF can obtain the second model from the ADRF;

[0387] or,

[0388] A third message is sent to the NRF so that when the NRF receives the fourth message sent by the AnLF, it can send a fifth message to the AnLF; the third message is used to register the second model for service; the fourth message is used to request the discovery of the second model; and the fifth message indicates that the second model is obtained from the second functional body.

[0389] In one embodiment, the first acquisition unit 601 is specifically used for:

[0390] Send a sixth message to the first functional body, the sixth message being used to request or subscribe to the first model;

[0391] Receive the seventh information sent by the first functional unit, wherein the seventh information includes one of the following:

[0392] The first model;

[0393] The URL corresponding to the first model.

[0394] In one embodiment, the fourth receiving unit 604 is further configured to receive the eighth information sent by the AnLF, the eighth information being used to request or subscribe to the second model;

[0395] Accordingly, the third sending unit 603 is further configured to send a ninth message to the AnLF, the ninth message including one of the following:

[0396] The second model;

[0397] The URL corresponding to the second model.

[0398] In practical applications, the first acquisition unit 601 can be implemented by a processor in the communication device combined with a communication interface; the first processing unit 602 can be implemented by a processor in the communication device; and the third sending unit 603 and the fourth receiving unit 604 can be implemented by a communication interface in the communication device.

[0399] To implement the NRF-side method of this application embodiment, this application embodiment also provides a communication device disposed on the NRF, such as... Figure 7 As shown, the device includes:

[0400] The first receiving unit 701 is used to receive the tenth information sent by the first functional body, the tenth information being used to perform service registration for at least one model pre-trained by the second MTLF.

[0401] The second processing unit 702 is used to perform service registration on at least one model pre-trained by the second MTLF using the tenth information;

[0402] The second receiving unit 703 is used to receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF.

[0403] The first sending unit 704 is used to send second information to the first MTLF, the second information indicating that the first model is obtained from the first functional body so that the first MTLF can perform model fine-tuning on the first model.

[0404] In one embodiment, the second receiving unit 703 is further configured to receive third information sent by the second functional body, the third information being used to register a service for the second model, the second model being obtained by the first MTLF through model fine-tuning of the first model;

[0405] The second processing unit 702 is further configured to register the second model for service using the third information;

[0406] The second receiving unit 703 is further configured to receive fourth information sent by AnLF, the fourth information being used to request the discovery of the second model;

[0407] The first sending unit 704 is further configured to send fifth information to the AnLF, the fifth information indicating that the second model is obtained from the second functional body.

[0408] In practical applications, the first receiving unit 701, the second receiving unit 703, and the first sending unit 704 can be implemented by the communication interface in the communication device; the second processing unit 702 can be implemented by the processor in the communication device.

[0409] To implement the method on the AnLF side of this application embodiment, this application embodiment also provides a communication device, disposed on the AnLF, such as... Figure 8 As shown, the device includes:

[0410] The second sending unit 801 is used to send fourth information to the NRF, the fourth information being used to request the discovery of a second model, the second model being obtained by the first MTLF by fine-tuning the first model, and the first model containing at least one model pre-trained by the second MTLF.

[0411] The third receiving unit 802 is used to receive the fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body;

[0412] The second acquisition unit 803 is used to acquire the second model from the second functional body in order to perform inference analysis.

[0413] In one embodiment, the second acquisition unit 803 is specifically used for:

[0414] Send an eighth message to the second functional body, the eighth message being used to request or subscribe to the second model;

[0415] Receive the ninth message sent by the second functional body, the ninth message including one of the following:

[0416] The second model;

[0417] The URL corresponding to the second model.

[0418] In practical applications, the second sending unit 801 and the third receiving unit 802 can be implemented by the communication interface in the communication device; the second acquiring unit 803 can be implemented by the processor in the communication device in conjunction with the communication interface.

[0419] It should be noted that the communication device provided in the above embodiments is only illustrated by the division of the above program modules during communication. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the communication device and communication method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0420] Based on the hardware implementation of the above program modules, and in order to implement the method on the first MTLF side of the embodiments of this application, the embodiments of this application also provide a first MTLF, such as... Figure 9 As shown, the first MTLF 900 includes:

[0421] The first communication interface 901 is capable of exchanging information with the terminal and / or other functional entities (such as the second MTLF, NRF, ADRF, AnLF, the first functional entity, etc.);

[0422] The first processor 902 is connected to the first communication interface 901 to enable information interaction with the terminal and / or other functional entities, and to execute the methods provided by one or more technical solutions on the first MTLF side when running a computer program;

[0423] The computer program is stored in the first memory 903.

[0424] Specifically, the first processor 902 is used to obtain a first model from the first functional body through the first communication interface 901, the first model including at least one model pre-trained by the second MTLF; and to fine-tune the first model to obtain a second model, the second model being used for AnLF to perform inference analysis.

[0425] In one embodiment, the first processor 902 is further configured to:

[0426] The first information is sent to the NRF through the first communication interface 901, and the first information is used to request the discovery of the first model.

[0427] The first communication interface 901 receives the second information sent by the NRF, the second information indicating that the first model is obtained from the first functional body;

[0428] The first model is obtained from the first functional body based on the second information.

[0429] In one embodiment, the first communication interface 901 is used for:

[0430] Send the second model to the AnLF;

[0431] or,

[0432] Send the second model to the ADRF so that the AnLF can obtain the second model from the ADRF;

[0433] or,

[0434] A third message is sent to the NRF so that when the NRF receives the fourth message sent by the AnLF, it can send a fifth message to the AnLF; the third message is used to register the second model for service; the fourth message is used to request the discovery of the second model; and the fifth message indicates that the second model is obtained from the second functional body.

[0435] In one embodiment, the first processor 902 is further configured to:

[0436] The sixth information is sent to the first functional body through the first communication interface 901, and the sixth information is used to request or subscribe to the first model.

[0437] The seventh information sent by the first functional body is received through the first communication interface 901, and the seventh information includes one of the following:

[0438] The first model;

[0439] The URL corresponding to the first model.

[0440] In one embodiment, the first communication interface 901 is further configured to:

[0441] Receive the eighth message sent by AnLF, the eighth message being used to request or subscribe to the second model;

[0442] Send a ninth message to the AnLF, the ninth message including one of the following:

[0443] The second model;

[0444] The URL corresponding to the second model.

[0445] It should be noted that the specific processing procedures of the first communication interface 901 and the first processor 902 can be understood by referring to the above method, and will not be repeated here.

[0446] Of course, in practical applications, the various components in the first MTLF 900 are coupled together via a bus system 904. It can be understood that the bus system 904 is used to implement communication between these components. In addition to the data bus, the bus system 904 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 9The general designated all buses as Bus System 904.

[0447] The first memory 903 in this embodiment is used to store various types of data to support the operation of the first MTLF 900. Examples of such data include any computer program used to operate on the first MTLF 900.

[0448] The methods disclosed in the embodiments of this application can be applied to the first processor 902, or implemented by the first processor 902. The first processor 902 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware or by instructions in the form of software in the first processor 902. The first processor 902 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The first processor 902 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in the first memory 903. The first processor 902 reads the information in the first memory 903 and completes the steps of the aforementioned method in combination with its hardware.

[0449] In an exemplary embodiment, the first MTLF 900 may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0450] Based on the hardware implementation of the above program modules, and in order to implement the NRF-side method of the embodiments of this application, the embodiments of this application also provide an NRF, such as... Figure 10 As shown, the NRF 1000 includes:

[0451] The second communication interface 1001 is capable of exchanging information with the terminal and / or other functional entities (such as the first MTLF, the second MTLF, ADRF, AnLF, the first functional entity, the second functional entity, etc.);

[0452] The second processor 1002 is connected to the second communication interface 1001 to enable information interaction with the terminal and / or other functional entities, and to execute the methods provided by one or more of the above-mentioned NRF side technical solutions when running computer programs;

[0453] The computer program is stored in the second memory 1003.

[0454] Specifically, the second communication interface 1001 is used to receive tenth information sent by the first functional body, the tenth information being used to perform service registration for at least one model pre-trained by the second MTLF.

[0455] The second processor 1002 is used to perform service registration on at least one model pre-trained by the second MTLF using the tenth information;

[0456] The second communication interface 1001 is further configured to receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF; and to send second information to the first MTLF, the second information indicating that the first model is obtained from the first functional body, so that the first MTLF can fine-tune the first model.

[0457] In one embodiment, the second communication interface 1001 is further configured to receive third information sent by the second functional body, the third information being used to register a service for the second model, the second model being obtained by the first MTLF through model fine-tuning of the first model;

[0458] The second processor 1002 is further configured to register the second model for service using the third information;

[0459] The second communication interface 1001 is also used to receive a fourth message sent by AnLF, the fourth message being used to request the discovery of the second model; and to send a fifth message to AnLF, the fifth message being used to indicate that the second model is obtained from the second functional body.

[0460] It should be noted that the specific processing procedures of the second communication interface 1001 and the second processor 1002 can be understood by referring to the above method, and will not be repeated here.

[0461] Of course, in practical applications, the various components in the NRF 1000 are coupled together via bus system 1004. It can be understood that bus system 1004 is used to implement communication between these components. In addition to the data bus, bus system 1004 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 10 The general labeled all buses as Bus System 1004.

[0462] The second memory 1003 in this embodiment is used to store various types of data to support the operation of the NRF 1000. Examples of such data include any computer program used to operate on the NRF 1000.

[0463] The methods disclosed in the embodiments of this application can be applied to the second processor 1002, or implemented by the second processor 1002. The second processor 1002 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware or by instructions in the form of software in the second processor 1002. The second processor 1002 may be a general-purpose processor, a DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The second processor 1002 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in the second memory 1003. The second processor 1002 reads the information in the second memory 1003 and completes the steps of the aforementioned method in combination with its hardware.

[0464] In an exemplary embodiment, the NRF 1000 may be implemented by one or more ASICs, DSPs, PLDs, CPLDs, FPGAs, general-purpose processors, controllers, MCUs, microprocessors, or other electronic components to perform the aforementioned method.

[0465] Based on the hardware implementation of the above program modules, and in order to implement the AnLF side method of the embodiments of this application, the embodiments of this application also provide an AnLF, such as... Figure 11 As shown, the AnLF 1100 includes:

[0466] The third communication interface 1101 is capable of exchanging information with the terminal and / or other functional entities (such as the first MTLF, the second MTLF, ADRF, NRF, the second functional entity, etc.);

[0467] The third processor 1102 is connected to the third communication interface 1101 to enable information interaction with the terminal and / or other functional entities, and to execute the methods provided by one or more of the AnLF side technical solutions when running computer programs;

[0468] The computer program is stored in the third memory 1103.

[0469] Specifically, the third communication interface 1101 is used to send fourth information to the NRF, the fourth information being used to request the discovery of a second model, the second model being obtained by the first MTLF by fine-tuning the first model, the first model including at least one model pre-trained by the second MTLF; and to receive fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body;

[0470] The third processor 1102 is used to obtain the second model from the second functional body through the third communication interface 1101 in order to perform inference analysis.

[0471] In one embodiment, the third processor 1102 is further configured to:

[0472] The eighth information is sent to the second functional body through the third communication interface 1101, and the eighth information is used to request or subscribe to the second model.

[0473] The ninth message sent by the second functional body is received through the third communication interface 1101, and the ninth message includes one of the following:

[0474] The second model;

[0475] The URL corresponding to the second model.

[0476] It should be noted that the specific processing procedures of the third communication interface 1101 and the third processor 1102 can be understood by referring to the above method, and will not be repeated here.

[0477] Of course, in practical applications, the various components in the AnLF 1100 are coupled together via bus system 1104. It can be understood that bus system 1104 is used to implement communication between these components. In addition to the data bus, bus system 1104 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 11 The general designated all buses as Bus System 1104.

[0478] The third memory 1103 in this embodiment is used to store various types of data to support the operation of AnLF 1100. Examples of such data include any computer program used to operate on AnLF 1100.

[0479] The methods disclosed in the embodiments of this application can be applied to, or implemented by, the third processor 1102. The third processor 1102 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by the integrated logic circuitry of the hardware or by instructions in the software form of the third processor 1102. The third processor 1102 may be a general-purpose processor, a DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The third processor 1102 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, specifically a third memory 1103. The third processor 1102 reads information from the third memory 1103 and, in conjunction with its hardware, completes the steps of the aforementioned method.

[0480] In an exemplary embodiment, AnLF 1100 may be implemented by one or more ASICs, DSPs, PLDs, CPLDs, FPGAs, general-purpose processors, controllers, MCUs, microprocessors, or other electronic components to perform the aforementioned method.

[0481] It is understood that the memories (first memory 903, second memory 1003, and third memory 1103) in the embodiments of this application can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memories.

[0482] To implement the method provided in the embodiments of this application, the embodiments of this application also provide a communication system, such as... Figure 12 As shown, the system includes: a first MTLF 1201, an NRF 1202, and an AnLF 1203.

[0483] It should be noted that the specific processing procedures of the first MTLF 1201, NRF 1202 and AnLF 1203 have been detailed above and will not be repeated here.

[0484] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium. For example, it may include a first memory 903 storing a computer program, which can be executed by a first processor 902 of a first MTLF 900 to complete the steps described in the aforementioned first MTLF-side method. Another example is a second memory 1003 storing a computer program, which can be executed by a second processor 1002 of an NRF 1000 to complete the steps described in the aforementioned NRF-side method. Yet another example is a third memory 1103 storing a computer program, which can be executed by a third processor 1102 of an AnLF 1100 to complete the steps described in the aforementioned AnLF-side method. The computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0485] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0486] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0487] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. A communication method, characterized in that, The MTLF, applied to the first model training logic function, includes: A first model is obtained from a first functional body, wherein the first model contains at least one model pre-trained by a second MTLF; The first model is fine-tuned to obtain a second model, which is used by the AnLF analysis logic function to perform inference analysis; wherein... The step of obtaining the first model from the first functional body includes: Send a first message to the Network Repository Function (NRF), the first message being used to request the discovery of a first model; Receive second information sent by the NRF, the second information indicating that the first model is obtained from the first functional body; Obtain the first model from the first functional body; The method further includes: A third message is sent to the NRF so that when the NRF receives the fourth message sent by the AnLF, it can send a fifth message to the AnLF; the third message is used to register the second model for service; the fourth message is used to request the discovery of the second model; and the fifth message indicates that the second model is obtained from the second functional body.

2. The method according to claim 1, characterized in that, The first functional body includes the second MTLF or analytical data repository function ADRF.

3. The method according to claim 1, characterized in that, The first information includes at least the identifier of the first model.

4. The method according to claim 3, characterized in that, The first information also includes at least one of the following: The type of the first model; The first model supports an artificial intelligence (AI) platform; The AI ​​framework supported by the first model; The first model supports the following device manufacturers; Does the first model support model sharing? Does the first model support cross-platform model format conversion? Does the first model support cross-language model format conversion? The conversion formats supported by the first model.

5. The method according to claim 1, characterized in that, The second information includes at least one of the following: The address of the first functional unit; The identifier of the first functional unit.

6. The method according to claim 1, characterized in that, The second functional unit includes a first MTLF or ADRF.

7. The method according to claim 1, characterized in that, The third information includes at least the identifier of the second model.

8. The method according to claim 7, characterized in that, The third information also includes at least one of the following: The address of the second functional unit; The identifier of the second functional entity; The type of the second model; The second model supports AI platforms; The second model supports the following AI frameworks; The second model supports the following equipment manufacturers; Does the second model support model sharing? Does the second model support cross-platform model format conversion? Does the second model support cross-language model format conversion? The conversion formats supported by the second model.

9. The method according to claim 1, characterized in that, The fourth information includes at least the identifier of the second model.

10. The method according to claim 9, characterized in that, The fourth information also includes at least one of the following: The type of the second model; The second model supports AI platforms; The second model supports the following AI frameworks; The second model supports the following equipment manufacturers; Does the second model support model sharing? Does the second model support cross-platform model format conversion? Does the second model support cross-language model format conversion? The conversion formats supported by the second model.

11. The method according to claim 1, characterized in that, The fifth piece of information includes at least one of the following: The address of the second functional unit; The identifier of the second functional body.

12. The method according to any one of claims 1 to 5, characterized in that, The step of obtaining the first model from the first functional body includes: Send a sixth message to the first functional body, the sixth message being used to request or subscribe to the first model; Receive the seventh information sent by the first functional unit, wherein the seventh information includes one of the following: The first model; The Uniform Resource Locator (URL) corresponding to the first model.

13. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Receive the eighth message sent by AnLF, the eighth message being used to request or subscribe to the second model; Send a ninth message to the AnLF, the ninth message including one of the following: The second model; The URL corresponding to the second model.

14. A communication method, characterized in that, Applied to NRF, including: Receive the tenth information sent by the first functional body, the tenth information being used for service registration of at least one model pre-trained by the second MTLF; The tenth information is used to register services for at least one model pre-trained by the second MTLF; Receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF; A second message is sent to the first MTLF, indicating that the first model should be obtained from the first functional body, so that the first MTLF can perform model fine-tuning on the first model; wherein... The method further includes: Receive third information sent by the second functional body, the third information being used to register the second model as a service, the second model being obtained by the first MTLF by fine-tuning the first model; The second model is registered for service using the third information; Receive the fourth message sent by AnLF, the fourth message being used to request the discovery of the second model; Send a fifth message to the AnLF, the fifth message indicating that the second model is obtained from the second functional body.

15. The method according to claim 14, characterized in that, The first functional unit includes the second MTLF or ADRF.

16. The method according to claim 14, characterized in that, The tenth piece of information includes at least the identifier of the at least one model.

17. The method according to claim 16, characterized in that, The tenth information also includes at least one of the following: The address of the first functional unit; The identifier of the first functional unit; The type of at least one model; The AI ​​platform supported by at least one model; The AI ​​framework supported by at least one model; The at least one model supports the equipment manufacturers; Does the at least one model support model sharing? Does the at least one model support cross-platform model format conversion? Does the at least one model support cross-language model format conversion? The conversion formats supported by at least one of the models.

18. The method according to claim 14, characterized in that, The first information includes at least the identifier of the first model.

19. The method according to claim 18, characterized in that, The first information also includes at least one of the following: The type of the first model; The AI ​​platforms supported by the first model; The AI ​​framework supported by the first model; The first model supports the following device manufacturers; Does the first model support model sharing? Does the first model support cross-platform model format conversion? Does the first model support cross-language model format conversion? The conversion formats supported by the first model.

20. The method according to claim 14, characterized in that, The second information includes at least one of the following: The address of the first functional unit; The identifier of the first functional unit.

21. The method according to claim 14, characterized in that, The second functional unit includes the first MTLF or ADRF.

22. The method according to claim 14, characterized in that, The third information includes at least the identifier of the second model.

23. The method according to claim 22, characterized in that, The third information also includes at least one of the following: The address of the second functional unit; The identifier of the second functional entity; The type of the second model; The second model supports AI platforms; The second model supports the following AI frameworks; The second model supports the following equipment manufacturers; Does the second model support model sharing? Does the second model support cross-platform model format conversion? Does the second model support cross-language model format conversion? The conversion formats supported by the second model.

24. The method according to claim 14, characterized in that, The fourth information includes at least the identifier of the second model.

25. The method according to claim 24, characterized in that, The fourth information also includes at least one of the following: The type of the second model; The second model supports AI platforms; The second model supports the following AI frameworks; The second model supports the following equipment manufacturers; Does the second model support model sharing? Does the second model support cross-platform model format conversion? Does the second model support cross-language model format conversion? The conversion formats supported by the second model.

26. The method according to claim 14, characterized in that, The fifth piece of information includes at least one of the following: The address of the second functional unit; The identifier of the second functional body.

27. A communication method, characterized in that, Applied to AnLF, including: Send a fourth message to the NRF, the fourth message being used to request the discovery of a second model, the second model being obtained by fine-tuning the first model by the first MTLF, the first model including at least one model pre-trained by the second MTLF; Receive the fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body; The second model is obtained from the second functional body to perform inference analysis; wherein, The second functional unit includes the first MTLF or ADRF; The step of obtaining the second model from the second functional body includes: Send an eighth message to the second functional body, the eighth message being used to request or subscribe to the second model; Receive the ninth message sent by the second functional body, the ninth message including one of the following: The second model; The URL corresponding to the second model.

28. The method according to claim 27, characterized in that, The fourth information includes at least the identifier of the second model.

29. The method according to claim 28, characterized in that, The fourth information also includes at least one of the following: The type of the second model; The second model supports AI platforms; The second model supports the following AI frameworks; The second model supports the following equipment manufacturers; Does the second model support model sharing? Does the second model support cross-platform model format conversion? Does the second model support cross-language model format conversion? The conversion formats supported by the second model.

30. The method according to claim 27, characterized in that, The fifth piece of information includes at least one of the following: The address of the second functional unit; The identifier of the second functional body.

31. A communication device, characterized in that, include: The first acquisition unit is configured to acquire a first model from a first functional body, wherein the first model includes at least one model pre-trained by a second MTLF. The first processing unit is used to fine-tune the first model to obtain a second model, which is then used by AnLF to perform inference analysis; wherein... The third sending unit is used to send first information to the NRF, the first information being used to request the discovery of the first model; The fourth receiving unit is configured to receive second information transmitted by the NRF, the second information indicating that the first model is obtained from the first functional body; The first acquisition unit is specifically used to acquire the first model from the first functional body based on the second information; The third sending unit is further configured to send third information to the NRF so that when the NRF receives the fourth information sent by the AnLF, it can send fifth information to the AnLF; the third information is used to register the second model for service; the fourth information is used to request the discovery of the second model; and the fifth information indicates that the second model is obtained from the second functional body.

32. A communication device, characterized in that, include: The first receiving unit is used to receive the tenth information sent by the first functional body, the tenth information being used to perform service registration for at least one model pre-trained by the second MTLF. The second processing unit is used to perform service registration on at least one model pre-trained by the second MTLF using the tenth information; The second receiving unit is used to receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF; A first transmitting unit is configured to send second information to the first MTLF, the second information indicating that the first model is obtained from the first functional body, so that the first MTLF can perform model fine-tuning on the first model; wherein... The second receiving unit is further configured to receive third information sent by the second functional body, the third information being used to register the second model for service, the second model being obtained by the first MTLF through model fine-tuning of the first model; The second processing unit is further configured to register the second model for service using the third information; The second receiving unit is further configured to receive fourth information sent by AnLF, the fourth information being used to request the discovery of the second model; The first sending unit is further configured to send fifth information to the AnLF, the fifth information indicating that the second model is obtained from the second functional body.

33. A communication device, characterized in that, include: The second sending unit is used to send fourth information to the NRF, the fourth information being used to request the discovery of a second model, the second model being obtained by the first MTLF by fine-tuning the first model, and the first model containing at least one model pre-trained by the second MTLF. The third receiving unit is used to receive the fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body; The second acquisition unit is used to acquire the second model from the second functional body to perform inference analysis; wherein, The second functional unit includes the first MTLF or ADRF; The second acquisition unit is specifically configured to send eighth information to the second functional entity, the eighth information being used to request or subscribe to the second model; and to receive ninth information sent by the second functional entity, the ninth information including one of the following: The second model; The URL corresponding to the second model.

34. A first MTLF, characterized in that, include: A first communication interface and a first processor; wherein... The first processor is configured to obtain a first model from a first functional unit through the first communication interface, the first model including at least one model pre-trained by a second MTLF; and to fine-tune the first model to obtain a second model, the second model being used for inference analysis by AnLF. The first processor is further configured to send first information to the NRF via the first communication interface, the first information being used to request the discovery of a first model; and to receive second information sent by the NRF via the first communication interface, the second information being used to indicate obtaining the first model from the first functional body; The first communication interface is used to send third information to the NRF so that when the NRF receives the fourth information sent by the AnLF, it can send fifth information to the AnLF; the third information is used to register the second model for service; the fourth information is used to request the discovery of the second model; and the fifth information indicates that the second model is obtained from the second functional body.

35. An NRF, characterized in that, include: The second communication interface is used to receive the tenth information sent by the first functional body, the tenth information being used to register services for at least one model pre-trained by the second MTLF. The second processor is used to perform service registration on at least one model pre-trained by the second MTLF using the tenth information; The second communication interface is further configured to receive first information sent by the first MTLF, the first information being used to request the discovery of a first model, the first model including at least one model pre-trained by the second MTLF; and to send second information to the first MTLF, the second information indicating that the first model is obtained from the first functional body, so that the first MTLF can fine-tune the first model. The second communication interface is also used to receive third information sent by the second functional body. The third information is used to register the second model as a service. The second model is obtained by the first MTLF by fine-tuning the first model. The second processor is further configured to register the second model for services using the third information; The second communication interface is also used to receive a fourth message sent by AnLF, the fourth message being used to request the discovery of the second model; and to send a fifth message to AnLF, the fifth message indicating that the second model is obtained from the second functional body.

36. An AnLF, characterized in that, include: The third communication interface is used to send fourth information to the NRF, the fourth information being used to request the discovery of a second model, the second model being obtained by the first MTLF by fine-tuning the first model, the first model including at least one model pre-trained by the second MTLF; and to receive fifth information sent by the NRF, the fifth information indicating that the second model is obtained from the second functional body; A third processor is configured to obtain the second model from the second functional unit via the third communication interface, in order to perform inference analysis; wherein, The second functional unit includes the first MTLF or ADRF; The third processor is further configured to send eighth information to the second functional body through the third communication interface, the eighth information being used to request or subscribe to the second model; The third processor is further configured to receive ninth information sent by the second functional body via the third communication interface, the ninth information including one of the following: The second model; The URL corresponding to the second model.

37. A first MTLF, characterized in that, include: A first processor and a first memory for storing computer programs capable of running on the processor. Wherein, when the first processor is used to run the computer program, it performs the steps of the method according to any one of claims 1 to 13.

38. An NRF, characterized in that, include: A second processor and a second memory for storing computer programs that can run on the processor. Wherein, when the second processor is used to run the computer program, it performs the steps of the method according to any one of claims 14 to 26.

39. An AnLF, characterized in that, include: A third processor and a third memory for storing computer programs that can run on the processor. When the third processor runs the computer program, it performs the steps of the method according to any one of claims 27 to 30.

40. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 13, or the steps of the method according to any one of claims 14 to 26, or the steps of the method according to any one of claims 27 to 30.

Images