Communication method and device

Abstract

The embodiment of the invention provides a communication method and device. The method comprises the steps that a first management network element receives first request information sent by a second management network element; and the first management network element performs fine tuning on the machine learning pre-training model according to the first request information to generate a fine-tuned machine learning model. Through the embodiment of the invention, the problems of low efficiency and resource waste caused by retraining the model from the beginning when the machine learning model is generated in the related technology are solved, and the effect of quickly generating the machine learning model by using less resources is achieved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of communications, in particular, to a communication method and apparatus. BACKGROUND

[0002] SA5 is studying and finalizing other aspects of the Artificial Intelligence (AI) / Machine Learning (ML) management specification for Rel-18, with a particular focus on enabling and managing AI / ML functions in various domains of the 5G system, including management and orchestration (e.g., MDA defined in TS 28.104), 5GC (e.g., NWDAF defined in TS 23.288), and NG-RAN (e.g., RAN intelligence defined in TS 38.300 and TS 38.401). The AI / ML management functions defined in SA5 Rel-18 include management and operation of ML training, ML testing, AI / ML simulation, ML entity deployment, and AI / ML inference.

[0003] In the current AIML lifecycle management (LCM) process, if the performance of a model does not meet expectations, retraining (i.e., training the model from scratch) needs to be triggered by the consumer (Consumer) or the producer (Producer) actively. Training a new model from scratch means that data needs to be collected again, which is likely to cause waste of resources. SUMMARY

[0004] Embodiments of the present application provide a communication method and apparatus to at least solve the problem of low efficiency and resource waste in retraining a model from scratch when generating a machine learning model in the related art.

[0005] According to an embodiment of the present application, a communication method is provided, comprising: a first management network element receiving first request information sent by a second management network element; the first management network element fine-tuning a machine learning pre-training model according to the first request information to generate a fine-tuned machine learning model.

[0006] According to another embodiment of the present application, a communication method is provided, comprising: a third management network element receiving second request information sent by a fourth management network element for requesting the third management network element to perform pre-training on a machine learning model, wherein the second request information includes pre-training model request indication information and / or pre-training data information; the third management network element performing model pre-training according to the second request information to generate a machine learning pre-training model.

[0007] According to still another embodiment of the present application, a computer program product is also provided, comprising computer programs, instructions, wherein the computer programs, instructions are arranged, when executed by a processor, to perform the steps of any of the above methods.

[0008] According to still another embodiment of the present application, a computer readable storage medium is also provided, having stored therein a computer program, wherein the computer program is arranged, when executed by a processor, to perform the steps of any of the above method embodiments.

[0009] According to still another embodiment of the present application, an electronic device is also provided, comprising a memory and a processor, the memory having stored therein a computer program, the processor being arranged to execute the computer program to perform the steps of any of the above method embodiments.

[0010] According to the present application, after receiving the first request information sent by the second management network element, the first management network element fine-tunes the existing machine learning pre-training model according to the first request information to generate a fine-tuned machine learning model, instead of generating a machine learning model from scratch. Therefore, the problem of low efficiency and resource waste when retraining a model from scratch to generate a machine learning model in the related art can be solved, and the effect of quickly generating a machine learning model using fewer resources is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is a hardware structure block diagram of a computer terminal according to the communication method of the embodiment of the present application;

[0012] Figure 2 is a schematic diagram of a network management architecture according to the communication method of the embodiment of the present application;

[0013] Figure 3 is a flowchart of the communication method according to the embodiment of the present application;

[0014] Figure 4 is a flowchart of the communication method according to the embodiment of the present application;

[0015] Figure 5 is a schematic diagram of a machine learning pre-training model according to the embodiment of the present application;

[0016] Figure 6 is a flowchart of the machine learning pre-training model according to the embodiment of the present application;

[0017] Figure 7 is a flowchart of the machine learning pre-training model according to the embodiment of the present application;

[0018] Figure 8is a machine learning pre-trained model diagram according to yet another embodiment of the present application;

[0019] Figure 9 is a machine learning pre-trained model generation flow diagram according to yet another embodiment of the present application;

[0020] Figure 10 is a machine learning model generation flow diagram according to an embodiment of the present application;

[0021] Figure 11 is a machine learning model generation flow diagram according to yet another embodiment of the present application;

[0022] Figure 12 is a machine learning model life cycle flow diagram according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Embodiments of the present application will be described below in detail with reference to the accompanying drawings and in conjunction with embodiments.

[0024] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and in the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence.

[0025] The following is an explanation of related terms of the embodiments of the present application:

[0026] ML model: A manageable artifact of an ML model algorithm. An ML model can contain metadata related to the model algorithm. The metadata can include, for example, applicable runtime contexts of the ML model algorithm.

[0027] ML model algorithm: A mathematical algorithm that can be "trained" by data and human expert input as examples to replicate decisions made by experts when given the same information.

[0028] ML model training: A process performed by an ML training function to take training data, run that data through an ML model, derive a related loss, and adjust the parameterization of the ML model based on the calculated loss.

[0029] ML initial training: An ML model training that generates an initial version of an ML entity.

[0030] ML retraining: A process of retraining a previously trained ML model. The newly trained ML entity supports the same type of inference as the previous version of the ML entity, i.e., the data types of the inference input and the data types of the inference output remain the same between the two versions of the ML entity, but the parameter values of the retrained model can be different.

[0031] ML federated training: ML training against a set of trained and inference-oriented ML models.

[0032] ML training: refers to the end-to-end process that enables the ML training function to perform initial training or retraining of an ML model. ML training can include interactions with other parties to collect and format data required for training of an ML model.

[0033] ML training function: a logical function with the capability of training an ML model, also referred to as MLT function.

[0034] AI / ML inference: refers to the process of running a set of input data through a trained ML entity to generate a set of output data (e.g., predictions).

[0035] AI / ML inference function: a logical function that uses an ML model for inference.

[0036] Model complexity parameters include model parameter quantity: parameters. Model parameter quantity refers to the number of learnable parameters in a machine learning model, which are usually weights and biases, used to represent the features and relationships of the model. The number of model parameters directly affects the complexity and capacity of the model, and is usually closely related to the expressive power and performance of the model. A larger number of parameters may mean a more complex model, but also increases the risk of overfitting. Model computation: the computational resources required by the model when performing inference or training. Computation is usually measured by indicators such as FLOPs, MACs, and MAdds. FLOPs (floating-point operations per second), MACs (multiply-accumulate operations), and MAdds (multiply-accumulate operations) are metrics used to quantify the computational complexity of algorithms, particularly in the fields of machine learning and signal processing. Here are the meanings of each term:

[0037] FLOPs (floating-point operations per second): represents the number of floating-point operations (such as addition, subtraction, multiplication, division) performed by a computing device in one second. It measures the raw computing power of a system. FLOPs can also be used to measure the computational complexity of an algorithm, indicating the number of floating-point operations required to perform a specific task.

[0038] MACs (multiply-accumulate operations): specifically refers to the number of multiply-accumulate operations performed by an algorithm or hardware component. In many machine learning models (such as convolutional neural networks CNN), most of the computational work comes from MAC operations, which involve multiplying two numbers and then adding the result to an accumulator.

[0039] MAdds (multiply-accumulate operations): similar to MACs, but separately counts the number of multiplication and addition operations. Therefore, for each MAC operation, two operations (one multiplication and one addition) are counted. Compared with MACs, MAdds provides a more detailed breakdown of computational complexity.

[0040] The above indicators are very important for understanding and comparing the efficiency of different algorithms, architectures, and hardware platforms. They help optimize algorithms in resource-constrained environments, such as mobile devices or edge computing devices, to improve speed and resource consumption efficiency.

[0041] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal, or a similar computing device. Taking the case of running on a computer terminal, Figure 1 is a hardware structure block diagram of a computer terminal of a communication method of the embodiments of the present application. As shown in Figure 1 , the computer terminal can include one or more (only one is shown in Figure 1 ) processors 102 (the processor 102 can include but is not limited to a processing device such as a microcontroller unit (MCU) or a field programmable gate array (FPGA)) and a memory 104 for storing data, wherein the above computer terminal can further include a transmission device 106 for communication function and an input and output device 108. Those skilled in the art can understand that Figure 1 the structure shown is only schematic, which does not limit the structure of the above computer terminal. For example, the computer terminal can further include more or less components than those shown in Figure 1 , or have a different configuration from that shown in Figure 1 .

[0042] The memory 104 can be used to store computer programs, for example, software programs of application software and modules, such as the computer program corresponding to the communication method in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, implements the above method. The memory 104 can include a high-speed random access memory, and can also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory 104 can further include a memory remotely arranged with respect to the processor 102, which can be connected to the computer terminal through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0043] The transmission device 106 is configured to receive or send data via a network. The network can include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network interface controller (NIC) that can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module configured to communicate with the Internet through a wireless manner.

[0044] Embodiments of the present application can run on a network architecture as shown in Figure 2 Figure 2 The network architecture includes a business support system (BSS), a cross domain management function (CD-MnF), a domain management function (Domain-MnF), and a network element (NE). The cross domain management function is configured to manage one or more domain management functions. The domain management function is configured to manage one or more network elements. The following will introduce each unit respectively.

[0045] The business support system is configured to provide functions and management services for communication services, such as charging, settlement, accounting, customer service, business, network monitoring, communication service lifecycle management, service intent translation, etc. The business support system can be an operation system of an operator, or a vertical OT system.

[0046] ​The cross-domain management function unit is also called a network management function unit (NMF), which can be a network management system (NMS), a network management service producer (MnS Producer), a network management service consumer (MnS Consumer), a network function management service consumer (NFMS_C), and the like. The cross-domain management function unit provides one or more of the following management functions or management services: network lifecycle management, network deployment, network fault management, network performance management, network configuration management, network assurance, network optimization function, and translation of the network intent (Intent-CSP) from a communication service provider. The network referred to in the above management functions or management services can include one or more network elements or sub-networks, or a network slice. That is, the network management function unit can be a network slice management function unit (NSMF), or a cross-domain management data analysis function unit (MDAF), or a cross-domain self-organization network function (SON Function), or a cross-domain intent-driven management function unit (Intent-Driven Management Service, Intent Driven MnS).

[0047] The domain management function unit is also called a network subnet management function (NSMF) or an element management function. It can be a wireless automation engine (MAE), an element management system (EMS), a network function management service provider (NFMS_P), a network slice subnet management function (NSSMF), a domain management data analysis function (Domain MDAF), a domain self-organization network function (SON Function), a domain intent management function, a MnSProducer, a MnS Consumer, and other element management entities. The domain management function unit can be classified in the following ways, including: by network type, which can be classified into a radio access network (RAN) domain management function (RAN domain MnF), a core network domain management function (CN domain MnF), a transport network domain management function (TN domain MnF), and the like. It should be noted that the domain management function unit can also be a domain network management system that manages one or more of an access network, a core network, or a transport network; by administrative region, which can be classified into a domain management function unit for a certain region, such as an A city domain management function unit, a B city domain management function unit, and the like. The domain management function unit provides one or more of the following functions or management services: lifecycle management of a subnet or an element, deployment of a subnet or an element, fault management of a subnet or an element, performance management of a subnet or an element, assurance of a subnet or an element, optimization function of a subnet or an element, and translation of an intent from a network operator (Intent-NOP), and the like. The subnet here includes one or more elements.The sub-network can also include a sub-network, that is, one or more sub-networks form a larger sub-network. The sub-network here can also be a network slice sub-network.

[0048] A network element is an entity for providing network services. The network element can include a core network element, a radio access network element, or a transport network element, etc. Specifically, the core network element can include, but is not limited to, an access and mobility management function (AMF) entity, a session management function (SMF) entity, a policy control function (PCF) entity, a network data analysis function (NWDAF) entity, a network repository function (NRF), a gateway, etc. The radio access network element can include, but is not limited to, various types of base stations (such as a generation node B (gNB), an evolved node B (eNB), a central unit control panel (CUCP), a central unit (CU), a distributed unit (DU), a central unit user panel (CUUP), etc.). In this application, a network function (NF) is also referred to as a network element (NE). The network element can provide one or more of the following management functions or management services: lifecycle management of the network element, deployment of the network element, fault management of the network element, performance management of the network element, assurance of the network element, optimization function of the network element, and translation of the intent of the network element, etc.

[0049] The cross-domain management function unit can be used for model training of the AI / ML model and model inference of the AI / ML model; the domain management function unit can be used for model training of the AI / ML model and model inference of the AI / ML model; the network element is an entity for providing network services, including a core network element and an access network element, and can provide at least one of model training of the AI / ML model and model inference of the AI / ML model.

[0050] In this embodiment, a communication method running on the above computer terminal or network architecture is provided, which is applied to a first management network element, Figure 3 is a flowchart of the communication method according to the embodiments of the present application, as Figure 3 shown, the flow includes the following steps:

[0051] Step S302, the first management network element receives the first request information sent by the second management network element.

[0052] In one embodiment, the first management network element is an ML Fine-tuning Producer (machine learning fine-tuning producer), which can be an existing ML Training Function (machine learning training function network element), or a ML Fine-tuning Function (machine learning fine-tuning function network element) defined specifically to support the Fine tuning function, to generate a new ML Model (machine learning model) based on the Pre-training ML Model (machine learning pre-training model) fine tuning. The second management network element is a machine learning fine-tuning consumer or a machine learning inference consumer.

[0053] In step S302 of the embodiments of the present application, the first request information includes at least one of the following: fine tuning indication information, used to indicate that a machine learning model needs to be generated based on machine learning pre-training model fine tuning; pre-training model identification information, used to indicate the identification of the machine learning pre-training model; model requirement information, used to indicate the requirement of the machine learning model.

[0054] In one exemplary embodiment, the model requirement information includes: model capability information, used to indicate the inference capability supported by the machine learning model, such as one or more MAD Type, Analytic ID, etc.; model performance information, used to indicate the performance of the machine learning model, such as accuracy, F1 score, etc.; model complexity information, used to indicate the complexity of the machine learning model, such as the number of parameters of the model, the size of the model, the number of layers of the internal neural network of the model; model energy consumption information, used to indicate the energy consumption of the machine learning model, which can be the energy consumption of the current training or inference task, or the energy consumption requirement of the inference execution of the fine-tuned model.

[0055] In one embodiment, the model complexity information is affected by the number of model parameters, which refers to the number of learnable parameters in the machine learning model, which are usually weights and biases, used to represent the characteristics and relationships of the model. The model calculation amount also affects the model complexity, which is the computing resource required by the model during inference or training. The calculation amount is usually measured by indicators such as FLOPs, MACs and MAdds. FLOPs (floating point operations per second), MACs (multiply-accumulate operations) and MAdds (multiply-accumulate operations) are metrics used to quantify the computational complexity of algorithms, especially in the fields of machine learning and signal processing.

[0056] In different scenarios, there can be different granularity requirements for model energy consumption information, for example, different physical environments (chips) have different energy consumptions, indicating the maximum energy consumption requirement of this training / inference task and indicating the energy consumption requirement of the trained model in runtime.

[0057] In an example embodiment, the first request information further includes fine-tuning data information for indicating fine-tuning based on the machine learning pre-trained model, wherein the fine-tuning data information includes at least one of the following: data address information, feature information, and sample identification information.

[0058] In an embodiment, the feature information includes PM and KPI, and the sample identification information includes UE ID (terminal identification) and S-NSSAI (service-network slice selection identification).

[0059] In an example embodiment, the first management network element receives the first request information sent by the second management network element, including: the first management network element receives an inference request message sent by the second management network element; the first management network element performs inference according to the inference request message and sends the inference result to the second management network element; and the first management network element receives the first request information sent by the second management network element under a preset condition, wherein the preset condition includes that the inference performance of the machine learning model evaluated by the second management network element based on the inference result does not meet a preset threshold.

[0060] In an example embodiment, the first management network element receives the inference request message sent by the second management network element; the first management network element performs inference according to the inference request message and evaluates the performance of the machine learning model; and in the case that the inference performance of the machine learning model evaluated by the first management network element does not meet a preset threshold, the first management network element fine-tunes the machine learning pre-trained model to generate a fine-tuned machine learning model.

[0061] In step S304 of the embodiments of the present application, the machine learning pre-trained model is the machine learning pre-trained model generated by the third management network element performing model pre-training.

[0062] In step S304 of the embodiments of the present application, the machine learning pre-trained model is the machine learning pre-trained model generated by the third management network element performing model pre-training.

[0063] In an example embodiment, the first management network element sends a machine learning fine-tuning report to the second management network element.

[0064] In an example embodiment, the machine learning fine-tuning report comprises at least one of: fine-tuning indication information indicating that the machine learning model is fine-tuned; fine-tuning background information indicating background information of fine-tuning of the machine learning model, wherein the fine-tuning background information comprises at least one of: fine-tuning time, fine-tuning environment such as a physical environment in which the fine-tuning task is run; pre-training model identification indicating identification information of the machine learning pre-training model; fine-tuning data indication information indicating a source of fine-tuning data.

[0065] In an embodiment, the source of fine-tuning data can be a consumer, and the fine-tuning data indication information is used to indicate whether non-consumer-provided data is used.

[0066] In an example embodiment, the first management network element creates a machine learning model instance; wherein the machine learning model instance comprises at least one of: fine-tuning model indication information of the machine learning model indicating that the machine learning model is a fine-tuned machine learning model; indication information of the machine learning model indicating that the machine learning model is a machine learning fine-tuning model; pre-training model identification of the index of the machine learning model indicating a machine learning pre-training model on which the fine-tuned machine learning model is based; identification information of the machine learning model indicating identification of the machine learning model; fine-tuning version information of the machine learning model indicating a fine-tuning version of the machine learning model; complexity information of the machine learning model; performance information of the machine learning model; capability information of the machine learning model; context information of the machine learning model.

[0067] In an embodiment, the complexity information of the machine learning model is used to indicate model complexity of the fine-tuned machine learning model; the performance information of the machine learning model is used to indicate performance of the fine-tuned machine learning model, performance indicators supported by the fine-tuned machine learning model, and performance scores of the fine-tuned machine learning model; the capability information of the machine learning model is used to indicate capabilities of the fine-tuned machine learning model, such as aIMLInferenceName(Analytic ID, MDA Type) supported after fine-tuning; and the context information of the machine learning model is used to indicate context information of the fine-tuned machine learning model, such as running environment, compatibility, running energy consumption and time in different running environments, etc.

[0068] In this embodiment, another communication method is provided, which is applied to a second management network element and comprises: the second management network element sending first request information to a first management network element to make the first management network element fine-tune a machine learning pre-training model according to the first request information to generate a fine-tuned machine learning model.

[0069] In an example embodiment, the first request information comprises at least one of the following: fine-tuning indication information, used to indicate that the machine learning model required is generated based on fine-tuning of the machine learning pre-training model; pre-training model identification information, used to indicate the identification of the machine learning pre-training model; and model requirement information, used to indicate the capability of the machine learning model.

[0070] In an example embodiment, the first request information further comprises fine-tuning data information used to indicate fine-tuning based on the machine learning pre-training model, wherein the fine-tuning data information comprises at least one of the following: data address information, feature information, and sample identification information.

[0071] In an example embodiment, the model requirement information comprises: model capability information, used to indicate the inference performance requirement of the machine learning model; model complexity information, used to indicate the complexity of the machine learning model; and model energy consumption information, used to indicate the energy consumption of the machine learning model.

[0072] In an example embodiment, the second management network element sends the first request information to the first management network element, comprising: the second management network element sends an inference request message to the first management network element; the second management network element receives an inference result sent by the first management network element after the first management network element performs inference according to the inference request message; and the second management network element sends the first request information to the first management network element under a first preset condition, wherein the preset condition comprises that the second management network element evaluates the inference performance of the machine learning model based on the inference result and the inference performance does not satisfy a preset threshold.

[0073] In an example embodiment, the second management network element sends an inference request message to the first management network element, so that the first management network element performs inference and evaluates the performance of the machine learning model according to the inference request message.

[0074] In an example embodiment, the machine learning pre-training model is a machine learning pre-training model generated by the third management network element performing model pre-training.

[0075] In an example embodiment, the second management network element receives the machine learning fine-tuning report sent by the first management network element.

[0076] In an example embodiment, the machine learning fine-tuning report comprises at least one of the following: fine-tuning indication information, used to indicate that the machine learning model is fine-tuned; fine-tuning background information, used to indicate the background of the fine-tuning of the machine learning model, wherein the fine-tuning background information comprises fine-tuning time and fine-tuning environment; pre-training model identification, used to indicate the identification information of the machine learning pre-training model; and fine-tuning data, used to indicate the machine learning fine-tuning data source.

[0077] In the embodiment, a communication method is also provided, applied to a third management network element,Figure 4 is a flowchart of a communication method according to yet another embodiment of the present application, as shown, the flow includes the following steps: Figure 4

[0078] At step S402, the third management network element receives second request information sent by the fourth management network element for requesting the third management network element to perform pre-training on the machine learning model, wherein the second request information includes at least one of the following: pre-training model request indication information; pre-training model requirement information; and pre-training data information.

[0079] In one embodiment, the third management network element is a machine learning pre-training model producer, which can be an MLTraining Function, and the fourth management network element is a machine learning pre-training model consumer.

[0080] At step S404, the third management network element performs model pre-training according to the second request information to generate a machine learning pre-training model.

[0081] In one exemplary embodiment, the third management network element creates a machine learning model instance, wherein the machine learning model instance includes at least one of the following: indication information of the machine learning model, used to indicate that the machine learning model is a machine learning pre-training model; identification information of the machine learning model, used to indicate the identification of the machine learning pre-training model; version information of the machine learning model; model complexity information of the machine learning model; performance information of the machine learning model; capability information of the machine learning model, used to indicate the capability of the pre-training model; and context information of the machine learning model.

[0082] In one embodiment, the complexity information of the machine learning model is used to indicate the model complexity of the pre-training model of the machine learning model; the performance information of the machine learning model is used to indicate the performance of the pre-training model of the machine learning model, the performance indicators supported by the pre-training model of the machine learning model, and the performance score of the pre-training model of the machine learning model; the capability information of the machine learning model is used to indicate the capability of the pre-training model of the machine learning model, such as the aIMLInferenceName (Analytic ID, MDA Type) supported after pre-training; and the context information of the machine learning model is used to indicate the context information of the pre-training model of the machine learning model, such as the running environment, compatibility, running energy consumption and time under different running environments, etc.

[0083] ​In an example embodiment, the third management network element sends a machine learning pre-training model report to the fourth management network element; wherein the machine learning pre-training model report comprises at least one of the following: identification information of the machine learning pre-training model; context information of the machine learning pre-training model; capability information of the machine learning pre-training model; performance information of the machine learning pre-training model.

[0084] In an example embodiment, the identification information comprises a distinguished name (DN) of the machine learning model, the context information of the machine learning pre-training model comprises energy consumption, memory, time, and training node information of the pre-training task, and the performance information of the machine learning pre-training model indicates pre-training model performance such as accuracy.

[0085] In an example embodiment, the third management network element generates a machine learning pre-training model instance according to the machine learning model, wherein the machine learning pre-training model instance comprises at least one of the following: indication information of the machine learning pre-training model, used to indicate the machine learning pre-training model; version information of the machine learning pre-training model; model complexity information of the machine learning pre-training model; performance information of the machine learning pre-training model; capability information of the machine learning pre-training model; context information of the machine learning pre-training model.

[0086] In an example embodiment, the performance information of the machine learning pre-training model is used to indicate performance information of the machine learning pre-training model, such as performance indicators supported by the pre-training model, performance scores of the pre-training model, and precision (FP16, FP32, BF16) of the pre-training model; the capability information of the machine learning pre-training model is used to indicate capability information of the machine learning pre-training model, such as aIMLInferenceName (Analytic ID, MDA Type) supported after pre-training; and the context information of the machine learning pre-training model is used to indicate context information of running of the machine learning pre-training model, such as a running environment, compatibility, and running energy consumption and time in different running environments.

[0087] In an example embodiment, the third management network element generates an enhanced pre-training request instance and performs enhanced pre-training according to third request information sent by the fourth management network element; and the third management network element updates the machine learning pre-training model instance according to an enhanced pre-training result; wherein the third request information comprises at least one of the following: indication information used to indicate that the machine learning pre-training model needs to be subjected to enhanced pre-training; identification information of the machine learning pre-training model subjected to enhanced pre-training; enhanced pre-training data information of data used by the machine learning pre-training model to perform enhanced pre-training; and demand information of the machine learning pre-training model to perform enhanced pre-training.

[0088] In an embodiment, the identification information is used to indicate a pre-trained model for enhanced pre-training, such as a DN of the model; the enhanced pre-training data information is used to indicate a data source for enhanced pre-training, such as an address of the enhanced pre-training data; and the requirement information is used to indicate a requirement for enhanced pre-training, such as an energy consumption requirement, a performance requirement, and the like.

[0089] In an example embodiment, the third management network element evaluates the machine learning pre-trained model and sends a machine learning pre-trained model report to the fourth management network element.

[0090] In an example embodiment, in a case where the machine learning pre-trained model does not meet the pre-training task requirement, the third management network element performs enhanced pre-training on the machine learning pre-trained model and updates the machine learning model instance or the machine learning pre-trained model instance.

[0091] In an example embodiment, after the third management network element performs enhanced pre-training on the machine learning pre-trained model, the method further includes: the third management network element sending a machine learning enhanced pre-trained model report to the fourth management network element; and the machine learning enhanced pre-trained model report includes at least one of the following: indication information used to indicate that the machine learning enhanced pre-trained model has undergone enhanced pre-training; and enhanced pre-training data indication information used to indicate data of the machine learning enhanced pre-trained model.

[0092] In this embodiment, another method for generating a machine learning pre-trained model is provided, which is applied to a fourth management network element and includes: the fourth management network element sending second request information used to request the third management network element to perform pre-training on a machine learning model to the third management network element, so that the third management network element performs model pre-training according to the second request information to generate the machine learning pre-trained model; and the second request information includes at least one of the following: pre-trained model request indication information; pre-trained model requirement information; and pre-training data information.

[0093] In an example embodiment, the fourth management network element receives the machine learning pre-trained model instance sent by the third management network element, and the machine learning pre-trained model instance includes at least one of the following: indication information of the machine learning pre-trained model, used to indicate the machine learning pre-trained model; version information of the machine learning pre-trained model; model complexity information of the machine learning pre-trained model; performance information of the machine learning pre-trained model; capability information of the machine learning pre-trained model; and context information of the machine learning pre-trained model.

[0094] In an example embodiment, the fourth management network element sends third request information to the third management network element to make the third management network element generate an enhanced pre-training request instance and perform enhanced pre-training; wherein the third request information comprises at least one of the following: indication information for indicating that the machine learning pre-training model needs to be subjected to enhanced pre-training; identification information for indicating the machine learning pre-training model subjected to enhanced pre-training; enhanced pre-training data information for indicating the data subjected to enhanced pre-training by the machine learning pre-training model; and requirement information for indicating that the machine learning pre-training model is subjected to enhanced pre-training.

[0095] In an example embodiment, the fourth management network element receives a machine learning pre-training model report sent by the third management network element after the third management network element performs evaluation on the machine learning pre-training model.

[0096] In an example embodiment, the fourth management network element receives a machine learning enhanced pre-training model report sent by the third management network element; wherein the machine learning enhanced pre-training model report comprises at least one of the following: indication information for indicating that the machine learning enhanced pre-training model is subjected to enhanced pre-training; and enhanced pre-training data indication information for indicating the data of the machine learning enhanced pre-training model.

[0097] From the above description of the embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software and the necessary general hardware platform, and of course, it can also be implemented by hardware, but in many cases, the former is a better implementation. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as a ROM / RAM, a magnetic disk, or an optical disk) and includes a plurality of instructions for causing a terminal device (which can be a mobile phone, a computer, a server, or a network device) to execute the methods described in the various embodiments of the present application.

[0098] In order to facilitate the understanding of the technical solutions provided by the embodiments of the application, the following describes specific embodiments in conjunction with specific scenarios.

[0099] Figure 5 is a schematic diagram of a machine learning pre-training model according to an embodiment of the present application. In the present embodiment, the existing Information Object Class (IOC) model is reused, as shown in Figure 5 The Pre-training MLModel reuses the existing ML Model IOC. The ML Pre-training Producer is the ML Training Function.Figure 5 IOC or sub-network IOC.

[0100] Figure 6 is a machine learning pre-training model generation flowchart according to an embodiment of the present application, as shown in the figure, the flow includes the following steps: Figure 6

[0101] Step S601, the ML pre-training consumer requests the ML pre-training producer (MLTrainingFunction) to train a Pre-training ML Model (machine learning pre-training model). The ML pre-training consumer creates an MLTrainingRequest (machine learning training request) instance on the ML pre-training producer, which is equivalent to the fourth management network element sending the second request information to the third management network element in the above embodiment. The MLTrainingRequest instance includes at least one of the following:

[0102] Pre-training model request indication information: indicating that the model requested to be trained this time is a pre-training model;

[0103] Pre-training data information: indicating the data for training the Pre-training ML Model, which can be data address, feature (PM, KPI, etc.), sample ID (UE ID, S-NSSAI, etc.), etc.;

[0104] Pre-training model requirement information.

[0105] Step S602, the ML pre-training producer executes model pre-training according to the consumer request.

[0106] Step S603, after the pre-training model training is completed, the ML pre-training producer creates an ML Model instance, and the ML Model instance includes at least one of the following:

[0107] Machine learning pre-training model indication information: indicating that the machine learning model is an ML Pre-training Model;

[0108] Pre-training model identification information, used to indicate the identification of the machine learning pre-training model;

[0109] Pre-training model version information: indicating the version of the ML Pre-training Model;

[0110] Pre-training model complexity information: indicating the model complexity of the ML Pre-training Model;

[0111] ​Pre-training model performance information: indicating the performance information of the ML Pre-training Model, such as the performance indicators supported by the pre-training model, the performance score of the pre-training model, and the accuracy (FP16, FP32, BF16) of the pre-training model;

[0112] Pre-training model capability information: indicating the capability information of the ML Pre-training Model, such as the aIMLInferenceName (Analytic ID, MDA Type) supported after fine-tuning.

[0113] Pre-training model context information: indicating the context information of the running of the ML Pre-training Model, such as the running environment, compatibility, running energy consumption and time in different running environments, etc.

[0114] In an example embodiment, the machine learning model instance includes at least one of the following: indication information of the machine learning model, used to indicate that the machine learning model is a machine learning pre-training model; identification information of the machine learning model, used to indicate the identification of the machine learning pre-training model; version information of the machine learning model; model complexity information of the machine learning model; performance information of the machine learning model; capability information of the machine learning model, used to indicate the capability of the pre-training model; and context information of the machine learning model.

[0115] In step S604, the ML pre-training producer creates an MLTrainingReport (machine learning training report) instance at the same time when creating the MLTrainingRequest instance in step S601, fills in the parameters in the MLTrainingReport instance when completing the ML pre-training, and sends the ML Model Pre-training report to the Consumer. The report information includes at least one of the following:

[0116] Identification information of the machine learning pre-training model: DN of the ML model;

[0117] Context information of the machine learning pre-training model: energy consumption, memory, time, training node information, etc. of this pre-training task;

[0118] Capability information of the machine learning pre-training model;

[0119] Performance information of the machine learning pre-training model, indicating the performance of the pre-training model, such as accuracy.

[0120] In an example embodiment, the machine learning pre-training model in the present embodiment is as shown in Figure 5 .

[0121] Figure 7 is a machine learning pre-training model generation flowchart according to still another embodiment of the present application, as shown in the figure, the flow includes the following steps: Figure 7

[0122] Step S701, the ML pre-training consumer requests the ML pre-training producer (MLTrainingFunction) to train a Pre-training ML Model. The ML pre-training consumer creates an MLTrainingRequest instance on the ML pre-training producer, which is equivalent to the fourth management network element sending the second request information to the third management network element in the above embodiment. The MLTrainingRequest instance includes at least one of the following:

[0123] Pre-training model request indication information: indicating that the model requested to be trained this time is a pre-training model;

[0124] Pre-training data information: indicating the data for training the Pre-training ML Model, which can be data address, feature (PM, KPI, etc.), sample ID (UE ID, S-NSSAI, etc.), etc.

[0125] Pre-training model requirement information.

[0126] Step S702, the ML pre-training producer performs model pre-training according to the consumer request.

[0127] Step S703, the ML pre-training producer (MLTrainingFunction) evaluates whether the performance of the pre-training model meets the demand of the pre-training task.

[0128] Specifically, in the case where the performance meets the demand, step S704 is entered, and in the case where the performance does not meet the demand, step S706 is entered.

[0129] Step S704, after the pre-training model training is completed, the ML pre-training producer creates an ML Model instance, and machine learning pre-training model indication information: indicating that the machine learning model is an ML Pre-training Model;

[0130] Pre-training model identification information, used to indicate the identification of the machine learning pre-training model;

[0131] Pre-training model version information: indicating the version of the ML Pre-training Model;

[0132] ​Pre-training model complexity information: indicating the model complexity of the ML Pre-training Model;

[0133] Pre-training model performance information: indicating the performance information of the ML Pre-training Model, such as the performance indicators supported by the pre-training model, the performance score of the pre-training model, the precision (FP16, FP32, BF16) of the pre-training model;

[0134] Pre-training model capability information: indicating the capability information of the ML Pre-training Model, such as the aIMLInferenceName (Analytic ID, MDAType) supported after fine-tuning.

[0135] Pre-training model context information: indicating the context information of the running of the ML Pre-training Model, such as the running environment, compatibility, running energy consumption and time under different running environments, etc.

[0136] In an example embodiment, the machine learning model instance includes at least one of the following: indication information of the machine learning model, used to indicate that the machine learning model is a machine learning pre-training model; identification information of the machine learning model, used to indicate the identification of the machine learning pre-training model; version information of the machine learning model; model complexity information of the machine learning model; performance information of the machine learning model; capability information of the machine learning model, used to indicate the capability of the pre-training model; context information of the machine learning model.

[0137] In step S705, the ML pre-training producer creates an MLTrainingReport instance at the same time when creating the MLTrainingRequest instance in step S701, fills in the parameters in the MLTrainingReport instance when completing the ML pre-training, and sends the ML Model Pre-training report to the Consumer, which includes at least one of the following:

[0138] Pre-training model identification information: DN of the ML model;

[0139] Pre-training model context information: energy consumption, memory, time, training node information, etc. of this pre-training task;

[0140] Capability information of the machine learning pre-training model;

[0141] Pre-training model performance information, indicating the performance of the pre-training model, such as accuracy, etc.

[0142] Step S706, after the pre-training model training is completed, the ML pre-training producer creates an ML Model instance, and the ML Model instance includes at least one of the following:

[0143] Pre-training model indication information: indicates that the model is an ML Pre-training Model;

[0144] Pre-training model version information: indicates the version of the ML Pre-training Model;

[0145] Pre-training model complexity information: indicates the model complexity of the ML Pre-training Model;

[0146] Pre-training model performance information: indicates the performance information of the ML Pre-training Model, such as the performance indicators supported by the pre-training model, the performance score of the pre-training model, and the precision (FP16, FP32, BF16) of the pre-training model;

[0147] Pre-training model capability information: indicates the capability information of the ML Pre-training Model, such as the aIMLInferenceName (Analytic ID, MDAType) supported after fine-tuning.

[0148] Pre-training model context: indicates the running context information of the ML Pre-training Model, such as the running environment, compatibility, running energy consumption and time in different running environments, etc.

[0149] Step S707, the ML pre-training producer performs enhanced pre-training, and the producer creates an AugmentPretrainingrequest instance, which includes at least one of the following:

[0150] Augment pre-training indication information, indicating that the pre-training model needs to be augmented pre-trained;

[0151] Pre-training model identifier, indicating the pre-training model used for augmented pre-training, such as the DN of the model;

[0152] Augmented pre-training data, indicating the data source for augmented pre-training, such as the address of the augmented pre-training data;

[0153] Augmented pre-training requirement information, indicating the requirements for augmented pre-training, such as energy consumption requirements and performance requirements.

[0154] Step S708, the ML pre-training producer updates the ML Model instance created in step S704, indicating that the pre-training model has been augmented pre-trained.

[0155] Step S709, the ML pre-training producer sends an ML Pre-training report to the consumer, additionally including the following enhanced pre-training related information:

[0156] Enhanced pre-training indication information, indicating that the model has undergone enhanced pre-training;

[0157] Enhanced pre-training data, indicating the source of the data for enhanced pre-training, such as the address of additional enhanced pre-training data.

[0158] Figure 8 Fig. 7 is a schematic diagram of a machine learning pre-training model according to another embodiment of the present application. In this embodiment, a new IOC model is defined, such as Figure 8 As shown, the Pre-training ML Model is defined as a new IOC, and the MLModelRepository name is contained. The ML Pre-training Producer is also a newly defined MLPreTraningFunction. Figure 8 for representing a management unit IOC or a sub-network IOC.

[0159] Figure 9 Fig. 8 is a flowchart of a machine learning pre-training model generation process according to another embodiment of the present application. As shown in Figure 9 the flowchart includes the following steps:

[0160] Step S901, the ML pre-training consumer requests to train a Pre-training ML Model in the ML pre-training producer (MLPreTrainingFunction). The ML pre-training consumer creates an MLPreTrainingRequest instance on the ML pre-training producer, which contains at least one of the following:

[0161] Pre-training model request indication information: indicating that the model requested to be trained this time is a pre-training model;

[0162] Pre-training data information: indicating the data for training the Pre-training ML Model, which can be a data address, a feature (PM, KPI, etc.), a sample ID (UE ID, S-NSSAI, etc.), etc.

[0163] Step S902, the ML pre-training producer, i.e., the MLPreTrainingFunction, performs model pre-training according to the consumer request.

[0164] Step S903, after the pre-training model training is completed, the ML pre-training producer creates a Pre-training ML Model instance, and the Pre-training ML Model instance includes at least one of the following:

[0165] Machine learning pre-training model indication information: indicates that the model is an ML Pre-training Model;

[0166] Machine learning pre-training model version information: indicates the version of the ML Pre-training Model;

[0167] Machine learning pre-training model complexity information: indicates the model complexity of the ML Pre-training Model;

[0168] Machine learning pre-training model performance information: indicates the performance information of the ML Pre-training Model;

[0169] Machine learning pre-training model capability information: indicates the capability information of the Pre-training ML Model, such as aIMLInferenceName (Analytic ID, MDA Type) supported after fine-tuning.

[0170] Machine learning pre-training model context: indicates the running context information of the ML Pre-training Model, such as a running environment, compatibility, running energy consumption and time in different running environments, etc.

[0171] Step S904, the ML pre-training producer creates an MLPreTrainingReport instance at the same time when creating the MLPreTrainingRequest instance in step S901, fills in the parameters in the MLPreTrainingReport instance when the ML pre-training is completed, and sends an ML Model Pre-training report to the Consumer, and the report includes at least one of the following:

[0172] Machine learning pre-training model identification information: DN of the Pre-training ML Model.

[0173] Machine learning pre-training context: energy consumption, memory, time, training node information, etc. of the pre-training task;

[0174] Machine learning pre-training model performance information, indicating the pre-training model performance, such as accuracy, etc.

[0175] Step S905, the ML pre-training consumer requests the ML pre-training producer to perform the enhanced pre-training, which is equivalent to the fourth management network element sending the third request information to the third management network element in the above embodiment. The producer creates an AugmentPretrainingrequest instance, which contains at least one of the following:

[0176] AugmentPretraining indication information, indicating that the pre-training model needs to be enhanced pre-trained;

[0177] Pre-training model identifier, indicating the pre-training model used for enhanced pre-training, which can be the DN of the model;

[0178] AugmentPretraining data, indicating the data source for enhanced pre-training, such as the address of the enhanced pre-training data.

[0179] AugmentPretraining requirement information, indicating the requirement for enhanced pre-training, such as energy consumption requirement, performance requirement, etc.

[0180] Step S906, the ML pre-training producer performs the enhanced pre-training.

[0181] Step S907, the ML pre-training producer updates the Pre-training ML Model instance created in step S903.

[0182] Step S908, the ML pre-training producer creates an AugmentPreTrainingReport instance at the same time of creating the AugmentPretrainingrequest instance in step S905, and sends an ML Pre-training report to the consumer when the training is completed. The report includes the following enhanced pre-training related information:

[0183] AugmentPretraining requirement implementation, indicating that the model has been enhanced pre-trained;

[0184] AugmentPretraining data usage, indicating the additional use of the data source for enhanced pre-training, such as the address of the enhanced pre-training data.

[0185] The machine learning pre-training model in the embodiment is as shown in the following figure. Figure 8

[0186] Figure 10 ​is a machine learning model generation flowchart according to an embodiment of the present application. In the present embodiment, the ML Fine-tuning Producer can be an existing ML Training Function, or can define an ML Fine-tuning Function dedicated to supporting Fine tuning functions, to implement fine tuning based on a pre-training ML Model to generate a new ML Model. The present embodiment takes the ML Fine-tuning Function as an example, as shown in Figure 10 The flowchart includes the following steps:

[0187] Step S1001, the ML Fine-tuning Consumer requests the ML Fine-tuning Producer (ML Fine-tuning Function) to generate the required ML Model based on fine tuning. The ML Fine-tuning Consumer creates an ML Fine-tuning Request instance (equivalent to the first request information in the above embodiment) on the ML Fine-tuning Producer, which at least contains one of the following information:

[0188] Fine-tuning model request indication information (equivalent to the fine-tuning indication information in the above embodiment): indicating the required ML Model based on fine tuning;

[0189] Fine-tuning model requirement information (equivalent to the model requirement information in the above embodiment): indicating the requirement information of the model generated based on fine tuning, such as capability information, performance information, energy consumption information, complexity information of the model, etc.

[0190] Fine-tuning data information: indicating the data used to train the Fine-tuning ML Model, which can be data address, feature (PM, KPI, etc.), sample ID (UE ID, S-NSSAI, etc.), etc.

[0191] Pre-training model indication information: identification information of the pre-training model, indicating the pre-training model based on which the present fine tuning task is performed.

[0192] Fine-tuning data information: indicating the data used to train the Fine-tuning ML Model, which can be data address, feature (PM, KPI, etc.), sample ID (UE ID, S-NSSAI, etc.), etc.

[0193] Step S1002, the ML Fine-tuning Producer performs the ML Model Fine Tuning process.

[0194] Specifically, the Consumer can request to start, pause, restart and abort the Fine Tuning process.

[0195] Step S1003, after the ML fine-tuning is completed, the ML fine-tuning producer creates an ML Model instance, which includes the following information:

[0196] ML fine-tuning model indication information (equivalent to the fine-tuning model indication information of the machine learning model in the above embodiment): indicating that the model is an ML Fine-tuning Model;

[0197] ML fine-tuning version information: indicating the version of the ML Fine-tuning Model;

[0198] ML fine-tuning model complexity information: indicating the model complexity of the ML Fine-tuning Model;

[0199] ML fine-tuning model performance information: indicating the performance information of the ML Fine-tuning Model, the performance indicators supported by the ML Fine-tuning, and the performance score of the ML Fine-tuning;

[0200] ML fine-tuning model capability information: indicating the capability information of the ML Fine-tuning Model, such as the supported aIMLInferenceName (Analytic ID, MDA Type) after fine-tuning.

[0201] ML fine-tuning model context information: indicating the context information of the running of the ML Fine-tuning Model, such as the running environment, compatibility, running energy consumption and time under different running environments, etc.

[0202] Step S1004, the ML fine-tuning producer sends an ML fine-tuning report to the ML fine-tuning consumer, which at least contains one of the following:

[0203] ML fine-tuning background information: indicating that the model is subjected to ML fine-tuning and a pre-training model based on ML fine-tuning;

[0204] ML fine-tuning data: indicating the data source for enhancing the pre-training, such as the address of the enhanced pre-training data.

[0205] Figure 11 is a machine learning model generation flowchart according to another embodiment of the present application, as shown in Figure 11 the flowchart includes the following steps:

[0206] Step S1101, the AI / ML inference consumer requests the AI / ML inference producer (AI / ML Inference Function) to perform AI / ML inference, and the inference is used for energy saving strategy of a certain area. The AI / ML inference consumer creates an instance of AIMLInferenceRequest on the AI / ML inference producer.

[0207] Step S1102, the AI / ML inference producer performs AI / ML inference according to the demand of the consumer.

[0208] Step S1103, the AI / ML inference producer evaluates the inference performance of the ML Model, and in this example, the inference result does not meet the expectation.

[0209] Step S1104, the AI / ML inference producer requests the ML fine tuning function to perform ML Model fine tuning, and the request contains the following information:

[0210] Fine tuning model request indication information (equivalent to fine tuning indication information in the above embodiment): indicating the required ML Model based on fine tuning;

[0211] Fine tuning model demand information: indicating the demand information for generating the model based on fine tuning, such as energy consumption of the model, complexity of the model, etc.

[0212] Fine tuning data information: indicating the data used for training the Fine-tuning ML Model, which can be data address, feature (PM, KPI, etc.), sample ID (UE ID, S-NSSAI, etc.), etc.

[0213] Pre-trained model indication information: identification of the pre-trained model, indicating the pre-trained model based on which the fine tuning task is performed.

[0214] Step S1105, the ML fine tuning producer (ML fine tuning function) performs ML Model FineTuning process.

[0215] Step S1106, the ML fine tuning producer sends a ML fine tuning report, which contains at least one of the following:

[0216] ML fine tuning background information, indicating that the model is fine tuned based on the pre-trained model.

[0217] ML fine tuning data, indicating the source of the data for enhancing the pre-trained model, such as the address of the enhanced pre-trained data.

[0218] Step S1107, the AI / ML inference producer performs inference according to the fine tuned model.

[0219] Step S1108, the AI / ML inference producer sends an AI / ML inference report to the consumer after the inference is completed, and the report at least contains an MLModel reference indicating the Fine Tuning ML Model used for this inference task.

[0220] According to the description in the above embodiments, in the case where the performance of the ML model does not meet the expectation, the present embodiment avoids starting from scratch to retrain the model, such as Figure 12 as shown, a Pre-training and Fine-tuning stage is added to the existing LCM process, and a new ML model is generated by fine-tuning the pre-trained model, avoiding training the model from scratch. Figure 12 Machine learning testing refers to the process of verifying and evaluating machine learning models; after machine learning testing, the machine learning simulation stage is entered, in which machine learning models and algorithms are used to simulate the real world; after simulation, the trained machine learning model is read from the storage medium (such as hard disk, cloud storage, etc.) into the memory for prediction or inference; finally, the inference stage is entered, which is the process of running a set of input data through the trained ML entity to generate a set of output data (such as prediction), and in the case where the inference result is not ideal, the training stage of the machine learning model can be answered.

[0221] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0222] In an example embodiment, the above computer readable storage medium can include, but is not limited to, a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic or optical disk, and various media that can store computer programs.

[0223] The embodiments of the present application also provide an electronic device, which includes a memory storing a computer program and a processor configured to execute the computer program to perform the steps in any of the above method embodiments.

[0224] In an example embodiment, the above electronic device can further include a transmission device connected to the processor and an input / output device connected to the processor.

[0225] The specific examples in the present embodiment can refer to the examples described in the above embodiments and example embodiments, which will not be described herein again.

[0226] It should be apparent to those skilled in the art that the steps of the application described above can be implemented with general computing devices, which can be centralized on a single computing device or distributed across a network of multiple computing devices, which can be implemented with program code executable by a computing device, which can be stored in a storage device for execution by a computing device, and in some cases, the steps shown or described can be performed in a different order than shown, or made into individual integrated circuit modules, or multiple steps made into a single integrated circuit module. Thus, the present application is not limited to any particular combination of hardware and software.

[0227] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the principles of the present application should be included in the protection scope of the present application.

Claims

1. A communication method, characterized in that, include: The first management network element receives the first request information sent by the second management network element; The first management network element fine-tunes the machine learning pre-trained model based on the first request information to generate a fine-tuned machine learning model.

2. The method according to claim 1, characterized in that, The first request information includes at least one of the following: Fine-tuning instruction information is used to instruct the fine-tuning of the machine learning pre-trained model to generate the required machine learning model; Pre-trained model identification information, used to indicate the identifier of the machine learning pre-trained model; Model requirement information, used to indicate the requirements of the machine learning model.

3. The method according to claim 2, characterized in that, The first request information further includes: fine-tuning data information for instructing fine-tuning based on the machine learning pre-trained model, wherein the fine-tuning data information includes at least one of the following: data address information, feature information, and sample identification information.

4. The method according to claim 2, characterized in that, The model requirements information includes: Model capability information, used to indicate the reasoning capabilities supported by the machine learning model; Model performance information, used to indicate the performance of the machine learning model; Model complexity information, used to indicate the complexity of the machine learning model; Model energy consumption information, used to indicate the energy consumption of the machine learning model.

5. The method according to any one of claims 2 to 4, characterized in that, The first management network element receives a first request message sent by the second management network element, including: The first management network element receives the inference request message sent by the second management network element; The first management network element performs inference based on the inference request message and sends the inference result to the second management network element; The first management network element receives a first request message sent by the second management network element under preset conditions, wherein the preset conditions include the second management network element's assessment, based on the inference result, that the inference performance of the machine learning model does not meet a preset threshold.

6. The method according to claim 1, characterized in that, The method further includes: The first management network element receives the inference request message sent by the second management network element; The first management network element performs inference based on the inference request message and evaluates the performance of the machine learning model; If the inference performance of the machine learning model evaluated by the first management network element does not meet the preset threshold, the first management network element fine-tunes the machine learning pre-trained model to generate the fine-tuned machine learning model.

7. The method according to claim 1, characterized in that, The method further includes: the first management network element sending a machine learning fine-tuning report to the second management network element.

8. The method according to claim 7, characterized in that, The machine learning fine-tuning report includes at least one of the following: Fine-tuning instruction information is used to indicate that the machine learning model has undergone machine learning fine-tuning; Fine-tuning background information, used to indicate the background information after which the machine learning model has been fine-tuned, wherein the fine-tuning background information includes at least one of the following: fine-tuning time and fine-tuning environment; Pre-trained model identifier, used to indicate the identifier information of a machine learning pre-trained model; Fine-tuning data indication information is used to indicate the source of the fine-tuning data.

9. The method according to claim 1, characterized in that, The method further includes: the first management network element creating a machine learning model instance; The machine learning model instance includes at least one of the following: The fine-tuning model indication information of the machine learning model is used to indicate that the machine learning model is a fine-tuned machine learning model. The indication information of the machine learning model is used to indicate that the machine learning model is a machine learning fine-tuning model; The pre-trained model identifier of the index of the machine learning model is used to indicate the machine learning pre-trained model on which the fine-tuned machine learning model is based. The identification information of the machine learning model is used to indicate the identification of the machine learning model; The fine-tuning version information of the machine learning model is used to indicate the fine-tuning version of the machine learning model; The complexity information of the machine learning model; Performance information of the machine learning model; The capability information of the machine learning model; The contextual information of the machine learning model.

10. The method according to claim 1, characterized in that, The machine learning pre-trained model is a machine learning pre-trained model generated by pre-training the third management network element execution model.

11. A communication method, characterized in that, include: The third management network element receives a second request message from the fourth management network element requesting the third management network element to perform pre-training on the machine learning model, wherein the second request message includes at least one of the following: Pre-trained model requests instruction information; Pre-trained model requirements information; Pre-training data information; The third management network element performs model pre-training based on the second request information to generate the machine learning pre-trained model.

12. The method according to claim 11, characterized in that, The method further includes: the third management network element creating a machine learning model instance, wherein the machine learning model instance includes at least one of the following: The indication information of the machine learning model is used to indicate that the machine learning model is a machine learning pre-trained model; The identification information of the machine learning model is used to indicate the identification of the machine learning pre-trained model; Version information of the machine learning model; The model complexity information of the machine learning model; Performance information of the machine learning model; The capability information of the machine learning model is used to indicate the capability of the pre-trained model; The contextual information of the machine learning model.

13. The method according to claim 11, characterized in that, The method further includes: the third management network element sending a machine learning pre-trained model report to the fourth management network element; The machine learning pre-trained model report includes at least one of the following: The identification information of the machine learning pre-trained model; Contextual information of the machine learning pre-trained model; The capability information of the machine learning pre-trained model; The performance information of the machine learning pre-trained model.

14. The method according to claim 13, characterized in that, The method further includes: the third management network element generating a machine learning pre-trained model instance based on the machine learning model, wherein the machine learning pre-trained model instance includes at least one of the following: The instruction information of the machine learning pre-trained model is used to instruct the machine learning pre-trained model; Version information of the machine learning pre-trained model; The model complexity information of the machine learning pre-trained model; Performance information of the machine learning pre-trained model; The capability information of the machine learning pre-trained model; The contextual information of the machine learning pre-trained model.

15. The method according to claim 14, characterized in that, After the third management network element generates a machine learning pre-trained model instance based on the machine learning model, it also includes: The third management network element generates an enhanced pre-training request instance based on the third request information sent by the fourth management network element and performs enhanced pre-training. The third management network element updates the machine learning pre-trained model instance based on the enhanced pre-training results; The third request information includes at least one of the following: Indication information used to indicate that the machine learning pre-trained model needs to undergo augmented pre-training; Identification information used to indicate the machine learning pre-trained model undergoing augmentation pre-training; Augmented pre-training data information used to instruct the machine learning pre-training model to perform augmented pre-training data; Information used to indicate the need for enhanced pre-training of the machine learning pre-trained model.

16. The method according to claim 11, characterized in that, The method further includes: the third management network element evaluating the machine learning pre-trained model and sending a machine learning pre-trained model report to the fourth management network element.

17. The method according to claim 16, characterized in that, The method further includes: If the machine learning pre-trained model does not meet the requirements of the pre-training task, the third management network element performs enhanced pre-training on the machine learning pre-trained model and updates the machine learning model instance or the machine learning pre-trained model instance.

18. The method according to claim 15 or 17, characterized in that, After the third management network element performs augmented pre-training on the machine learning pre-trained model, it further includes: The third management network element sends a machine learning enhancement pre-trained model report to the fourth management network element; The machine learning augmented pre-trained model report includes at least one of the following: Indicative information used to indicate that the machine learning augmentation pre-trained model has undergone augmentation pre-training; Augmented pre-training data indication information used to indicate the data of the machine learning augmented pre-training model.

19. A computer program product, comprising a computer program and instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 4, 6 to 10, or 11 to 17.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of claims 1 to 4, 6 to 10, or 11 to 17.

21. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 4, 6 to 10, or 11 to 17.

Images