WO2023185726A1 - 模型获取方法、信息发送方法、信息接收方法、装置及网元 - Google Patents

模型获取方法、信息发送方法、信息接收方法、装置及网元 Download PDF

Info

Publication number
WO2023185726A1
WO2023185726A1 PCT/CN2023/084048 CN2023084048W WO2023185726A1 WO 2023185726 A1 WO2023185726 A1 WO 2023185726A1 CN 2023084048 W CN2023084048 W CN 2023084048W WO 2023185726 A1 WO2023185726 A1 WO 2023185726A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
information
network element
request message
requirement
Prior art date
Application number
PCT/CN2023/084048
Other languages
English (en)
French (fr)
Inventor
崇卫微
Original Assignee
维沃移动通信有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 维沃移动通信有限公司 filed Critical 维沃移动通信有限公司
Publication of WO2023185726A1 publication Critical patent/WO2023185726A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/10—File systems; File servers
    • G06F16/14—Details of searching files based on file metadata
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/10—File systems; File servers
    • G06F16/14—Details of searching files based on file metadata
    • G06F16/148—File search processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/10—File systems; File servers
    • G06F16/17—Details of further file system functions
    • G06F16/172—Caching, prefetching or hoarding of files
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/56—Provisioning of proxy services
    • H04L67/568—Storing data temporarily at an intermediate stage, e.g. caching
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/02—Traffic management, e.g. flow control or congestion control
    • H04W28/10—Flow control between communication endpoints
    • H04W28/14—Flow control between communication endpoints using intermediate storage

Definitions

  • This application belongs to the field of communication technology, and specifically relates to a model acquisition method, an information sending method, an information receiving method, a device and a network element.
  • the training function entity in the communication network can be used to train the model and provide the trained model to model consumers. Since each training function entity is deployed in a decentralized manner, the generated model is distributed on different training function entities. Before model consumers request model operations, they need to use proprietary network element equipment to perform training functions and model discovery operations.
  • Embodiments of the present application provide a model acquisition method, information sending method, information receiving method, device and network element, which can solve the problem of high signaling and data transmission overhead when multiple model consumers acquire models.
  • the first aspect provides a model acquisition method, which includes:
  • the first network element receives a model storage request message sent by the second network element, where the model storage request message includes model information of at least one model, and the model information includes at least one of a model file and model file storage information;
  • the first network element stores model information of the at least one model.
  • an information sending method includes:
  • the second network element sends a model storage request message to the first network element, where the model storage request message includes model information of at least one model, and the model information includes at least one of a model file and model file storage information.
  • the third aspect provides a model acquisition method, including:
  • the third network element sends a first model request message to the first network element, where the first model request message includes model requirement information;
  • the third network element receives the model information of the target model sent by the first network element, and the target model and the model The model requires information matching, and the model information includes at least one of a model file and model file storage information.
  • the fourth aspect provides a method of receiving information, including:
  • the fifth network element receives the network element query request message sent by the first network element.
  • the network element query request message includes network element capability requirement information.
  • the network element capability requirement information is determined by the first network element according to the model requirement information.
  • the network element capability requirement information is used to determine a fourth network element, and the fourth network element can provide model information of a model that matches the model requirement information;
  • the fifth network element sends a network element query response message to the first network element, where the network element query response message includes network element information matching the network element capability requirement information.
  • a model acquisition device including:
  • a receiving module configured to receive a model storage request message sent by the second network element, where the model storage request message includes model information of at least one model, and the model information includes at least one of a model file and model file storage information;
  • a storage module configured to store model information of the at least one model.
  • an information sending device including:
  • a sending module configured to send a model storage request message to the first network element, where the model storage request message includes model information of at least one model, and the model information includes at least one of a model file and model file storage information.
  • a model acquisition device including:
  • a sending module configured to send a first model request message to the first network element, where the first model request message includes model requirement information
  • a receiving module configured to receive model information of a target model sent by the first network element, where the target model matches the model requirement information, and the model information includes at least one of a model file and model file storage information.
  • an information receiving device including:
  • a receiving module configured to receive a network element query request message sent by the first network element.
  • the network element query request message includes network element capability requirement information.
  • the network element capability requirement information is the first network element requirement information according to the model. It is determined that the network element capability requirement information is used to determine a fourth network element, and the fourth network element can provide model information of a model that matches the model requirement information;
  • a sending module configured to send a network element query response message to the first network element, where the network element query response message includes network element information matching the network element capability requirement information.
  • a first network element including a processor and a communication interface.
  • the communication interface is configured to receive a model storage request message sent by a second network element.
  • the model storage request message includes a model of at least one model.
  • Information, the model information includes at least one of a model file and model file storage information; the processor is configured to store model information of the at least one model.
  • a second network element including a processor and a communication interface.
  • the communication interface is used to send a model storage request message to the first network element.
  • the model storage request message includes model information of at least one model.
  • the model information includes at least one of a model file and model file storage information.
  • a third network element including a processor and a communication interface, the communication interface being used to A network element sends a first model request message, the first model request message includes model requirement information; receives the model information of the target model sent by the first network element, and the target model matches the model requirement information, so The model information includes at least one of model files and model file storage information.
  • a fifth network element including a processor and a communication interface.
  • the communication interface is used to receive a network element query request message sent by the first network element.
  • the network element query request message includes a network element query request message.
  • Capability requirement information is determined by the first network element according to the model requirement information.
  • the network element capability requirement information is used to determine a fourth network element.
  • the fourth network element can provide the required information. model information of a model that matches the model requirement information; and sends a network element query response message to the first network element, where the network element query response message includes network element information that matches the network element capability requirement information.
  • a communication system including: a first network element, a second network element, a third network element, a fourth network element and a fifth network element, the first network element being used to perform the first
  • the second network element is used to perform the method described in the second aspect
  • the third network element is used to perform the method described in the third aspect
  • the fourth network element is used to perform the third aspect.
  • the fifth network element is used to perform the method described in the fourth aspect.
  • a readable storage medium is provided. Programs or instructions are stored on the readable storage medium. When the programs or instructions are executed by a processor, the implementation of the first aspect, the second aspect, the third aspect, or The steps of the method described in the fourth aspect.
  • a chip in a fifteenth aspect, includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement the first aspect and the second aspect. The steps of the method described in the aspect, the third aspect or the fourth aspect.
  • a computer program/program product is provided, the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement the first aspect, the The steps of the method described in the second, third or fourth aspect.
  • the first network element receives a model storage request message sent by the second network element.
  • the model storage request message includes model information of at least one model.
  • the model information includes a model file and model file storage information. At least one item of; the first network element stores model information of the at least one model.
  • Figure 1 is a block diagram of a wireless communication system applicable to the embodiment of the present application.
  • Figure 2 is one of the flow charts of the model acquisition method provided by the embodiment of the present application.
  • Figure 3 is a flow chart of the information sending method provided by the embodiment of the present application.
  • Figure 4 is the second flow chart of the model acquisition method provided by the embodiment of the present application.
  • Figure 5 is a flow chart of an information receiving method provided by an embodiment of the present application.
  • Figure 6a is one of the schematic diagrams of information interaction between network elements provided by the embodiment of the present application.
  • Figure 6b is the second schematic diagram of information interaction between network elements provided by the embodiment of the present application.
  • Figure 7 is one of the structural diagrams of the model acquisition device provided by the embodiment of the present application.
  • Figure 8 is a structural diagram of an information sending device provided by an embodiment of the present application.
  • Figure 9 is the second structural diagram of the model acquisition device provided by the embodiment of the present application.
  • Figure 10 is a structural diagram of an information receiving device provided by an embodiment of the present application.
  • Figure 11 is a structural diagram of a communication device provided by an embodiment of the present application.
  • Figure 12 is a structural diagram of a network-side device provided by an embodiment of the present application.
  • first, second, etc. in the description and claims of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances so that the embodiments of the present application can be practiced in sequences other than those illustrated or described herein, and that "first" and “second” are distinguished objects It is usually one type, and the number of objects is not limited.
  • the first object can be one or multiple.
  • “and/or” in the description and claims indicates at least one of the connected objects, and the character “/" generally indicates that the related objects are in an "or” relationship.
  • LTE Long Term Evolution
  • LTE-Advanced, LTE-A Long Term Evolution
  • LTE-A Long Term Evolution
  • CDMA Code Division Multiple Access
  • TDMA Time Division Multiple Access
  • FDMA Frequency Division Multiple Access
  • OFDMA Orthogonal Frequency Division Multiple Access
  • SC-FDMA Single-carrier Frequency Division Multiple Access
  • NR New Radio
  • FIG. 1 shows a block diagram of a wireless communication system to which embodiments of the present application are applicable.
  • the wireless communication system includes a terminal 11 and a network side device 12.
  • the terminal 11 may be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer), or a notebook computer, a personal digital assistant (Personal Digital Assistant, PDA), a palmtop computer, a netbook, or a super mobile personal computer.
  • Tablet Personal Computer Tablet Personal Computer
  • laptop computer laptop computer
  • PDA Personal Digital Assistant
  • PDA Personal Digital Assistant
  • UMPC ultra-mobile personal computer
  • UMPC mobile Internet device
  • MID mobile Internet Device
  • AR augmented reality
  • VR virtual reality
  • robots wearable devices
  • VUE vehicle user equipment
  • PUE pedestrian terminal
  • smart home with wireless Home equipment with online communication functions, such as refrigerators, TVs, washing machines or furniture, etc.
  • PCs personal computers
  • Wearable devices include: smart watches, smart phones Rings, smart headphones, smart glasses, smart jewelry (smart bracelets, smart bracelets, smart rings, smart necklaces, smart anklets, smart anklets, etc.), smart wristbands, smart clothing, etc.
  • the network side device 12 may include an access network device or a core network device, where the access network device may also be called a radio access network device, a radio access network (Radio Access Network, RAN), a radio access network function or a wireless access network unit.
  • Access network equipment may include a base station, a Wireless Local Area Network (WLAN) access point or a WiFi node, etc.
  • WLAN Wireless Local Area Network
  • the base station may be called a Node B, an Evolved Node B (eNB), an access point, a base transceiver station ( Base Transceiver Station (BTS), radio base station, radio transceiver, Basic Service Set (BSS), Extended Service Set (ESS), home B-node, home evolved B-node, transmitting and receiving point ( Transmitting Receiving Point (TRP) or some other appropriate terminology in the field, as long as the same technical effect is achieved, the base station is not limited to specific technical terms. It should be noted that in the embodiment of this application, only in the NR system The base station is introduced as an example, and the specific type of base station is not limited.
  • this embodiment of the present application provides a model acquisition method, which includes the following steps:
  • Step 201 The first network element receives a model storage request message sent by the second network element.
  • the model storage request message includes model information of at least one model.
  • the model information includes at least one of a model file and model file storage information. .
  • Model files are used to store information related to the model, such as the structure of the model, parameters of the model, etc.
  • the model file storage information may be network element information (network element identification, network element address, etc.) stored in the model file or model file download address information.
  • the first network element may include a data storage function network element, such as an Analytics Data Repository Function (ADRF) network element, and the second network element may include a model training function (Model Training Logical Function, MTLF) network element.
  • ADRF Analytics Data Repository Function
  • MTLF Model Training Logical Function
  • NWDAF Network Data Analysis Function
  • the NWDAF contains MTLF (containing MLTF).
  • At least one model may be a model obtained by training on the second network element, or may be a model obtained by training on other network elements.
  • the other network elements send the model information of at least one model to the second network element, and the second network element The network element is then sent to the first network element for storage.
  • Step 202 The first network element stores model information of the at least one model.
  • the first network element can store the model file locally, and when other network elements request the model, it can send the model file of the requested model to other network elements; the first network element can also store the model file storage information locally, and in When other network elements request a model, the model file storage information of the requested model can be sent to other network elements, and the other network elements can then obtain the model file based on the storage information.
  • the first network element receives a model storage request message sent by the second network element.
  • the model storage request message The information includes model information of at least one model, and the model information includes at least one of a model file and model file storage information; the first network element stores the model information of the at least one model.
  • the first network element can store the model information of at least one model locally, and the model consumer only needs to request the first network element to obtain the model, which can simplify the model acquisition process and reduce signaling and data transmission overhead.
  • model information also includes model description information
  • model description information includes at least one of the following:
  • the framework information (framework) on which the model is based is used to indicate the artificial intelligence (AI) training framework (training framework) on which the model training is based; for example, common model frameworks include TensorFlow, Pytorch, Caffe2, etc.
  • the model description method information includes model format information or model language information, which is used to indicate that the trained model is represented in the model format or the model language; the model format information Or model language information includes Open Neural Network Exchange (ONNX) language.
  • the description method information is used to indicate the platform framework or language that the model needs to be based on for interoperability between different functional entities (such as analytical logic function (Analytics Logical Function, AnLF) and MTLF, or between two MTLFs).
  • optimization algorithm information based on the model is used to indicate the algorithm used for model convergence during the model training process, such as gradient descent (Gradient Descent) method, stochastic gradient descent (Stochastic Gradient Descent, SGD) method, mini-batch Gradient descent method (mini-batch gradient descent), momentum method (Momentum), etc.
  • Accuracy information that the model can achieve (also called accuracy information); this information is used to indicate the accuracy of the output results that the trained model can achieve, specifically indicating the performance of the model during the training phase or testing phase.
  • the accuracy of the model output results For example, MTLF can set up a verification data set to evaluate the accuracy of the model.
  • the verification data set includes the input data of the model and the corresponding label data.
  • MTLF inputs the input data into the trained model to obtain the output data, and then compares the outputs. Whether the data is consistent with the label data, and then the accuracy of the model is obtained based on the corresponding algorithm (such as the proportion of consistent results).
  • Model storage space information including the storage space occupied by the model itself or the total storage space required to run the model.
  • Model computing power requirement information used to indicate the computing power requirements required to run the model for inference tasks, such as "floating-point operations per second", “peak speed per second” (i.e. Floating-point operations per second ,FLOPS).
  • Model version information the version value corresponding to the model instance.
  • model description information may also include at least one of the following:
  • Model function information model function or type (model type) information, which is used to indicate the function or role of the stored model, such as for image recognition, business classification, etc.;
  • Model identification information model ID, which is used as an index to refer to the model.
  • Manufacturer information used to indicate the manufacturer of the MTLF or NWDAF (containing MLTF) of the training model.
  • model storage request message also includes other information, which includes one or more of the following:
  • Analysis ID (Analytics ID) this information is used to indicate that the model instance corresponds to the inference task indicated by the analytics ID, or is used to indicate that the model can be used as the model inference corresponding to the analytics ID.
  • Model target time information the time period for which the model is applicable
  • Model applicable object for example, used to indicate that the model is trained for a certain UE, UE group, or all UEs; or, used to indicate that the model is suitable for a certain UE, UE group, or all UEs;
  • Model filter information including Single Network Slice Selection Assistance Information (S-NSSAI), Data Network Name (DNN), area of interest (area of interest) )wait.
  • S-NSSAI Single Network Slice Selection Assistance Information
  • DNN Data Network Name
  • area of interest area of interest
  • the first network element stores other information. Further, the first network element can also store the network element identifier, network element address and other information of the second network element.
  • the method further includes:
  • the first network element receives a first model request message sent by the third network element, where the first model request message includes model requirement information;
  • the first network element determines a target model according to the model requirement information, and the target model matches the model requirement information;
  • the first network element sends the model information of the target model to the third network element.
  • the third network element may be a model consumption network element, such as an analytical logic function (Analytics Logical Function, AnLF) network element or NWDAF (containing AnLF).
  • AnLF Analytics Logical Function
  • NWDAF containing AnLF
  • the model requirement information includes at least one of the following:
  • the model consuming network element Based on some kind of AI training framework;
  • the description method information on which the model is based uses this information to indicate that the requested model must be based on a certain language or format; that is, the model consuming network element indicates the requested model in order to understand or run the model.
  • the model must be based on some model description method or format, such as ONNX.
  • optimization algorithm information on which the model is based.
  • the model consuming network element uses this information to specify that the requested model must be based on a certain optimization algorithm.
  • the model consuming network element uses this information to specify the accuracy requirements of the output results that the requested model must be able to achieve.
  • Model storage space information The model consuming network element uses this information to specify the storage space size requirement occupied by the requested model itself or the total storage space size requirement occupied during running the model. For example, the model itself is less than 5M or the total storage space during running the model is less than 10M.
  • Model computing power requirement information The model consuming network element uses this information to indicate requesting a model that meets the computing power requirement. For example, when running the model, the computing power requirement is no more than 500 floating-point operations per second (FLOPS).
  • FLOPS floating-point operations per second
  • Model version requirement information version requirements corresponding to model instances.
  • Model requirements information also includes one or more of the following:
  • Model function or type (model type) information.
  • the model informant uses this information to indicate the function or role of the requested model, such as for image recognition, business classification, etc.;
  • Model ID Model identification (model ID) information.
  • the model messager uses the model ID to refer to the request for a specific model instance.
  • Vendor requirement information used to indicate the vendor requirements of the MTLF or NWDAF (containing MLTF) that generates the model.
  • the first model request message also includes other requirement information, and the other requirement information includes at least one of the following:
  • Analytics ID this information is used to indicate that the model consumer requests the model corresponding to the inference task indicated by the analytics ID, or is used to indicate that the requested model can be used as the model inference corresponding to the analytics ID.
  • Model target time information the time period for which the model is applicable
  • Model applicable object for example, used to indicate that the model is trained for a certain UE, UE group, or all UEs; or, used to indicate that the model needs to be applied to a certain UE, UE group, or all UEs;
  • Model filter information including S-NSSAI, DNN, area of interest, etc.
  • the first network element determines the target model according to the model requirement information and the stored model description information.
  • the target model may include one or more models, and the model description information of the one or more models (models) can meet or match the model requirement information. .
  • one or more models are determined based on the model function or type (model type) information in the model requirement information, and the model function or type (model type) information in the model description information of the one or more models corresponds thereto;
  • one or more models are determined according to the framework requirements or description method requirement information on which the model is based in the model requirement information.
  • the framework or description method information on which the model is based in the model description information of the one or more models can be Meet the framework requirements or description method requirements information on which the above model is based;
  • one or more models are determined according to the precision/accuracy requirement information in the model requirement information, and the precision/accuracy information in the model description information of the one or more models can satisfy the precision/accuracy requirement information. . And so on.
  • the model description information of the target model matches the model requirement information one-to-one.
  • the framework information on which the target model is based is the same as the framework information on which the model in the model requirement information is based, and/or, the model in the model description information
  • the achievable accuracy information is the same as the accuracy requirement information in the model requirement information
  • the model storage space information in the model description information is the same as the storage space requirement information in the model requirement information, and so on.
  • the first network element After determining the target model, the first network element sends a model feedback message to the third network element.
  • the model feedback message Carry model information of the target model.
  • the model feedback message includes all or part of the model description information corresponding to one or more models
  • the model feedback message also includes all or part of other information corresponding to one or more models.
  • the first network element matches the locally stored model description information according to the model requirement information. If the match is successful, that is, when the first network element includes a model that matches the model requirement information, the first network element A model matching the model requirement information is determined as the target model. If the matching is unsuccessful, the first network element sends a response message to the second network element, and the response message is used to indicate that the model request failed. Optionally, the response message may carry the reason value for model request failure: the model does not exist.
  • the matching is unsuccessful, that is, if the first network element does not include a model that matches the model requirement information, first obtain a model that matches the model requirement information, specifically as follows:
  • the first network element sends a second model request message to a fourth network element, where the second model request message includes the model requirement information; the first network element receives a model response message sent by the fourth network element.
  • the model response message includes model information of a model that matches the model requirement information; and the first network element determines the model that matches the model requirement information as the target model.
  • the fourth network element can be an MTLF or NWDAF (containing MTLF) network element. There may be multiple fourth network elements, and the second network element and the fourth network element may be different network elements.
  • the first network element determines the network element capability requirement information according to the first model requirement information, and then matches the network element capability requirement information with the pre-obtained network element capability information to determine the fourth network element, or the first network element
  • the network element determines the fourth network element by querying the fifth network element.
  • the specific process is:
  • the first network element sends a network element query request message to the fifth network element.
  • the network element query request message includes network element capability requirement information, wherein the network element capability requirement information is the first network element based on the The above-mentioned first model requires information to be determined;
  • the first network element receives a network element query response message sent by the fifth network element, where the network element query response message includes network element information matching the network element capability requirement information;
  • the first network element determines the fourth network element based on the network element information, and the fourth network element can provide model information of a model that matches the model requirement information.
  • the fifth network element includes a Network Repository Function (NRF) network element.
  • NRF Network Repository Function
  • Network element capability requirement information includes at least one of the following:
  • Model function or type (model type) information.
  • the second network element can publish and store the trained model to the first network element, and at the same time, also publish the description information corresponding to the model to the third network element. on one network element.
  • the model consuming network element uniformly queries the first network element to obtain the model matching the requirements. It avoids the need for different model consuming network elements to obtain models from different training function entities in a scattered manner, thus avoiding a large number of point-to-point network element discovery operations. It also eliminates the process of repeatedly requesting to send models in the point-to-point method, saving time and money. Network signaling and data overhead.
  • this embodiment of the present application also provides an information sending method, which includes the following steps:
  • Step 301 The second network element sends a model storage request message to the first network element, where the model storage request message includes model information of at least one model, and the model information includes at least one of a model file and model file storage information.
  • Model files are used to store information related to the model, such as the structure of the model, parameters of the model, etc.
  • the model file storage information may be network element information (network element identification, network element address, etc.) stored in the model file or model file download address information.
  • the first network element may include a data storage function network element, such as an ADRF network element, and the second network element may include an MTLF network element or a NWDAF (containing MTLF) network element.
  • a data storage function network element such as an ADRF network element
  • the second network element may include an MTLF network element or a NWDAF (containing MTLF) network element.
  • At least one model may be a model obtained by training on the second network element, or may be a model obtained by training on other network elements.
  • the other network elements send the model information of at least one model to the second network element, and the second network element The network element is then sent to the first network element for storage.
  • the second network element sends a model storage request message to the first network element.
  • the model storage request message includes model information of at least one model.
  • the model information includes at least one of a model file and model file storage information. item.
  • the second network element sends the obtained model information of at least one model to the first network element for storage.
  • the model consumer only needs to request the first network element to obtain the model, which can simplify the model acquisition process and reduce signaling. and data transmission overhead.
  • model information also includes model description information
  • model description information includes at least one of the following:
  • the framework information (framework) on which the model is based is used to indicate the AI training framework on which the model training is based; for example, common model frameworks include TensorFlow, Pytorch, Caffe2, etc.
  • the model description method information includes model format information or model language information, which is used to indicate that the trained model is represented in the model format or the model language; the model format information Or model language information includes ONNX language.
  • the description method information is used to indicate the platform framework or language that the model needs to be based on for interoperability between different functional entities (such as AnLF and MTLF, or between two MTLFs).
  • optimization algorithm information is used to indicate the algorithm used for model convergence during the model training process, such as the gradient descent (Gradient Descent) method, the stochastic gradient descent (Stochastic Gradient Descent, SGD) method, Mini-batch gradient descent method (mini-batch gradient descent), momentum method (Momentum) wait.
  • gradient descent Gradient Descent
  • stochastic gradient descent stochastic Gradient Descent, SGD
  • Mini-batch gradient descent mini-batch gradient descent
  • momentum method Momentum
  • the precision information that the model can achieve (also called accuracy information); this precision information is used to indicate the accuracy of the output results that the trained model can achieve, specifically indicating the accuracy of the model in the training phase or testing phase. How accurate the presented model output is.
  • MTLF can set up a verification data set to evaluate the accuracy of the model.
  • the verification data set includes the input data of the model and the corresponding label data.
  • MTLF inputs the input data into the trained model to obtain the output data, and then compares the outputs. Whether the data is consistent with the label data, and then the accuracy of the model is obtained based on the corresponding algorithm (such as the proportion of consistent results).
  • Model storage space information including the storage space occupied by the model itself or the total storage space required to run the model.
  • Model computing power requirement information used to indicate the computing power requirements required to run the model for inference tasks, such as FLOPS (i.e. "floating point operations per second", “peak speed per second”).
  • Model version information the version value corresponding to the model instance.
  • model description information may also include at least one of the following:
  • Model function information model function or type (model type) information, which is used to indicate the function or role of the stored model, such as for image recognition, business classification, etc.;
  • Model identification information model ID, which is used as an index to refer to the model.
  • Manufacturer information used to indicate the manufacturer of the MTLF or NWDAF (containing MLTF) of the training model.
  • model storage request message also includes other information, which includes one or more of the following:
  • Analysis ID (Analytics ID) this information is used to indicate that the model instance corresponds to the inference task indicated by the analytics ID, or is used to indicate that the model can be used as the model inference corresponding to the analytics ID.
  • Model target time information the time period for which the model is applicable
  • Model applicable object for example, used to indicate that the model is trained for a certain UE, UE group, or all UEs; or, used to indicate that the model is suitable for a certain UE, UE group, or all UEs;
  • Model filter information including S-NSSAI, DNN, area of interest, etc.
  • the method before the second network element sends a model storage request message to the first network element, the method further includes:
  • the second network element trains the at least one model and obtains model information of the at least one model.
  • the second network element trains to obtain the at least one model
  • the second network element receives the model information of the at least one model.
  • the second network element obtains the model information of the at least one model from other network elements.
  • the above methods before the second network element sends a model storage request message to the first network element, the above methods also include:
  • the second network element receives a second model request message sent by the first network element, the second model request message includes model requirement information, and the at least one model is a model that matches the model requirement information.
  • the second network element receives the second model request message, determines at least one model according to the model requirement information, and carries the at least one model in the model storage request message and sends it to the first network element.
  • the model requirement information includes at least one of the following:
  • the model consuming network element Based on some kind of AI training framework;
  • the description method information on which the model is based uses this information to indicate that the requested model must be based on a certain language or format; that is, the model consuming network element indicates the requested model in order to understand or run the model.
  • the model must be based on some model description method or format, such as ONNX.
  • optimization algorithm information on which the model is based.
  • the model consuming network element uses this information to specify that the requested model must be based on a certain optimization algorithm.
  • the model consuming network element uses this information to specify the accuracy requirements of the output results that the requested model must be able to achieve.
  • Model storage space information The model consuming network element uses this information to specify the storage space size requirement occupied by the requested model itself or the total storage space size requirement occupied during running the model. For example, the model itself is less than 5M or the total storage space during running the model is less than 10M.
  • Model computing power requirement information The model consuming network element uses this information to indicate requesting a model that meets the computing power requirement. For example, when running this model, the computing power requirement is not higher than 500 FLOPS.
  • Model version requirement information version requirements corresponding to model instances.
  • Model requirements information also includes one or more of the following:
  • Model function or type (model type) information.
  • the model informant uses this information to indicate the function or role of the requested model, such as for image recognition, business classification, etc.;
  • Model ID Model identification (model ID) information.
  • the model messager uses the model ID to refer to the request for a specific model instance.
  • Vendor requirement information used to indicate the vendor requirements of the MTLF or NWDAF (containing MLTF) that generates the model.
  • the first model request message also includes other requirement information, and the other requirement information includes at least one of the following:
  • Analytics ID this information is used to indicate that the model consumer requests the model corresponding to the inference task indicated by the analytics ID, or is used to indicate that the requested model can be used as the model inference corresponding to the analytics ID.
  • Model target time information the time period for which the model is applicable
  • Model applicable object such as used to indicate that the model is trained for a certain UE, UE group or all UEs; Or, used to indicate that the model is applicable to a certain UE, UE group or all UEs;
  • Model filter information including S-NSSAI, DNN, area of interest, etc.
  • the first network element determines the target model according to the model requirement information and the stored model description information.
  • the target model may include one or more models, and the model description information of the one or more models (models) can meet or match the model requirement information. .
  • one or more models are determined based on the model function or type (model type) information in the model requirement information, and the model function or type (model type) information in the model description information of the one or more models corresponds thereto;
  • one or more models are determined according to the framework requirements or description method requirement information on which the model is based in the model requirement information.
  • the framework or description method information on which the model is based in the model description information of the one or more models can be Meet the framework requirements or description method requirements information on which the above model is based;
  • one or more models are determined according to the precision/accuracy requirement information in the model requirement information, and the precision/accuracy information in the model description information of the one or more models can satisfy the precision/accuracy requirement information. . And so on.
  • the at least one model is a model that matches the model requirement information. It can be understood that the model description information of the model in the at least one model matches the model requirement information one by one.
  • the framework information on which the model is based matches the framework information on which the model is based in the model requirement information, and/or, the achievable accuracy information of the model in the model description information matches the accuracy requirement information in the model requirement information, and/or,
  • the model storage space information in the model description information matches the storage space requirement information in the model requirement information, and so on.
  • this embodiment of the present application also provides a model acquisition method, which includes the following steps:
  • Step 401 The third network element sends a first model request message to the first network element, where the first model request message includes model requirement information.
  • the first model request message is used to request model information of a model that matches the model requirement information.
  • the first network element may include a data storage function network element, such as an ADRF network element, and the third network element may be a model consumption network element, such as an AnLF network element or NWDAF (containing AnLF).
  • Step 402 The third network element receives the model information of the target model sent by the first network element.
  • the target model matches the model requirement information.
  • the model information includes at least one of a model file and model file storage information. item.
  • Model files are used to store information related to the model, such as the structure of the model, parameters of the model, etc.
  • the model file storage information may be network element information (network element identification, network element address, etc.) stored in the model file or model file download address information.
  • the third network element sends a first model request message to the first network element, where the first model request message includes model requirement information; the third network element receives the model of the target model sent by the first network element. information, the target model matches the model requirement information, and the model information includes at least one of a model file and model file storage information.
  • the third network element can request a model that matches the model requirement information from the first network element, which can simplify the model acquisition process and reduce signaling and data transmission overhead.
  • model information also includes model description information
  • model description information includes at least one of the following:
  • the framework information (framework) on which the model is based is used to indicate the AI training framework on which model training is based; for example, common model frameworks include TensorFlow, Pytorch, Caffe2, etc.
  • the model description method information includes model format information or model language information, which is used to indicate that the trained model is represented in the model format or the model language; the model format information Or model language information includes ONNX language.
  • the description method information is used to indicate the platform framework or language that the model needs to be based on for interoperability between different functional entities (such as AnLF and MTLF, or between two MTLFs).
  • optimization algorithm information based on the model is used to indicate the algorithm used for model convergence during the model training process, such as gradient descent (Gradient Descent) method, stochastic gradient descent (Stochastic Gradient Descent, SGD) method, mini-batch Gradient descent method (mini-batch gradient descent), momentum method (Momentum), etc.
  • Accuracy information that the model can achieve (also called accuracy information); this information is used to indicate the accuracy of the output results that the trained model can achieve, specifically indicating the performance of the model during the training phase or testing phase.
  • the accuracy of the model output results For example, MTLF can set up a verification data set to evaluate the accuracy of the model.
  • the verification data set includes the input data of the model and the corresponding label data.
  • MTLF inputs the input data into the trained model to obtain the output data, and then compares the outputs. Whether the data is consistent with the label data, and then the accuracy of the model is obtained based on the corresponding algorithm (such as the proportion of consistent results).
  • Model storage space information including the storage space occupied by the model itself or the total storage space required to run the model.
  • Model computing power requirement information used to indicate the computing power requirements required to run the model for inference tasks, such as FLOPS (i.e. "floating point operations per second", “peak speed per second”).
  • Model version information the version value corresponding to the model instance.
  • model description information may also include at least one of the following:
  • Model function information model function or type (model type) information, which is used to indicate the function or role of the stored model, such as for image recognition, business classification, etc.;
  • Model identification information model ID, which is used as an index to refer to the model.
  • Manufacturer information used to indicate the manufacturer of the MTLF or NWDAF (containing MLTF) of the training model.
  • model storage request message also includes other information, which includes one or more of the following:
  • Analysis ID (Analytics ID) this information is used to indicate that the model instance corresponds to the inference task indicated by the analytics ID, or is used to indicate that the model can be used as the model inference corresponding to the analytics ID.
  • Model target time information the time period for which the model is applicable
  • Model applicable object for example, used to indicate that the model is trained for a certain UE, UE group (group), or all UEs; or, used to indicate that the model is suitable for a certain UE, UE group, or all UEs;
  • Model filter information including S-NSSAI, DNN, area of interest, etc.
  • the model requirement information includes at least one of the following:
  • the model consuming network element Based on some kind of AI training framework;
  • the description method information on which the model is based uses this information to indicate that the requested model must be based on a certain language or format; that is, the model consuming network element indicates the requested model in order to understand or run the model.
  • the model must be based on some model description method or format, such as ONNX.
  • optimization algorithm information on which the model is based.
  • the model consuming network element uses this information to specify that the requested model must be based on a certain optimization algorithm.
  • the model consuming network element uses this information to specify the accuracy requirements of the output results that the requested model must be able to achieve.
  • Model storage space information The model consuming network element uses this information to specify the storage space size requirement occupied by the requested model itself or the total storage space size requirement occupied during running the model. For example, the model itself is less than 5M or the total storage space during running the model is less than 10M.
  • Model computing power requirement information The model consuming network element uses this information to indicate requesting a model that meets the computing power requirement. For example, when running this model, the computing power requirement is not higher than 500 FLOPS.
  • Model version requirement information version requirements corresponding to model instances.
  • Model requirements information also includes one or more of the following:
  • Model function or type (model type) information.
  • the model informant uses this information to indicate the function or role of the requested model, such as for image recognition, business classification, etc.;
  • Model ID Model identification (model ID) information.
  • the model messager uses the model ID to refer to the request for a specific model instance.
  • Vendor requirement information used to indicate the vendor requirements of the MTLF or NWDAF (containing MLTF) that generates the model.
  • the first model request message also includes other requirement information, and the other requirement information includes at least one of the following:
  • Analytics ID this information is used to indicate that the model consumer requests the model corresponding to the inference task indicated by the analytics ID, or is used to indicate that the requested model can be used as the model inference corresponding to the analytics ID.
  • Model target time information the time period for which the model is applicable
  • Model applicable object for example, used to indicate that the model is trained for a certain UE, UE group, or all UEs; or, used to indicate that the model is suitable for a certain UE, UE group, or all UEs;
  • Model filter information including S-NSSAI, DNN, area of interest, etc.
  • this embodiment of the present application also provides an information receiving method, which includes the following steps:
  • Step 501 The fifth network element receives the network element query request message sent by the first network element.
  • the network element query request message Including network element capability requirement information, the network element capability requirement information is determined by the first network element according to the model requirement information, the network element capability requirement information is used to determine the fourth network element, the fourth network element can Model information for a model that matches the model requirement information is provided.
  • the fifth network element can determine the fourth network element according to the network element capability requirement information.
  • the fourth network element is used to provide model information of a model that matches the model requirement information.
  • the fourth network element can be trained to obtain the model that matches the model requirement information. Models that require information matching.
  • the first network element may include a data storage function network element, such as an ADRF network element, and the fifth network element may include an NRF network element.
  • Step 502 The fifth network element sends a network element query response message to the first network element.
  • the network element query response message includes network element information matching the network element capability requirement information, that is, the fourth network element network element information.
  • the fifth network element receives the network element query request message sent by the first network element.
  • the network element query request message includes network element capability requirement information.
  • the network element capability requirement information is the first network element. Determined according to the model requirement information, the network element capability requirement information is used to determine a fourth network element.
  • the fourth network element can provide model information of a model that matches the model requirement information.
  • the fifth network element provides The first network element sends a network element query response message, and the network element query response message includes network element information matching the network element capability requirement information.
  • the fifth network element can determine the fourth network element based on the network element capability requirement information sent by the first network element, and send the network element information of the fourth network element to the first network element to facilitate the first network element from The fourth network element obtains the model information of the model that matches the model requirement information, so that when the model consumer requests the first network element to obtain the model, the model acquisition process can be simplified and signaling and data transmission overhead can be reduced.
  • the model requirement information includes at least one of the following:
  • the model consuming network element Based on some kind of AI training framework;
  • the description method information on which the model is based uses this information to indicate that the requested model must be based on a certain language or format; that is, the model consuming network element indicates the requested model in order to understand or run the model.
  • the model must be based on some model description method or format, such as ONNX.
  • optimization algorithm information on which the model is based.
  • the model consuming network element uses this information to specify that the requested model must be based on a certain optimization algorithm.
  • the model consuming network element uses this information to specify the accuracy requirements of the output results that the requested model must be able to achieve.
  • Model storage space information The model consuming network element uses this information to specify the storage space size requirement occupied by the requested model itself or the total storage space size requirement occupied during running the model. For example, the model itself is less than 5M or the total storage space during running the model is less than 10M.
  • Model computing power requirement information The model consuming network element uses this information to indicate requesting a model that meets the computing power requirement. For example, when running this model, the computing power requirement is not higher than 500 FLOPS.
  • Model version requirement information version requirements corresponding to model instances.
  • Model requirements information also includes one or more of the following:
  • Model function or type (model type) information.
  • the model informant uses this information to indicate the function or role of the requested model, such as for image recognition, business classification, etc.;
  • Model ID Model identification (model ID) information.
  • the model messager uses the model ID to refer to the request for a specific model instance.
  • Vendor requirement information used to indicate the vendor requirements of the MTLF or NWDAF (containing MLTF) that generates the model.
  • the first model request message also includes other requirement information, and the other requirement information includes at least one of the following:
  • Analytics ID this information is used to indicate that the model consumer requests the model corresponding to the inference task indicated by the analytics ID, or is used to indicate that the requested model can be used as the model inference corresponding to the analytics ID.
  • Model target time information the time period for which the model is applicable
  • Model applicable object for example, used to indicate that the model is trained for a certain UE, UE group, or all UEs; or, used to indicate that the model is suitable for a certain UE, UE group, or all UEs;
  • Model filter information including S-NSSAI, DNN, area of interest, etc.
  • Figure 6a shows a schematic diagram of information interaction between network elements provided by this application. As shown in Figure 6a, it includes the following steps:
  • Step 600 MTLF trains a model for a certain AI business (such as image recognition, user classification, etc.), which involves data collection, and performs model training operations based on the collection.
  • the data collection or model training process is not limited in this application.
  • Step 601 After the model training is completed, the model training function (such as MTLF) sends a model storage request message to the model platform (such as ADRF) for publishing or storing a specific AI model to the model platform.
  • the model training function such as MTLF
  • the model platform such as ADRF
  • model storage request message includes the model file (model file) or model file storage information (such as download address information), and model description information.
  • Model file or model file download address information is used to store the trained model file itself in the model platform, or to store the model file download address in the model platform.
  • model storage request message and model description information can be found in the previous records and will not be described in detail here.
  • Step 602 After receiving the model request message, the model platform (such as ADRF) stores relevant model information.
  • the model platform such as ADRF
  • model information stored in the model platform is the model file (model file) or model file storage information sent by MTLF in step 601, as well as model description information.
  • the model platform (such as ADRF) also stores other information included in the model storage request message;
  • the model platform (such as ADRF) also stores information such as the network element identifier and network element address of MTLF.
  • Step 603 The model consuming network element (such as AnLF or other MTLF) sends a model request message to ADRF to request a specific model.
  • the model request message includes model requirement information.
  • the content included in the model requirement information can be found in the previous records and will not be described in detail here.
  • the model request message also includes other requirement information.
  • the other requirement information please refer to the previous description and will not be described again here.
  • Step 604 The model platform (such as ADRF) determines one or more models based on the model requirement information and the stored model description information, where the model description information of the one or more models can meet or match the model requirement information.
  • the model platform such as ADRF
  • one or more models are determined based on the model function or type (model type) information in the model requirement information, and the model function or type (model type) information in the model description information of the one or more models corresponds thereto;
  • one or more models are determined according to the framework requirements or description method requirement information on which the model is based in the model requirement information.
  • the framework or description method information on which the model is based in the model description information of the one or more models can be Meet the framework requirements or description method requirements information on which the above model is based;
  • one or more models are determined according to the precision/accuracy requirement information in the model requirement information, and the precision/accuracy information in the model description information of the one or more models can satisfy the precision/accuracy requirement information.
  • model platform (such as ADRF) can also make joint judgments based on multiple requirement information in the model requirement information at the same time.
  • Step 605 The model platform (such as ADRF) sends a model feedback message to the model consuming network element.
  • the model feedback message includes model files or model file storage information corresponding to the determined one or more models.
  • the model feedback message includes all or part of the model description information corresponding to one or more models
  • the model feedback message includes all or part of other information corresponding to one or more models.
  • Figure 6b shows a schematic diagram of information interaction between network elements provided by this application. As shown in Figure 6b, it includes the following steps:
  • Step 611 The model consuming network element (such as AnLF or other MTLF) sends a first model request message to ADRF to request a specific model.
  • a model consuming network element such as AnLF or other MTLF
  • Step 612 The model platform (such as ADRF) determines that a suitable model is not matched according to the first model request message, and then executes the following steps 613-616.
  • the model platform such as ADRF
  • Step 613 The model platform (such as ADRF) determines the MTLF of the model to be obtained.
  • the model platform (such as ADRF) sends a query request message to the query network element (such as NRF) to obtain the MTLF that can provide the corresponding model.
  • the query request message includes NF type (such as MTLF type, NWDAF type, training UE type, etc.) information and MTLF capability requirement information, where the MTLF capability requirement information is determined based on the model requirement information in step 611.
  • NF type such as MTLF type, NWDAF type, training UE type, etc.
  • the MTLF capability requirement information is used by the NRF to determine one or more MTLFs that can meet or support the above MTLF capability requirement.
  • the MTLF capability requirement information is the network element capability requirement information.
  • the content included in the network element capability requirement information can be found in the above description and will not be described in detail here.
  • query request message also includes one or more of the following:
  • Analytics ID this information is used to indicate that the model consumer requests the model corresponding to the inference task indicated by analytics ID, or is used to indicate that the requested model can be used as the model inference corresponding to analytics ID.
  • Model target time information the time period for which the model is applicable
  • Model applicable object for example, used to indicate that the model is trained for a certain UE, UE group, or all UEs; or, used to indicate that the model is suitable for a certain UE, UE group, or all UEs;
  • Model filter information including S-NSSAI, DNN, area of interest, etc.
  • Step 614 the query network element sends a query response message to the model platform (such as ADRF), which contains the determined MTLF information.
  • the model platform such as ADRF
  • the query network element determines one or more MTLFs that meet the requirements based on one or more items in the MTLF capability requirement information.
  • steps 613 and 614 are optional steps.
  • the model platform (such as ADRF) can also determine one or more MTLFs that meet the requirements. This method requires MTLF to register its own capability information to the model platform (such as ADRF) in advance, where the registered capability information includes one or more capability information corresponding to the above-mentioned MTLF capability requirement information.
  • Step 615 The model platform (such as ADRF) sends a second model request message to the determined one or more MTLFs, which is used to request the model requested by the model consuming network element from the MTLF.
  • the model platform such as ADRF
  • the content of the second model request message may refer to the first model request message in step 611.
  • Step 616 MTLF sends a model response message to the model platform (such as ADRF), which contains the specific AI model requested.
  • the model platform such as ADRF
  • Step 617 The model platform (such as ADRF) stores relevant model information. Please refer to the record of step 602.
  • Step 618 The model platform (such as ADRF) sends a model feedback message to the model consuming network element.
  • the model feedback message includes model file information or model file storage information corresponding to the determined model or models. See the description of step 605.
  • the network deploys a unified or a small number of model platforms, enables each training function entity to publish and store the trained model on the model platform, and at the same time, also publishes the description information corresponding to the model. Go to the model platform.
  • the model consuming network element uniformly queries the model platform to obtain the model that matches the requirements.
  • This method avoids different model consuming network elements from going to different training function entities to obtain models, thereby avoiding a large number of point-to-point network element discovery operations, and also eliminating the process of repeatedly requesting to send models in the point-to-point method. Saves network signaling and data overhead.
  • Figure 7 shows a first model acquisition device provided by an embodiment of the present application, in which the first model acquisition device 700 includes:
  • the first receiving module 701 is configured to receive a model storage request message sent by the second network element.
  • the model storage request message includes model information of at least one model.
  • the model information includes at least one of a model file and model file storage information. item;
  • the storage module 702 is used to store model information of the at least one model.
  • model information also includes model description information, and the model description information includes at least one of the following:
  • the model stores spatial information
  • the first model acquisition device 700 also includes:
  • the second receiving module is configured to receive the first model request message sent by the third network element, where the first model request message includes model requirement information;
  • a first determination module configured to determine a target model according to the model requirement information, and the target model matches the model requirement information
  • the first sending module is configured to send model information of the target model to the third network element.
  • the model requirement information includes at least one of the following:
  • the model stores spatial information
  • the framework information is used to indicate the training framework on which the model is based;
  • the description method information is used to indicate model format information or model language information
  • the optimization algorithm information is used to indicate the algorithm for model convergence
  • the computing power requirement information is used to indicate the computing power required by the model when performing inference tasks.
  • the first determination module is configured to determine the model that matches the model requirement information as the target model when the first network element includes a model that matches the model requirement information. .
  • the first determination module includes:
  • a sending submodule configured to send a second model request message to the fourth network element when the first network element does not include a model that matches the model requirement information, where the second model request message includes the Model requirements information;
  • a receiving submodule configured to receive a model response message sent by the fourth network element, where the model response message includes model information of a model that matches the model requirement information;
  • a determining sub-module is used to determine a model that matches the model requirement information as the target model.
  • the first model acquisition device 700 also includes:
  • the second sending module is configured to send a network element query request message to the fifth network element, where the network element query request message includes network element capability requirement information, wherein the network element capability requirement information is the first network element based on The first model requires information to be determined;
  • a third receiving module configured for the first network element to receive a network element query response message sent by the fifth network element, where the network element query response message includes network element information matching the network element capability requirement information;
  • the second determination module is configured to determine the fourth network element according to the network element information.
  • the fourth network element can provide model information of a model that matches the model requirement information.
  • the first network element includes a data storage function network element.
  • the second network element or the fourth network element includes a model training function MTLF network element.
  • the third network element includes an analysis logic function AnLF network element.
  • the fifth network element includes a network storage function NRF network element.
  • the first model acquisition device 700 provided by the embodiment of the present application can implement each process implemented by the method embodiment in Figure 2 and achieve the same technical effect. To avoid duplication, details will not be described here.
  • FIG. 8 shows an information sending device provided by an embodiment of the present application.
  • the information sending device 800 includes:
  • Sending module 801 is configured to send a model storage request message to the first network element, where the model storage request message includes model information of at least one model, and the model information includes at least one of a model file and model file storage information.
  • model information also includes model description information, and the model description information includes at least one of the following:
  • the model stores spatial information
  • the information sending device 800 also includes an acquisition module, configured to train the at least one model and obtain model information of the at least one model;
  • the information sending device 800 further includes a receiving module for receiving the second model sent by the first network element.
  • request message the second model request message includes model requirement information
  • the at least one model is a model that matches the model requirement information.
  • the model requirement information includes at least one of the following:
  • the model stores spatial information
  • the framework information is used to indicate the training framework on which the model is based;
  • the description method information is used to indicate model format information or model language information
  • the optimization algorithm information is used to indicate the algorithm for model convergence
  • the computing power requirement information is used to indicate the computing power required by the model when performing inference tasks.
  • the first network element includes a data storage function network element.
  • the second network element includes a model training function MTLF network element.
  • the information sending device 800 provided by the embodiment of the present application can implement each process implemented by the method embodiment in Figure 3 and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • FIG. 9 shows a second model acquisition device provided by an embodiment of the present application.
  • the second model acquisition device 900 includes:
  • Sending module 901 configured to send a first model request message to the first network element, where the first model request message includes model requirement information
  • Receiving module 902 configured to receive model information of a target model sent by the first network element, where the target model matches the model requirement information, and the model information includes at least one of a model file and model file storage information.
  • model information also includes model description information, and the model description information includes at least one of the following:
  • the model stores spatial information
  • the model requirement information includes at least one of the following:
  • the model stores spatial information
  • the framework information is used to indicate the training framework on which the model is based;
  • the description method information is used to indicate model format information or model language information
  • the optimization algorithm information is used to indicate the algorithm for model convergence
  • the computing power requirement information is used to indicate the computing power required by the model when performing inference tasks.
  • the first network element includes a data storage function network element.
  • the third network element includes an analysis logic function AnLF network element.
  • the second model acquisition device 900 provided by the embodiment of the present application can implement each process implemented by the method embodiment in Figure 4 and achieve the same technical effect. To avoid duplication, details will not be described here.
  • FIG. 10 shows an information receiving device provided by an embodiment of the present application.
  • the information receiving device 1000 includes:
  • the receiving module 1001 is configured to receive a network element query request message sent by the first network element.
  • the network element query request message includes network element capability requirement information.
  • the network element capability requirement information is the first network element according to the model requirements. If the information is determined, the network element capability requirement information is used to determine a fourth network element, and the fourth network element can provide model information of a model that matches the model requirement information;
  • the sending module 1002 is configured to send a network element query response message to the first network element, where the network element query response message includes network element information matching the network element capability requirement information.
  • the model requirement information includes at least one of the following:
  • the model stores spatial information
  • the framework information is used to indicate the training framework on which the model is based;
  • the description method information is used to indicate model format information or model language information
  • the optimization algorithm information is used to indicate the algorithm for model convergence
  • the computing power requirement information is used to indicate the computing power required by the model when performing inference tasks.
  • the first network element includes a data storage function network element.
  • the fifth network element includes a network storage function NRF network element.
  • the information receiving device 1000 provided by the embodiment of the present application can implement each process implemented by the method embodiment in Figure 5 and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • the device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip.
  • the electronic device may be a terminal or other devices other than the terminal.
  • terminals may include but are not limited to the types of terminals 11 listed above, and other devices may be servers, network attached storage (Network Attached Storage, NAS), etc., which are not specifically limited in the embodiment of this application.
  • NAS Network Attached Storage
  • this embodiment of the present application also provides a communication device 1100, which includes a processor 1101 and a memory 1102.
  • the memory 1102 stores programs or instructions that can be run on the processor 1101.
  • the program or instruction is executed by the processor 1101, each step of the method embodiment shown in FIG. 2, FIG. 3, FIG. 4 or FIG. 5 is implemented, and the same technical effect can be achieved. To avoid repetition, the details will not be described here.
  • Embodiments of the present application also provide a first network element, including a processor and a communication interface.
  • the communication interface is used to receive a model storage request message sent by the second network element.
  • the model storage request message includes a model of at least one model.
  • Information, the model information includes at least one of a model file and model file storage information; the processor is configured to store model information of the at least one model.
  • the embodiment of the first network element corresponds to the above-mentioned method embodiment shown in Figure 2.
  • Each implementation process and implementation method of the above-mentioned method embodiment can be applied to the embodiment of the first network element, and can achieve the same technical effect. .
  • An embodiment of the present application also provides a second network element, including a processor and a communication interface.
  • the communication interface is used to send a model storage request message to the first network element.
  • the model storage request message includes model information of at least one model.
  • the model information includes at least one of a model file and model file storage information.
  • the embodiment of the second network element corresponds to the above-mentioned method embodiment shown in Figure 3. Each implementation process and implementation method of the above-mentioned method embodiment can be applied to the embodiment of the second network element, and can achieve the same technical effect. .
  • Embodiments of the present application also provide a third network element, including a processor and a communication interface.
  • the communication interface is used to send a first model request message to the first network element, where the first model request message includes model requirement information; Receive model information of a target model sent by the first network element, where the target model matches the model requirement information, and the model information includes at least one of a model file and model file storage information.
  • the embodiment of the third network element corresponds to the above-mentioned method embodiment shown in Figure 4. Each implementation process and implementation manner of the above-mentioned method embodiment can be applied to the embodiment of the third network element. , and can achieve the same technical effect.
  • Embodiments of the present application also provide a fifth network element, including a processor and a communication interface.
  • the communication interface is used to receive a network element query request message sent by the first network element.
  • the network element query request message includes network element capabilities.
  • Requirement information the network element capability requirement information is determined by the first network element according to the model requirement information, the network element capability requirement information is used to determine the fourth network element, the fourth network element can provide the Model information of the model whose model requirement information matches; sending a network element query response message to the first network element, where the network element query response message includes network element information that matches the network element capability requirement information.
  • the embodiment of the fifth network element corresponds to the above-mentioned method embodiment shown in Figure 5. Each implementation process and implementation method of the above-mentioned method embodiment can be applied to the embodiment of the fifth network element, and can achieve the same technical effect. .
  • the embodiment of the present application also provides a network side device.
  • the network side device 1200 includes: an antenna 121 , a radio frequency device 122 , a baseband device 123 , a processor 124 and a memory 125 .
  • the antenna 121 is connected to the radio frequency device 122 .
  • the radio frequency device 122 receives information through the antenna 121 and sends the received information to the baseband device 123 for processing.
  • the baseband device 123 processes the information to be sent and sends it to the radio frequency device 122.
  • the radio frequency device 122 processes the received information and then sends it out through the antenna 121.
  • the method performed by the network side device in the above embodiment can be implemented in the baseband device 123, which includes a baseband processor.
  • the baseband device 123 may include, for example, at least one baseband board on which multiple chips are disposed, as shown in FIG. Program to perform the network device operations shown in the above method embodiments.
  • the network side device may also include a network interface 126, which is, for example, a common public radio interface (CPRI).
  • a network interface 126 which is, for example, a common public radio interface (CPRI).
  • CPRI common public radio interface
  • the network side device 1200 in this embodiment of the present invention also includes: instructions or programs stored in the memory 125 and executable on the processor 124.
  • the processor 124 calls the instructions or programs in the memory 125 to execute Figures 7 and 8 , the method of executing each module shown in Figure 9 or Figure 10 and achieving the same technical effect will not be repeated here to avoid repetition.
  • Embodiments of the present application also provide a readable storage medium, with programs or instructions stored on the readable storage medium.
  • programs or instructions When the programs or instructions are executed by the processor, the above-mentioned shown in Figure 2, Figure 3, Figure 4 or Figure 5 can be implemented.
  • Each process of the method embodiment can achieve the same technical effect, so to avoid repetition, it will not be described again here.
  • the processor is the processor in the terminal described in the above embodiment.
  • the readable storage medium includes computer readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disk or optical disk, etc.
  • An embodiment of the present application further provides a chip.
  • the chip includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement the above-mentioned Figures 2, 3,
  • Each process of the method embodiment shown in Figure 4 or Figure 5 can achieve the same technical effect. To avoid repetition, it will not be described again here.
  • chips mentioned in the embodiments of this application may also be called system-on-chip, system-on-a-chip, system-on-chip or system-on-chip, etc.
  • Embodiments of the present application further provide a computer program/program product.
  • the computer program/program product is stored in a storage medium.
  • the computer program/program product is executed by at least one processor to implement the above-mentioned Figures 2 and 3. , each process of the method embodiment shown in Figure 4 or Figure 5, and can achieve the same technical effect, so to avoid repetition, they will not be described again here.
  • the embodiment of the present application also provides a communication system, including: a first network element, a second network element, a third network element, a fourth network element and a fifth network element.
  • the first network element can be used to perform the execution as shown in the figure above.
  • the second network element can be used to perform the steps of the method embodiment shown in Figure 3.
  • the third network element can be used to perform the steps of the method embodiment shown in Figure 4.
  • the fourth network element can be used to perform the steps of the method embodiment shown in Figure 2 above
  • the fifth network element can be used to perform the steps of the method embodiment shown in Figure 5 above.
  • the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation.
  • the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to the existing technology.
  • the computer software product is stored in a storage medium (such as ROM/RAM, disk , CD), including several instructions to cause a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of this application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Library & Information Science (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请公开了一种模型获取方法、信息发送方法、信息接收方法、装置及网元，属于通信技术领域，本申请实施例的模型获取方法包括：第一网元接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；所述第一网元存储所述至少一个模型的模型信息。

Description

模型获取方法、信息发送方法、信息接收方法、装置及网元
相关申请的交叉引用
本申请主张在2022年3月28日在中国提交的中国专利申请No.202210317181.5的优先权，其全部内容通过引用包含于此。
技术领域
本申请属于通信技术领域，具体涉及一种模型获取方法、信息发送方法、信息接收方法、装置及网元。
背景技术
通信网络中的训练功能实体可用于进行模型训练，并将训练好的模型提供给模型消费者使用。由于各训练功能实体分散部署，产生的模型分布在不同的训练功能实体上。模型消费者在请求获取模型操作之前，需要先借助专有的网元设备负责做训练功能和模型发现操作。
如果存在多个模型消费者需要从同一个训练功能实体中获取同种模型，则需要多次查询网元设备，还需要多次去同一个训练功能实体请求模型，信令和数据传输的开销比较大。
发明内容
本申请实施例提供一种模型获取方法、信息发送方法、信息接收方法、装置及网元，能够解决多个模型消费者在获取模型时，信令和数据传输开销较大的问题。
第一方面，提供了一种模型获取方法，该方法包括：
第一网元接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；
所述第一网元存储所述至少一个模型的模型信息。
第二方面，提供了一种信息发送方法，所述方法包括：
第二网元向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
第三方面，提供了一种模型获取方法，包括：
第三网元向第一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信息；
所述第三网元接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模 型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
第四方面，提供了一种信息接收方法，包括：
第五网元接收第一网元发送的网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息；
所述第五网元向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息。
第五方面，提供了一种模型获取装置，包括：
接收模块，用于接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；
存储模块，用于存储所述至少一个模型的模型信息。
第六方面，提供了一种信息发送装置，包括：
发送模块，用于向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
第七方面，提供了一种模型获取装置，包括：
发送模块，用于向第一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信息；
接收模块，用于接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
第八方面，提供了一种信息接收装置，包括：
接收模块，用于接收第一网元发送的网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息；
发送模块，用于向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息。
第九方面，提供了一种第一网元，包括处理器和通信接口，所述通信接口用于接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；所述处理器用于存储所述至少一个模型的模型信息。
第十方面，提供了一种第二网元，包括处理器和通信接口，所述通信接口用于向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
第十一方面，提供了一种第三网元，包括处理器和通信接口，所述通信接口用于向第 一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信息；接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
第十二方面，提供了一种第五网元，包括处理器和通信接口，所述通信接口用于接收第一网元发送的网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息；向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息。
第十三方面，提供了一种通信系统，包括：第一网元、第二网元、第三网元、第四网元和第五网元，所述第一网元用于执行第一方面所述的方法，所述第二网元用于执行第二方面所述的方法，所述第三网元用于执行第三方面所述的方法，所述第四网元用于执行第二方面所述的方法，所述第五网元用于执行第四方面所述的方法。
第十四方面，提供了一种可读存储介质，所述可读存储介质上存储程序或指令，所述程序或指令被处理器执行时实现如第一方面、第二方面、第三方面或第四方面所述的方法的步骤。
第十五方面，提供了一种芯片，所述芯片包括处理器和通信接口，所述通信接口和所述处理器耦合，所述处理器用于运行程序或指令，实现如第一方面、第二方面、第三方面或第四方面所述的方法的步骤。
第十六方面，提供了一种计算机程序/程序产品，所述计算机程序/程序产品被存储在存储介质中，所述计算机程序/程序产品被至少一个处理器执行以实现如第一方面、第二方面、第三方面或第四方面所述的方法的步骤。
在本申请实施例中，第一网元接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；所述第一网元存储所述至少一个模型的模型信息。通过上述方式，第一网元可将至少一个模型的模型信息存储于本地，模型消费者向第一网元请求获取模型即可，可简化模型获取过程，减少信令和数据传输开销。
附图说明
图1是本申请实施例可应用的一种无线通信系统的框图；
图2是本申请实施例提供的模型获取方法的流程图之一；
图3是本申请实施例提供的信息发送方法的流程图；
图4是本申请实施例提供的模型获取方法的流程图之二；
图5是本申请实施例提供的信息接收方法的流程图；
图6a是本申请实施例提供的网元之间的信息交互示意图之一；
图6b是本申请实施例提供的网元之间的信息交互示意图之二；
图7是本申请实施例提供的模型获取装置的结构图之一；
图8是本申请实施例提供的信息发送装置的结构图；
图9是本申请实施例提供的模型获取装置的结构图之二；
图10是本申请实施例提供的信息接收装置的结构图；
图11是本申请实施例提供的通信设备的结构图；
图12是本申请实施例提供的网络侧设备的结构图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员所获得的所有其他实施例，都属于本申请保护的范围。
本申请的说明书和权利要求书中的术语“第一”、“第二”等是用于区别类似的对象，而不用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换，以便本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施，且“第一”、“第二”所区别的对象通常为一类，并不限定对象的个数，例如第一对象可以是一个，也可以是多个。此外，说明书以及权利要求中“和/或”表示所连接对象的至少其中之一，字符“/”一般表示前后关联对象是一种“或”的关系。
值得指出的是，本申请实施例所描述的技术不限于长期演进型(Long Term Evolution，LTE)/LTE的演进(LTE-Advanced，LTE-A)系统，还可用于其他无线通信系统，诸如码分多址(Code Division Multiple Access，CDMA)、时分多址(Time Division Multiple Access，TDMA)、频分多址(Frequency Division Multiple Access，FDMA)、正交频分多址(Orthogonal Frequency Division Multiple Access，OFDMA)、单载波频分多址(Single-carrier Frequency Division Multiple Access，SC-FDMA)和其他系统。本申请实施例中的术语“系统”和“网络”常被可互换地使用，所描述的技术既可用于以上提及的系统和无线电技术，也可用于其他系统和无线电技术。以下描述出于示例目的描述了新空口(New Radio，NR)系统，并且在以下大部分描述中使用NR术语，但是这些技术也可应用于NR系统应用以外的应用，如第6代(6th Generation，6G)通信系统。
图1示出本申请实施例可应用的一种无线通信系统的框图。无线通信系统包括终端11和网络侧设备12。其中，终端11可以是手机、平板电脑(Tablet Personal Computer)、膝上型电脑(Laptop Computer)或称为笔记本电脑、个人数字助理(Personal Digital Assistant，PDA)、掌上电脑、上网本、超级移动个人计算机(ultra-mobile personal computer，UMPC)、移动上网装置(Mobile Internet Device，MID)、增强现实(augmented reality，AR)/虚拟现实(virtual reality，VR)设备、机器人、可穿戴式设备(Wearable Device)、车载设备(Vehicle User Equipment，VUE)、行人终端(Pedestrian User Equipment，PUE)、智能家居(具有无 线通信功能的家居设备，如冰箱、电视、洗衣机或者家具等)、游戏机、个人计算机(personal computer，PC)、柜员机或者自助机等终端侧设备，可穿戴式设备包括：智能手表、智能手环、智能耳机、智能眼镜、智能首饰(智能手镯、智能手链、智能戒指、智能项链、智能脚镯、智能脚链等)、智能腕带、智能服装等。需要说明的是，在本申请实施例并不限定终端11的具体类型。网络侧设备12可以包括接入网设备或核心网设备，其中，接入网设备也可以称为无线接入网设备、无线接入网(Radio Access Network,RAN)、无线接入网功能或无线接入网单元。接入网设备可以包括基站、无线局域网(Wireless Local Area Network，WLAN)接入点或WiFi节点等，基站可被称为节点B、演进节点B(eNB)、接入点、基收发机站(Base Transceiver Station，BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set，BSS)、扩展服务集(Extended Service Set，ESS)、家用B节点、家用演进型B节点、发送接收点(Transmitting Receiving Point，TRP)或所述领域中其他某个合适的术语，只要达到相同的技术效果，所述基站不限于特定技术词汇，需要说明的是，在本申请实施例中仅以NR系统中的基站为例进行介绍，并不限定基站的具体类型。
下面结合附图，通过一些实施例及其应用场景对本申请实施例提供的方法进行详细地说明。
如图2所示，本申请实施例提供了一种模型获取方法，包括如下步骤：
步骤201、第一网元接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
模型文件(model file)用于存储与模型相关的信息，例如，模型的结构，模型的参数等等。模型文件存储信息可以是模型文件所存储的网元信息(网元标识，网元地址等)或模型文件下载地址信息。
所述第一网元可以包括数据存储功能网元，如分析数据存储功能(Analytics Data Repository Function，ADRF)网元，所述第二网元可以包括模型训练功能(Model Training Logical Function，MTLF)网元或模型训练功能的网络数据分析功能(Network Data Analysis Function，NWDAF)网元，该NWDAF包含MTLF(containing MLTF)。
至少一个模型可以是第二网元训练获得的模型，也可以是其他网元训练获得的模型，此种情况下，其他网元将至少一个模型的模型信息发送给第二网元，第二网元再发送给第一网元进行存储。
步骤202、所述第一网元存储所述至少一个模型的模型信息。
第一网元可将模型文件存储于本地，在其他网元请求模型时，可以将请求的模型的模型文件发送给其他网元；第一网元也可以将模型文件存储信息存储于本地，在其他网元请求模型时，可以将请求的模型的模型文件存储信息发送给其他网元，其他网元再根据存储信息获取模型文件。
本实施例中，第一网元接收第二网元发送的模型存储请求消息，所述模型存储请求消 息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；所述第一网元存储所述至少一个模型的模型信息。通过上述方式，第一网元可将至少一个模型的模型信息存储于本地，模型消费者向第一网元请求获取模型即可，可简化模型获取过程，减少信令和数据传输开销。
上述中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
(1)模型所基于的框架信息(framework)，用于指示模型训练所基于的人工智能(Artificial Intelligence，AI)训练框架(training framework)；如常见的模型框架有TensorFlow，Pytorch，Caffe2等。
(2)模型所基于的描述方法信息，模型描述方法信息包括模型格式信息或模型语言信息，用于指示将所训练的模型以所述模型格式或所述模型语言来表示；所述模型格式信息或模型语言信息包括开放神经网络交换(Open Neural Network Exchange，ONNX)语言。描述方法信息用于指示模型在不同功能实体之间互通(如分析逻辑功能(Analytics Logical Function，AnLF)与MTLF，或者两个MTLF之间)需要基于的平台框架或语言。
(3)模型基于的优化算法信息；该信息用于指示该模型训练过程中用于模型收敛的算法，如梯度下降(Gradient Descent)法、随机梯度下降(Stochastic Gradient Descent，SGD)法、小批量梯度下降法(mini-batch gradient descent)、动量法(Momentum)等。
(4)模型可达到的精度信息(也可称为准确度信息)；该信息用于指示该训练完成的模型可达到的输出结果的准确程度，具体地指示模型在训练阶段或测试阶段所呈现的模型输出结果的准确程度。例如，MTLF可以设置一个验证数据集用于评估模型的准确度，该验证数据集中包括所述模型的输入数据和对应的标签数据，MTLF将输入数据输入训练后的模型得到输出数据，再比较输出数据与标签数据是否一致，进而根据相应算法(如结果一致的占比值)得到模型的准确度。
(5)模型存储空间信息，包括模型本身占用存储空间大小或运行该模型过程中需占用总存储空间大小。
(6)模型算力要求信息，用于指示运行该模型进行推理任务时所需要的算力要求，如“每秒浮点运算次数”，“每秒峰值速度”(即Floating-point operations per second，FLOPS)。
(7)模型版本信息，模型实例对应的版本值。
另外，所述模型描述信息还可包括如下至少一项：
(1)模型功能信息；模型功能或类型(model type)信息，该信息用于指示该存储的模型的功能或作用，如用于图像识别，业务分类等；
(2)模型标识信息，model ID，该信息用于作为索引(index)指代模型。
(3)模型输入数据类型信息。
(4)模型输出数据类型信息。
(5)厂商信息，用于指示训练模型的MTLF或NWDAF(containing MLTF)所属厂商。
可选地，模型存储请求消息中还包括其他信息，所述其他信息包括以下一项或多项：
(1)分析标识(Analytics ID)，该信息用于指示该模型实例与analytics ID所示推理任务对应，或者用于指示该模型可以用作analytics ID对应的模型推理。
(2)模型目标时间信息，模型适用的时间段；
(3)模型适用对象，如用于指示模型是针对某UE、UE group或者所有UE训练的；或者，用于指示模型适用于某UE、UE group或者所有UE；
(4)模型过滤器(Model filter)信息：包括单一网络切片选择辅助信息(Single Network Slice Selection Assistance Information，S-NSSAI)、数据网络名(Data Network Name，DNN)、感兴趣区域(area of interest)等。
第一网元对其他信息进行存储，进一步的，第一网元还可存储第二网元的网元标识、网元地址等信息。
在本申请一种实施例中，所述方法还包括：
所述第一网元接收第三网元发送的第一模型请求消息，所述第一模型请求消息包括模型要求信息；
所述第一网元根据所述模型要求信息确定目标模型，所述目标模型与所述模型要求信息匹配；
所述第一网元向所述第三网元发送所述目标模型的模型信息。
具体的，第三网元可为模型消费网元，例如分析逻辑功能(Analytics Logical Function，AnLF)网元或者NWDAF(containing AnLF)。第三网元向第一网元发送第一模型请求消息，用于请求与模型要求信息匹配的模型。
所述模型要求信息包括如下至少一项：
(1)模型所基于的框架信息；模型消费网元利用该信息指示所请求模型必须基于某平台框架；也就是说，模型消费网元为了可理解或可运行该模型而指示所请求的模型必须基于某种AI training framework；
(2)模型所基于的描述方法信息，模型消费网元利用该信息指示所请求模型必须基于某语言或格式；也就是说，模型消费网元为了可理解或可运行该模型而指示所请求的模型必须基于某种模型描述方法或格式，如ONNX。
(3)模型基于的优化算法信息，模型消费网元利用该信息指定所请求的模型必须基于某种优化算法。
(4)模型可达到的精度信息，模型消费网元利用该信息指定所请求的模型必须能达到的输出结果准确度要求。
(5)模型存储空间信息，模型消费网元利用该信息指定所请求的模型本身占用存储空间大小要求或运行该模型过程中需占用总存储空间大小要求。如，模型本身小于5M或者运行该模型过程中总存储空间小于10M。
(6)模型算力要求信息，模型消费网元利用该信息指示请求满足该算力要求的模型。 如运行该模型时算力要求不高于500每秒所执行的浮点运算次数(Floating-point Operations Per Second，FLOPS)。
(7)模型版本要求信息，模型实例对应的版本要求。
模型要求信息还包括以下一项或多项：
(1)模型功能或类型(model type)信息，模型消息者利用该信息指示所请求模型的功能或作用，如用于图像识别，业务分类等；
(2)模型标识(model ID)信息，模型消息者利用model ID指代请求某具体模型实例。
(3)厂商要求信息，用于指示产生模型的MTLF或NWDAF(containing MLTF)所属厂商要求。
可选地，第一模型请求消息还包括其他要求信息，所述其他要求信息包括以下至少一项：
(1)Analytics ID，该信息用于指示模型消费者请求与analytics ID所示推理任务对应的模型，或者用于指示所请求的模型可以用作analytics ID对应的模型推理。
(2)Model目标时间信息，模型适用的时间段；
(3)模型适用对象，如用于指示模型是针对某UE、UE group或者所有UE训练的；或者，用于指示模型需适用于某UE、UE group或者所有UE；
(4)Model filter信息：包括S-NSSAI、DNN、area of interest等。
第一网元根据模型要求信息和所存储的模型描述信息，确定目标模型，目标模型可包括一个或多个模型，这一个或多个模型(model)的模型描述信息可以满足或匹配模型要求信息。
例如，根据模型要求信息中的模型功能或类型(model type)信息，确定一个或多个model，该一个或多个model的模型描述信息中的模型功能或类型(model type)信息与之对应；
又如，根据模型要求信息中的模型所基于的框架要求或描述方法要求信息，确定一个或多个model，该一个或多个model的模型描述信息中的模型所基于的框架或描述方法信息可满足上述模型所基于的框架要求或描述方法要求信息；
又如，根据模型要求信息中的精度/准确度要求信息，确定一个或多个model，该一个或多个model的模型描述信息中的精度/准确度信息可满足所述精度/准确度要求信息。以此类推。
所述目标模型的模型描述信息与所述模型要求信息一一匹配，例如，目标模型所基于的框架信息与模型要求信息中的模型所基于的框架信息相同，和/或，模型描述信息中模型可达到的精度信息与模型要求信息中的精度要求信息相同，和/或，模型描述信息中模型存储空间信息与模型要求信息中的存储空间要求信息相同等等。
所述第一网元在确定目标模型后，向所述第三网元发送模型反馈消息，模型反馈消息 携带所述目标模型的模型信息。
可选地，所述模型反馈消息包括一个或多个模型所对应的模型描述信息的全部或部分信息；
可选地，所述模型反馈消息还包括一个或多个模型所对应的其他信息的全部或部分信息。
第一网元根据模型要求信息与本地存储的模型描述信息相匹配，若匹配成功，即在所述第一网元包括与所述模型要求信息匹配的模型的情况下，所述第一网元将与所述模型要求信息匹配的模型确定为所述目标模型。如匹配不成功，第一网元向第二网元发送响应消息，响应消息用于指示模型请求失败。可选地，响应消息可携带模型请求失败的原因值是模型不存在。
另一种情况下，若匹配不成功，即在所述第一网元不包括与所述模型要求信息匹配的模型的情况下，先获取与所述模型要求信息匹配的模型，具体为：
所述第一网元向第四网元发送第二模型请求消息，所述第二模型请求消息包括所述模型要求信息；所述第一网元接收所述第四网元发送的模型响应消息，所述模型响应消息包括与所述模型要求信息相匹配的模型的模型信息；所述第一网元将与所述模型要求信息相匹配的模型确定为所述目标模型。
第四网元可为MTLF或NWDAF(containing MTLF)网元。第四网元可有多个，第二网元和第四网元可以为不同网元。
所述第一网元根据所述第一模型要求信息确定网元能力要求信息，然后将网元能力要求信息与预先获取的网元能力信息进行匹配，确定第四网元，或者，第一网元通过向第五网元进行查询，确定第四网元，具体过程为：
所述第一网元向第五网元发送网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，其中，所述网元能力要求信息是所述第一网元根据所述第一模型要求信息确定的；
所述第一网元接收所述第五网元发送的网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息；
所述第一网元根据所述网元信息，确定所述第四网元，所述第四网元能够提供与所述模型要求信息相匹配的模型的模型信息。
上述中，第五网元包括网络存储功能(Network Repository Function，NRF)网元。
网元能力要求信息包括以下至少一项：
(1)模型功能或类型(model type)信息。
(2)模型标识信息，model ID。
(3)模型所基于的框架要求或描述方法信息。
(4)模型基于的优化算法信息。
(5)精度/准确度信息。
(6)存储空间信息。
(7)模型算力信息。
(8)厂商要求信息。
(9)版本要求信息。
上述实施例中的模型获取方法中，通过第一网元，使第二网元可将训练好的模型发布和存储到第一网元上，同时，还将模型对应的描述信息也发布到第一网元上。模型消费网元统一从第一网元查询获取匹配要求的模型。避免了不同的模型消费网元分散地去从不同训练功能实体中获取模型，从而避免了大量的点对点的网元发现操作，也可省去点对点方式中存在的重复请求发送模型的过程，节省了网络信令和数据开销。
如图3所示，本申请实施例还提供一种信息发送方法，包括如下步骤：
步骤301、第二网元向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
模型文件(model file)用于存储与模型相关的信息，例如，模型的结构，模型的参数等等。模型文件存储信息可以是模型文件所存储的网元信息(网元标识，网元地址等)或模型文件下载地址信息。
所述第一网元可以包括数据存储功能网元，如ADRF网元，所述第二网元可以包括MTLF网元或NWDAF(containing MTLF)网元。
至少一个模型可以是第二网元训练获得的模型，也可以是其他网元训练获得的模型，此种情况下，其他网元将至少一个模型的模型信息发送给第二网元，第二网元再发送给第一网元进行存储。
本实施例中，第二网元向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。通过上述方式，第二网元将获取到的至少一个模型的模型信息发送给第一网元进行存储，模型消费者向第一网元请求获取模型即可，可简化模型获取过程，减少信令和数据传输开销。
上述中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
(1)模型所基于的框架信息(framework)用于指示模型训练所基于的AI training framework；如常见的模型框架有TensorFlow，Pytorch，Caffe2等。
(2)模型所基于的描述方法信息，模型描述方法信息包括模型格式信息或模型语言信息，用于指示将所训练的模型以所述模型格式或所述模型语言来表示；所述模型格式信息或模型语言信息包括ONNX语言。描述方法信息用于指示模型在不同功能实体之间互通(如AnLF与MTLF，或者两个MTLF之间)需要基于的平台框架或语言。
(3)模型基于的优化算法信息；该优化算法信息用于指示该模型训练过程中用于模型收敛的算法，如梯度下降(Gradient Descent)法、随机梯度下降(Stochastic Gradient Descent，SGD)法、小批量梯度下降法(mini-batch gradient descent)、动量法(Momentum) 等。
(4)模型可达到的精度信息(也可称为准确度信息)；该精度信息用于指示该训练完成的模型可达到的输出结果的准确程度，具体地指示模型在训练阶段或测试阶段所呈现的模型输出结果的准确程度。例如，MTLF可以设置一个验证数据集用于评估模型的准确度，该验证数据集中包括所述模型的输入数据和对应的标签数据，MTLF将输入数据输入训练后的模型得到输出数据，再比较输出数据与标签数据是否一致，进而根据相应算法(如结果一致的占比值)得到模型的准确度。
(5)模型存储空间信息，包括模型本身占用存储空间大小或运行该模型过程中需占用总存储空间大小。
(6)模型算力要求信息，用于指示运行该模型进行推理任务时所需要的算力要求，如FLOPS(即“每秒浮点运算次数”，“每秒峰值速度”)。
(7)模型版本信息，模型实例对应的版本值。
另外，所述模型描述信息还可包括如下至少一项：
(1)模型功能信息；模型功能或类型(model type)信息，该信息用于指示该存储的模型的功能或作用，如用于图像识别，业务分类等；
(2)模型标识信息，model ID，该信息用于作为索引(index)指代模型。
(3)模型输入数据类型信息。
(4)模型输出数据类型信息。
(5)厂商信息，用于指示训练模型的MTLF或NWDAF(containing MLTF)所属厂商。
可选地，模型存储请求消息中还包括其他信息，所述其他信息包括以下一项或多项：
(1)分析标识(Analytics ID)，该信息用于指示该模型实例与analytics ID所示推理任务对应，或者用于指示该模型可以用作analytics ID对应的模型推理。
(2)模型目标时间信息，模型适用的时间段；
(3)模型适用对象，如用于指示模型是针对某UE、UE group或者所有UE训练的；或者，用于指示模型适用于某UE、UE group或者所有UE；
(4)模型过滤器(Model filter)信息：包括S-NSSAI、DNN、area of interest等。
在本申请一种实施例中，在所述第二网元向第一网元发送模型存储请求消息之前，所述方法还包括：
所述第二网元对所述至少一个模型进行训练，获得所述至少一个模型的模型信息，此种情况下，第二网元训练获得所述至少一个模型；
或者，
所述第二网元接收所述至少一个模型的模型信息，此种情况下，第二网元从其他网元获取到至少一个模型的模型信息。
在本申请一种实施例中，在所述第二网元向第一网元发送模型存储请求消息之前，所 述方法还包括：
所述第二网元接收所述第一网元发送的第二模型请求消息，所述第二模型请求消息包括模型要求信息，所述至少一个模型是与所述模型要求信息相匹配的模型。
具体的，第二网元接收第二模型请求消息，第二网元根据模型要求信息确定至少一个模型，并将至少一个模型携带在所述模型存储请求消息中发送给第一网元。
所述模型要求信息包括如下至少一项：
(1)模型所基于的框架信息；模型消费网元利用该信息指示所请求模型必须基于某平台框架；也就是说，模型消费网元为了可理解或可运行该模型而指示所请求的模型必须基于某种AI training framework；
(2)模型所基于的描述方法信息，模型消费网元利用该信息指示所请求模型必须基于某语言或格式；也就是说，模型消费网元为了可理解或可运行该模型而指示所请求的模型必须基于某种模型描述方法或格式，如ONNX。
(3)模型基于的优化算法信息，模型消费网元利用该信息指定所请求的模型必须基于某种优化算法。
(4)模型可达到的精度信息，模型消费网元利用该信息指定所请求的模型必须能达到的输出结果准确度要求。
(5)模型存储空间信息，模型消费网元利用该信息指定所请求的模型本身占用存储空间大小要求或运行该模型过程中需占用总存储空间大小要求。如，模型本身小于5M或者运行该模型过程中总存储空间小于10M。
(6)模型算力要求信息，模型消费网元利用该信息指示请求满足该算力要求的模型。如运行该模型时算力要求不高于500FLOPS。
(7)模型版本要求信息，模型实例对应的版本要求。
模型要求信息还包括以下一项或多项：
(1)模型功能或类型(model type)信息，模型消息者利用该信息指示所请求模型的功能或作用，如用于图像识别，业务分类等；
(2)模型标识(model ID)信息，模型消息者利用model ID指代请求某具体模型实例。
(3)厂商要求信息，用于指示产生模型的MTLF或NWDAF(containing MLTF)所属厂商要求。
可选地，第一模型请求消息还包括其他要求信息，所述其他要求信息包括以下至少一项：
(1)Analytics ID，该信息用于指示模型消费者请求与analytics ID所示推理任务对应的模型，或者用于指示所请求的模型可以用作analytics ID对应的模型推理。
(2)Model目标时间信息，模型适用的时间段；
(3)模型适用对象，如用于指示模型是针对某UE、UE group或者所有UE训练的； 或者，用于指示模型适用于某UE、UE group或者所有UE；
(4)Model filter信息：包括S-NSSAI、DNN、area of interest等。
第一网元根据模型要求信息和所存储的模型描述信息，确定目标模型，目标模型可包括一个或多个模型，这一个或多个模型(model)的模型描述信息可以满足或匹配模型要求信息。
例如，根据模型要求信息中的模型功能或类型(model type)信息，确定一个或多个model，该一个或多个model的模型描述信息中的模型功能或类型(model type)信息与之对应；
又如，根据模型要求信息中的模型所基于的框架要求或描述方法要求信息，确定一个或多个model，该一个或多个model的模型描述信息中的模型所基于的框架或描述方法信息可满足上述模型所基于的框架要求或描述方法要求信息；
又如，根据模型要求信息中的精度/准确度要求信息，确定一个或多个model，该一个或多个model的模型描述信息中的精度/准确度信息可满足所述精度/准确度要求信息。以此类推。
所述至少一个模型是与所述模型要求信息相匹配的模型，可以理解为所述至少一个模型中的模型的模型描述信息与所述模型要求信息一一匹配，例如，所述至少一个模型中的模型所基于的框架信息与模型要求信息中的模型所基于的框架信息匹配，和/或，模型描述信息中模型可达到的精度信息与模型要求信息中的精度要求信息匹配，和/或，模型描述信息中模型存储空间信息与模型要求信息中的存储空间要求信息匹配等等。
如图4所示，本申请实施例还提供一种模型获取方法，包括如下步骤：
步骤401、第三网元向第一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信息。
第一模型请求消息用于请求与模型要求信息相匹配的模型的模型信息。所述第一网元可以包括数据存储功能网元，如ADRF网元，第三网元可为模型消费网元，例如AnLF网元或者NWDAF(containing AnLF)。
步骤402、第三网元接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
模型文件(model file)用于存储与模型相关的信息，例如，模型的结构，模型的参数等等。模型文件存储信息可以是模型文件所存储的网元信息(网元标识，网元地址等)或模型文件下载地址信息。
本实施例中，第三网元向第一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信息；第三网元接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。通过上述方式，第三网元可向第一网元请求与模型要求信息相匹配的模型，可简化模型获取过程，减少信令和数据传输开销。
上述中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
(1)模型所基于的框架信息(framework)，用于指示模型训练所基于的AI training framework；如常见的模型框架有TensorFlow，Pytorch，Caffe2等。
(2)模型所基于的描述方法信息，模型描述方法信息包括模型格式信息或模型语言信息，用于指示将所训练的模型以所述模型格式或所述模型语言来表示；所述模型格式信息或模型语言信息包括ONNX语言。描述方法信息用于指示模型在不同功能实体之间互通(如AnLF与MTLF，或者两个MTLF之间)需要基于的平台框架或语言。
(3)模型基于的优化算法信息；该信息用于指示该模型训练过程中用于模型收敛的算法，如梯度下降(Gradient Descent)法、随机梯度下降(Stochastic Gradient Descent，SGD)法、小批量梯度下降法(mini-batch gradient descent)、动量法(Momentum)等。
(4)模型可达到的精度信息(也可称为准确度信息)；该信息用于指示该训练完成的模型可达到的输出结果的准确程度，具体地指示模型在训练阶段或测试阶段所呈现的模型输出结果的准确程度。例如，MTLF可以设置一个验证数据集用于评估模型的准确度，该验证数据集中包括所述模型的输入数据和对应的标签数据，MTLF将输入数据输入训练后的模型得到输出数据，再比较输出数据与标签数据是否一致，进而根据相应算法(如结果一致的占比值)得到模型的准确度。
(5)模型存储空间信息，包括模型本身占用存储空间大小或运行该模型过程中需占用总存储空间大小。
(6)模型算力要求信息，用于指示运行该模型进行推理任务时所需要的算力要求，如FLOPS(即“每秒浮点运算次数”，“每秒峰值速度”)。
(7)模型版本信息，模型实例对应的版本值。
另外，所述模型描述信息还可包括如下至少一项：
(1)模型功能信息；模型功能或类型(model type)信息，该信息用于指示该存储的模型的功能或作用，如用于图像识别，业务分类等；
(2)模型标识信息，model ID，该信息用于作为索引(index)指代模型。
(3)模型输入数据类型信息。
(4)模型输出数据类型信息。
(5)厂商信息，用于指示训练模型的MTLF或NWDAF(containing MLTF)所属厂商。
可选地，模型存储请求消息中还包括其他信息，所述其他信息包括以下一项或多项：
(1)分析标识(Analytics ID)，该信息用于指示该模型实例与analytics ID所示推理任务对应，或者用于指示该模型可以用作analytics ID对应的模型推理。
(2)模型目标时间信息，模型适用的时间段；
(3)模型适用对象，如用于指示模型是针对某UE、UE组(group)或者所有UE训练的；或者，用于指示模型适用于某UE、UE group或者所有UE；
(4)模型过滤器(Model filter)信息：包括S-NSSAI、DNN、area of interest等。
所述模型要求信息包括如下至少一项：
(1)模型所基于的框架信息；模型消费网元利用该信息指示所请求模型必须基于某平台框架；也就是说，模型消费网元为了可理解或可运行该模型而指示所请求的模型必须基于某种AI training framework；
(2)模型所基于的描述方法信息，模型消费网元利用该信息指示所请求模型必须基于某语言或格式；也就是说，模型消费网元为了可理解或可运行该模型而指示所请求的模型必须基于某种模型描述方法或格式，如ONNX。
(3)模型基于的优化算法信息，模型消费网元利用该信息指定所请求的模型必须基于某种优化算法。
(4)模型可达到的精度信息，模型消费网元利用该信息指定所请求的模型必须能达到的输出结果准确度要求。
(5)模型存储空间信息，模型消费网元利用该信息指定所请求的模型本身占用存储空间大小要求或运行该模型过程中需占用总存储空间大小要求。如，模型本身小于5M或者运行该模型过程中总存储空间小于10M。
(6)模型算力要求信息，模型消费网元利用该信息指示请求满足该算力要求的模型。如运行该模型时算力要求不高于500FLOPS。
(7)模型版本要求信息，模型实例对应的版本要求。
模型要求信息还包括以下一项或多项：
(1)模型功能或类型(model type)信息，模型消息者利用该信息指示所请求模型的功能或作用，如用于图像识别，业务分类等；
(2)模型标识(model ID)信息，模型消息者利用model ID指代请求某具体模型实例。
(3)厂商要求信息，用于指示产生模型的MTLF或NWDAF(containing MLTF)所属厂商要求。
可选地，第一模型请求消息还包括其他要求信息，所述其他要求信息包括以下至少一项：
(1)Analytics ID，该信息用于指示模型消费者请求与analytics ID所示推理任务对应的模型，或者用于指示所请求的模型可以用作analytics ID对应的模型推理。
(2)Model目标时间信息，模型适用的时间段；
(3)模型适用对象，如用于指示模型是针对某UE、UE group或者所有UE训练的；或者，用于指示模型适用于某UE、UE group或者所有UE；
(4)Model filter信息：包括S-NSSAI、DNN、area of interest等。
如图5所示，本申请实施例还提供一种信息接收方法，包括如下步骤：
步骤501、第五网元接收第一网元发送的网元查询请求消息，所述网元查询请求消息 包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息。
第五网元根据网元能力要求信息可确定第四网元，第四网元用于提供与所述模型要求信息匹配的模型的模型信息，例如，第四网元可训练获得与所述模型要求信息匹配的模型。
所述第一网元可以包括数据存储功能网元，如ADRF网元，所述第五网元包括NRF网元。
步骤502、所述第五网元向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息，即第四网元的网元信息。
本实施例中，第五网元接收第一网元发送的网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息，所述第五网元向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息。通过上述方式，第五网元可基于第一网元发送的网元能力要求信息确定第四网元，并将第四网元的网元信息发送给第一网元，便于第一网元从第四网元获取与模型要求信息匹配的模型的模型信息，使得模型消费者向第一网元请求获取模型时，可简化模型获取过程，减少信令和数据传输开销。
所述模型要求信息包括如下至少一项：
(1)模型所基于的框架信息；模型消费网元利用该信息指示所请求模型必须基于某平台框架；也就是说，模型消费网元为了可理解或可运行该模型而指示所请求的模型必须基于某种AI training framework；
(2)模型所基于的描述方法信息，模型消费网元利用该信息指示所请求模型必须基于某语言或格式；也就是说，模型消费网元为了可理解或可运行该模型而指示所请求的模型必须基于某种模型描述方法或格式，如ONNX。
(3)模型基于的优化算法信息，模型消费网元利用该信息指定所请求的模型必须基于某种优化算法。
(4)模型可达到的精度信息，模型消费网元利用该信息指定所请求的模型必须能达到的输出结果准确度要求。
(5)模型存储空间信息，模型消费网元利用该信息指定所请求的模型本身占用存储空间大小要求或运行该模型过程中需占用总存储空间大小要求。如，模型本身小于5M或者运行该模型过程中总存储空间小于10M。
(6)模型算力要求信息，模型消费网元利用该信息指示请求满足该算力要求的模型。如运行该模型时算力要求不高于500FLOPS。
(7)模型版本要求信息，模型实例对应的版本要求。
模型要求信息还包括以下一项或多项：
(1)模型功能或类型(model type)信息，模型消息者利用该信息指示所请求模型的功能或作用，如用于图像识别，业务分类等；
(2)模型标识(model ID)信息，模型消息者利用model ID指代请求某具体模型实例。
(3)厂商要求信息，用于指示产生模型的MTLF或NWDAF(containing MLTF)所属厂商要求。
可选地，第一模型请求消息还包括其他要求信息，所述其他要求信息包括以下至少一项：
(1)Analytics ID，该信息用于指示模型消费者请求与analytics ID所示推理任务对应的模型，或者用于指示所请求的模型可以用作analytics ID对应的模型推理。
(2)Model目标时间信息，模型适用的时间段；
(3)模型适用对象，如用于指示模型是针对某UE、UE group或者所有UE训练的；或者，用于指示模型适用于某UE、UE group或者所有UE；
(4)Model filter信息：包括S-NSSAI、DNN、area of interest等。
以下对本申请提供的方法进行如下举例说明。
图6a所示为本申请提供的网元之间的信息交互示意图，如图6a所示，包括如下步骤：
步骤600，MTLF针对某种AI业务(如图像识别，用户分类等)训练模型，其中涉及到数据收集，并基于收集收集进行模型训练操作。数据采集或模型训练过程本申请中不做限定。
步骤601，在模型训练完成后，模型训练功能(如MTLF)向模型平台(如ADRF)发送模型存储请求消息，用于向model平台发布或存储某具体的AI模型。
具体地，模型存储请求消息中包括模型文件(model file)或模型文件存储信息(例如下载地址信息)，以及模型描述信息。模型文件(model file)或模型文件下载地址信息，用于将训练好的模型文件本身存储于模型平台，或者将模型文件下载地址存储于模型平台。
模型存储请求消息以及模型描述信息包括的内容可参见前文记载，在此不做赘述。
步骤602，model平台(如ADRF)接收模型请求消息后，存储相关的模型信息。
具体地，model平台(如ADRF)所存储的模型信息是步骤601中MTLF发送的模型文件(model file)或模型文件存储信息，以及模型描述信息。
可选地，model平台(如ADRF)还存储了模型存储请求消息中包括其他信息；
可选地，model平台(如ADRF)还存储了MTLF的网元标识、网元地址等信息。
步骤603，模型消费网元(如AnLF或其他MTLF)向ADRF发送模型请求消息，用于请求某具体的模型。其中，模型请求消息中包括模型要求信息，模型要求信息包括的内容可参见前文记载，在此不做赘述。
可选地，模型请求消息还包括其他要求信息，所述其他要求信息包括的内容可参见前文记载，在此不做赘述。
步骤604，model平台(如ADRF)基于模型要求信息和所存储的模型描述信息，确定一个或多个model，其中该一个或多个model的模型描述信息可以满足或匹配模型要求信息。
例如，根据模型要求信息中的模型功能或类型(model type)信息，确定一个或多个model，该一个或多个model的模型描述信息中的模型功能或类型(model type)信息与之对应；
又如，根据模型要求信息中的模型所基于的框架要求或描述方法要求信息，确定一个或多个model，该一个或多个model的模型描述信息中的模型所基于的框架或描述方法信息可满足上述模型所基于的框架要求或描述方法要求信息；
又如，根据模型要求信息中的精度/准确度要求信息，确定一个或多个model，该一个或多个model的模型描述信息中的精度/准确度信息可满足所述精度/准确度要求信息。
依次类推。
另外，model平台(如ADRF)也可以同时基于模型要求信息中的多个要求信息共同判断。
步骤605，model平台(如ADRF)向模型消费网元发送模型反馈消息，所述模型反馈消息中包括所确定的一个或多个model对应的模型文件(model file)或者模型文件存储信息。
可选地，所述模型反馈消息中包括一个或多个模型所对应的模型描述信息的全部或部分信息；
可选地，所述模型反馈消息中包括一个或多个模型所对应的其他信息的全部或部分信息。
图6b所示为本申请提供的网元之间的信息交互示意图，如图6b所示，包括如下步骤：
步骤611、模型消费网元(如AnLF或其他MTLF)向ADRF发送第一模型请求消息，用于请求某具体的模型。本步骤可参见步骤603。
步骤612、model平台(如ADRF)根据第一模型请求消息，确定并未匹配到合适的模型，则执行下述步骤613-616。
具体匹配方法可参见步骤604。
步骤613、model平台(如ADRF)确定待获取model的MTLF。可选地，model平台(如ADRF)向查询网元(如NRF)发送查询请求消息，用于获取可提供对应model的MTLF。
具体地，查询请求消息中包括NF type(如MTLF type，NWDAF type，training UE type等)信息和MTLF能力要求信息，其中MTLF能力要求能力信息是根据所述步骤611中的模型要求信息确定的。
所述MTLF能力要求信息用于NRF确定一个或多个MTLF，该一个或多个MTLF可以满足或支持上述MTLF能力要求。
MTLF能力要求信息即网元能力要求信息，网元能力要求信息包括的内容可参见上文记载，在此不做赘述。
另外，查询请求消息中还包括以下一项或多项：
Analytics ID，该信息用于指示模型消费者请求与analytics ID所示推理任务对应的模型，或者用于指示所请求的模型可以用作analytics ID对应的模型推理。
Model目标时间信息，模型适用的时间段；
模型适用对象，如用于指示模型是针对某UE、UE group或者所有UE训练的；或者，用于指示模型适用于某UE、UE group或者所有UE；
Model filter信息：包括S-NSSAI、DNN、area of interest等。
步骤614、可选地，查询网元向model平台(如ADRF)发送查询响应消息，其中包含所确定的MTLF信息。
具体的，查询网元根据MTLF能力要求信息中的一项或多项确定满足要求的一个或多个MTLF。
上述中，步骤613和614是可选步骤，在model平台(如ADRF)中存在MTLF能力信息的场景下，model平台(如ADRF)也可以自己确定符合要求的一个或多个MTLF。该方式要求MTLF提前将自身能力信息注册到model平台(如ADRF)，其中注册的能力信息包含上述MTLF能力要求信息对应的一种或多种能力信息。
步骤615、model平台(如ADRF)向所确定的一个或多个MTLF发送第二模型请求消息，用于向MTLF请求模型消费网元请求的模型。
第二模型请求消息内容可参考步骤611中的第一模型请求消息。
步骤616、MTLF向model平台(如ADRF)发送模型响应消息，其中包含所请求的具体的AI模型。
模型响应消息的内容具体参考模型存储请求消息的内容。
步骤617、model平台(如ADRF)存储相关的模型信息，可参见步骤602的记载。
步骤618、model平台(如ADRF)向模型消费网元发送模型反馈消息，所述模型反馈消息中包括所确定的一个或多个model对应的模型文件(model file)信息或者模型文件存储信息，可参见步骤605的记载。
上述网元之间的信息交互方法，网络部署统一的或少量的model平台，使能各训练功能实体将训练好的模型发布和存储到model平台上，同时，还将模型对应的描述信息也发布到model平台上。模型消费网元统一从model平台查询获取匹配要求的模型。
该方法避免了不同的模型消费网元分散地去不同训练功能实体中去获取模型，从而避免了大量的点对点的网元发现操作，也可省去点对点方式中存在的重复请求发送模型的过程，节省了网络信令和数据开销。
如图7所示为本申请实施例提供的一种第一模型获取装置，其中第一模型获取装置700包括：
第一接收模块701，用于接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；
存储模块702，用于存储所述至少一个模型的模型信息。
可选地，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
模型所基于的框架信息；
模型所基于的描述方法信息；
模型基于的优化算法信息；
模型可达到的精度信息；
模型存储空间信息；
模型算力要求信息；
模型版本信息。
可选地，第一模型获取装置700还包括：
第二接收模块，用于接收第三网元发送的第一模型请求消息，所述第一模型请求消息包括模型要求信息；
第一确定模块，用于根据所述模型要求信息确定目标模型，所述目标模型与所述模型要求信息匹配；
第一发送模块，用于向所述第三网元发送所述目标模型的模型信息。
可选地，所述模型要求信息包括如下至少一项：
模型所基于的框架信息；
模型所基于的描述方法信息；
模型基于的优化算法信息；
模型可达到的精度信息；
模型存储空间信息；
模型算力要求信息；
模型版本要求信息。
可选地，所述框架信息用于指示模型所基于的训练框架；
和/或，
所述描述方法信息用于指示模型格式信息或模型语言信息；
和/或，
所述优化算法信息用于指示模型收敛的算法；
和/或，
所述算力要求信息用于指示模型在进行推理任务时所需要的算力。
可选地，所述第一确定模块，用于在所述第一网元包括与所述模型要求信息匹配的模型的情况下，将与所述模型要求信息匹配的模型确定为所述目标模型。
可选地，所述第一确定模块，包括：
发送子模块，用于在所述第一网元不包括与所述模型要求信息匹配的模型的情况下，向第四网元发送第二模型请求消息，所述第二模型请求消息包括所述模型要求信息；
接收子模块，用于接收所述第四网元发送的模型响应消息，所述模型响应消息包括与所述模型要求信息相匹配的模型的模型信息；
确定子模块，用于将与所述模型要求信息相匹配的模型确定为所述目标模型。
可选地，第一模型获取装置700还包括：
第二发送模块，用于向第五网元发送网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，其中，所述网元能力要求信息是所述第一网元根据所述第一模型要求信息确定的；
第三接收模块，用于所述第一网元接收所述第五网元发送的网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息；
第二确定模块，用于根据所述网元信息，确定所述第四网元，所述第四网元能够提供与所述模型要求信息相匹配的模型的模型信息。
可选地，所述第一网元包括数据存储功能网元。
可选地，所述第二网元或所述第四网元包括模型训练功能MTLF网元。
可选地，所述第三网元包括分析逻辑功能AnLF网元。
可选地，所述第五网元包括网络存储功能NRF网元。
本申请实施例提供的第一模型获取装置700能够实现图2的方法实施例实现的各个过程，并达到相同的技术效果，为避免重复，这里不再赘述。
如图8所示为本申请实施例提供的一种信息发送装置，信息发送装置800包括：
发送模块801，用于向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
可选地，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
模型所基于的框架信息；
模型所基于的描述方法信息；
模型基于的优化算法信息；
模型可达到的精度信息；
模型存储空间信息；
模型算力要求信息；
模型版本信息。
可选地，信息发送装置800还包括获取模块，用于对所述至少一个模型进行训练，获得所述至少一个模型的模型信息；
或者，接收所述至少一个模型的模型信息。
可选地，信息发送装置800还包括接收模块，用于接收所述第一网元发送的第二模型 请求消息，所述第二模型请求消息包括模型要求信息，所述至少一个模型是与所述模型要求信息相匹配的模型。
可选地，所述模型要求信息包括如下至少一项：
模型所基于的框架信息；
模型所基于的描述方法信息；
模型基于的优化算法信息；
模型可达到的精度信息；
模型存储空间信息；
模型算力要求信息；
模型版本要求信息。
可选地，所述框架信息用于指示模型所基于的训练框架；
和/或，
所述描述方法信息用于指示模型格式信息或模型语言信息；
和/或，
所述优化算法信息用于指示模型收敛的算法；
和/或，
所述算力要求信息用于指示模型在进行推理任务时所需要的算力。
可选地，所述第一网元包括数据存储功能网元。
可选地，所述第二网元包括模型训练功能MTLF网元。
本申请实施例提供的信息发送装置800能够实现图3的方法实施例实现的各个过程，并达到相同的技术效果，为避免重复，这里不再赘述。
如图9所示为本申请实施例提供的一种第二模型获取装置，第二模型获取装置900包括：
发送模块901，用于向第一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信息；
接收模块902，用于接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
可选地，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
模型所基于的框架信息；
模型所基于的描述方法信息；
模型基于的优化算法信息；
模型可达到的精度信息；
模型存储空间信息；
模型算力要求信息；
模型版本信息。
可选地，所述模型要求信息包括如下至少一项：
模型所基于的框架信息；
模型所基于的描述方法信息；
模型基于的优化算法信息；
模型可达到的精度信息；
模型存储空间信息；
模型算力要求信息；
模型版本要求信息。
可选地，所述框架信息用于指示模型所基于的训练框架；
和/或，
所述描述方法信息用于指示模型格式信息或模型语言信息；
和/或，
所述优化算法信息用于指示模型收敛的算法；
和/或，
所述算力要求信息用于指示模型在进行推理任务时所需要的算力。
可选地，所述第一网元包括数据存储功能网元。
可选地，所述第三网元包括分析逻辑功能AnLF网元。
本申请实施例提供的第二模型获取装置900能够实现图4的方法实施例实现的各个过程，并达到相同的技术效果，为避免重复，这里不再赘述。
如图10所示为本申请实施例提供的一种信息接收装置，信息接收装置1000包括：
接收模块1001，用于接收第一网元发送的网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息；
发送模块1002，用于向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息。
可选地，所述模型要求信息包括如下至少一项：
模型所基于的框架信息；
模型所基于的描述方法信息；
模型基于的优化算法信息；
模型可达到的精度信息；
模型存储空间信息；
模型算力要求信息；
模型版本要求信息。
可选地，所述框架信息用于指示模型所基于的训练框架；
和/或，
所述描述方法信息用于指示模型格式信息或模型语言信息；
和/或，
所述优化算法信息用于指示模型收敛的算法；
和/或，
所述算力要求信息用于指示模型在进行推理任务时所需要的算力。
可选地，所述第一网元包括数据存储功能网元。
可选地，所述第五网元包括网络存储功能NRF网元。
本申请实施例提供的信息接收装置1000能够实现图5的方法实施例实现的各个过程，并达到相同的技术效果，为避免重复，这里不再赘述。
本申请实施例中的装置可以是电子设备，例如具有操作系统的电子设备，也可以是电子设备中的部件，例如集成电路或芯片。该电子设备可以是终端，也可以为除终端之外的其他设备。示例性的，终端可以包括但不限于上述所列举的终端11的类型，其他设备可以为服务器、网络附属存储器(Network Attached Storage，NAS)等，本申请实施例不作具体限定。
可选的，如图11所示，本申请实施例还提供一种通信设备1100，包括处理器1101和存储器1102，存储器1102上存储有可在所述处理器1101上运行的程序或指令，该程序或指令被处理器1101执行时实现上述图2、图3、图4或图5所示方法实施例的各个步骤，且能达到相同的技术效果，为避免重复，这里不再赘述。
本申请实施例还提供一种第一网元，包括处理器和通信接口，所述通信接口用于接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；所述处理器用于存储所述至少一个模型的模型信息。该第一网元的实施例与上述图2所示方法实施例对应，上述方法实施例的各个实施过程和实现方式均可适用于第一网元的实施例中，且能达到相同的技术效果。
本申请实施例还提供一种第二网元，包括处理器和通信接口，所述通信接口用于向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。该第二网元的实施例与上述图3所示方法实施例对应，上述方法实施例的各个实施过程和实现方式均可适用于第二网元的实施例中，且能达到相同的技术效果。
本申请实施例还提供一种第三网元，包括处理器和通信接口，所述通信接口用于向第一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信息；接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。该第三网元的实施例与上述图4所示方法实施例对应，上述方法实施例的各个实施过程和实现方式均可适用于第三网元的实施例 中，且能达到相同的技术效果。
本申请实施例还提供一种第五网元，包括处理器和通信接口，所述通信接口用于接收第一网元发送的网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息；向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息。该第五网元的实施例与上述图5所示方法实施例对应，上述方法实施例的各个实施过程和实现方式均可适用于第五网元的实施例中，且能达到相同的技术效果。
具体地，本申请实施例还提供了一种网络侧设备。如图12所示，该网络侧设备1200包括：天线121、射频装置122、基带装置123、处理器124和存储器125。天线121与射频装置122连接。在上行方向上，射频装置122通过天线121接收信息，将接收的信息发送给基带装置123进行处理。在下行方向上，基带装置123对要发送的信息进行处理，并发送给射频装置122，射频装置122对收到的信息进行处理后经过天线121发送出去。
以上实施例中网络侧设备执行的方法可以在基带装置123中实现，该基带装置123包括基带处理器。
基带装置123例如可以包括至少一个基带板，该基带板上设置有多个芯片，如图12所示，其中一个芯片例如为基带处理器，通过总线接口与存储器125连接，以调用存储器125中的程序，执行以上方法实施例中所示的网络设备操作。
该网络侧设备还可以包括网络接口126，该接口例如为通用公共无线接口(common public radio interface，CPRI)。
具体地，本发明实施例的网络侧设备1200还包括：存储在存储器125上并可在处理器124上运行的指令或程序，处理器124调用存储器125中的指令或程序执行图7、图8、图9或图10所示各模块执行的方法，并达到相同的技术效果，为避免重复，故不在此赘述。
本申请实施例还提供一种可读存储介质，所述可读存储介质上存储有程序或指令，该程序或指令被处理器执行时实现上述图2、图3、图4或图5所示方法实施例的各个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
其中，所述处理器为上述实施例中所述的终端中的处理器。所述可读存储介质，包括计算机可读存储介质，如计算机只读存储器ROM、随机存取存储器RAM、磁碟或者光盘等。
本申请实施例另提供了一种芯片，所述芯片包括处理器和通信接口，所述通信接口和所述处理器耦合，所述处理器用于运行程序或指令，实现上述图2、图3、图4或图5所示方法实施例的各个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
应理解，本申请实施例提到的芯片还可以称为系统级芯片，系统芯片，芯片系统或片上系统芯片等。
本申请实施例另提供了一种计算机程序/程序产品，所述计算机程序/程序产品被存储在存储介质中，所述计算机程序/程序产品被至少一个处理器执行以实现上述图2、图3、图4或图5所示方法实施例的各个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
本申请实施例还提供了一种通信系统，包括：第一网元、第二网元、第三网元、第四网元和第五网元，所述第一网元可用于执行如上图2所示的方法实施例的步骤，所述第二网元可用于执行图3所示的方法实施例的步骤，所述第三网元可用于执行如上图4所示的方法实施例的步骤，所述第四网元可用于执行如上图2所示的方法实施例的步骤，所述第五网元可用于执行如上图5所示的方法实施例的步骤。
需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外，需要指出的是，本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能，还可包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能，例如，可以按不同于所描述的次序来执行所描述的方法，并且还可以添加、省去、或组合各种步骤。另外，参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以计算机软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述，但是本申请并不局限于上述的具体实施方式，上述的具体实施方式仅仅是示意性的，而不是限制性的，本领域的普通技术人员在本申请的启示下，在不脱离本申请宗旨和权利要求所保护的范围情况下，还可做出很多形式，均属于本申请的保护之内。

Claims (41)

  1. 一种模型获取方法，包括：
    第一网元接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；
    所述第一网元存储所述至少一个模型的模型信息。
  2. 根据权利要求1所述的方法，其中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本信息。
  3. 根据权利要求1所述的方法，其中，所述方法还包括：
    所述第一网元接收第三网元发送的第一模型请求消息，所述第一模型请求消息包括模型要求信息；
    所述第一网元根据所述模型要求信息确定目标模型，所述目标模型与所述模型要求信息匹配；
    所述第一网元向所述第三网元发送所述目标模型的模型信息。
  4. 根据权利要求3所述的方法，其中，所述模型要求信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本要求信息。
  5. 根据权利要求2或4所述的方法，其中，所述框架信息用于指示模型所基于的训练框架；
    和/或，
    所述描述方法信息用于指示模型格式信息或模型语言信息；
    和/或，
    所述优化算法信息用于指示模型收敛的算法；
    和/或，
    所述算力要求信息用于指示模型在进行推理任务时所需要的算力。
  6. 根据权利要求3所述的方法，其中，所述第一网元根据所述模型要求信息确定目标模型，包括：
    在所述第一网元包括与所述模型要求信息匹配的模型的情况下，所述第一网元将与所述模型要求信息匹配的模型确定为所述目标模型。
  7. 根据权利要求3所述的方法，其中，所述第一网元根据所述模型要求信息确定目标模型，包括：
    在所述第一网元不包括与所述模型要求信息匹配的模型的情况下，所述第一网元向第四网元发送第二模型请求消息，所述第二模型请求消息包括所述模型要求信息；
    所述第一网元接收所述第四网元发送的模型响应消息，所述模型响应消息包括与所述模型要求信息相匹配的模型的模型信息；
    所述第一网元将与所述模型要求信息相匹配的模型确定为所述目标模型。
  8. 根据权利要求7所述的方法，其中，所述方法还包括：
    所述第一网元向第五网元发送网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，其中，所述网元能力要求信息是所述第一网元根据所述第一模型要求信息确定的；
    所述第一网元接收所述第五网元发送的网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息；
    所述第一网元根据所述网元信息，确定所述第四网元，所述第四网元能够提供与所述模型要求信息相匹配的模型的模型信息。
  9. 根据权利要求1-2或4或6-8中任一项所述的方法，其中，所述第一网元包括数据存储功能网元。
  10. 根据权利要求7或8所述的方法，其中，所述第二网元或所述第四网元包括模型训练功能MTLF网元。
  11. 根据权利要求3所述的方法，其中，所述第三网元包括分析逻辑功能AnLF网元。
  12. 根据权利要求8所述的方法，其中，所述第五网元包括网络存储功能NRF网元。
  13. 一种信息发送方法，包括：
    第二网元向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
  14. 根据权利要求13所述的方法，其中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本信息。
  15. 根据权利要求13所述的方法，其中，在所述第二网元向第一网元发送模型存储请求消息之前，所述方法还包括：
    所述第二网元对所述至少一个模型进行训练，获得所述至少一个模型的模型信息；
    或者，
    所述第二网元接收所述至少一个模型的模型信息。
  16. 根据权利要求13所述的方法，其中，在所述第二网元向第一网元发送模型存储请求消息之前，所述方法还包括：
    所述第二网元接收所述第一网元发送的第二模型请求消息，所述第二模型请求消息包括模型要求信息，所述至少一个模型是与所述模型要求信息相匹配的模型。
  17. 根据权利要求16所述的方法，其中，所述模型要求信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本要求信息。
  18. 根据权利要求17所述的方法，其中，所述框架信息用于指示模型所基于的训练框架；
    和/或，
    所述描述方法信息用于指示模型格式信息或模型语言信息；
    和/或，
    所述优化算法信息用于指示模型收敛的算法；
    和/或，
    所述算力要求信息用于指示模型在进行推理任务时所需要的算力。
  19. 根据权利要求13-17中任一项所述的方法，其中，所述第一网元包括数据存储功能网元。
  20. 根据权利要求13-17中任一项所述的方法，其中，所述第二网元包括模型训练功能MTLF网元。
  21. 一种模型获取方法，包括：
    第三网元向第一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信 息；
    所述第三网元接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
  22. 根据权利要求21所述的方法，其中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本信息。
  23. 根据权利要求21所述的方法，其中，所述模型要求信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本要求信息。
  24. 根据权利要求22或23所述的方法，其中，所述框架信息用于指示模型所基于的训练框架；
    和/或，
    所述描述方法信息用于指示模型格式信息或模型语言信息；
    和/或，
    所述优化算法信息用于指示模型收敛的算法；
    和/或，
    所述算力要求信息用于指示模型在进行推理任务时所需要的算力。
  25. 根据权利要求21-23中任一项所述的方法，其中，所述第一网元包括数据存储功能网元。
  26. 根据权利要求21-23中任一项所述的方法，其中，所述第三网元包括分析逻辑功能AnLF网元。
  27. 一种信息接收方法，包括：
    第五网元接收第一网元发送的网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元 能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息；
    所述第五网元向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息。
  28. 根据权利要求27所述的方法，其中，所述模型要求信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本要求信息。
  29. 根据权利要求28所述的方法，其中，所述框架信息用于指示模型所基于的训练框架；
    和/或，
    所述描述方法信息用于指示模型格式信息或模型语言信息；
    和/或，
    所述优化算法信息用于指示模型收敛的算法；
    和/或，
    所述算力要求信息用于指示模型在进行推理任务时所需要的算力。
  30. 根据权利要求27或29所述的方法，其中，所述第一网元包括数据存储功能网元。
  31. 根据权利要求27或29所述的方法，其中，所述第五网元包括网络存储功能NRF网元。
  32. 一种模型获取装置，包括：
    接收模块，用于接收第二网元发送的模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项；
    存储模块，用于存储所述至少一个模型的模型信息。
  33. 根据权利要求32所述的装置，其中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本信息。
  34. 一种信息发送装置，包括：
    发送模块，用于向第一网元发送模型存储请求消息，所述模型存储请求消息包括至少一个模型的模型信息，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
  35. 根据权利要求34所述的装置，其中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本信息。
  36. 一种模型获取装置，包括：
    发送模块，用于向第一网元发送第一模型请求消息，所述第一模型请求消息包括模型要求信息；
    接收模块，用于接收所述第一网元发送的目标模型的模型信息，所述目标模型与所述模型要求信息匹配，所述模型信息包括模型文件和模型文件存储信息中的至少一项。
  37. 根据权利要求36所述的装置，其中，所述模型信息还包括模型描述信息，所述模型描述信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本信息。
  38. 一种信息接收装置，包括：
    接收模块，用于接收第一网元发送的网元查询请求消息，所述网元查询请求消息包括网元能力要求信息，所述网元能力要求信息是所述第一网元根据模型要求信息确定的，所述网元能力要求信息用于确定第四网元，所述第四网元能够提供与所述模型要求信息匹配的模型的模型信息；
    发送模块，用于向所述第一网元发送网元查询响应消息，所述网元查询响应消息包括与所述网元能力要求信息匹配的网元信息。
  39. 根据权利要求38所述的装置，其中，所述模型要求信息包括如下至少一项：
    模型所基于的框架信息；
    模型所基于的描述方法信息；
    模型基于的优化算法信息；
    模型可达到的精度信息；
    模型存储空间信息；
    模型算力要求信息；
    模型版本要求信息。
  40. 一种网元，包括处理器和存储器，所述存储器存储可在所述处理器上运行的程序或指令，其中，所述程序或指令被所述处理器执行时实现如权利要求1至31中任一项所述的方法的步骤。
  41. 一种可读存储介质，所述可读存储介质上存储程序或指令，其中，所述程序或指令被处理器执行时实现如权利要求1-31中任一项所述的方法的步骤。
PCT/CN2023/084048 2022-03-28 2023-03-27 模型获取方法、信息发送方法、信息接收方法、装置及网元 WO2023185726A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210317181.5A CN116866998A (zh) 2022-03-28 2022-03-28 模型获取方法、信息发送方法、信息接收方法、装置及网元
CN202210317181.5 2022-03-28

Publications (1)

Publication Number Publication Date
WO2023185726A1 true WO2023185726A1 (zh) 2023-10-05

Family

ID=88199408

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/084048 WO2023185726A1 (zh) 2022-03-28 2023-03-27 模型获取方法、信息发送方法、信息接收方法、装置及网元

Country Status (2)

Country Link
CN (1) CN116866998A (zh)
WO (1) WO2023185726A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112346870A (zh) * 2020-11-18 2021-02-09 脸萌有限公司 模型处理方法及系统
WO2021163895A1 (zh) * 2020-02-18 2021-08-26 Oppo广东移动通信有限公司 网络模型的管理方法及建立或修改会话的方法、装置
CN113469371A (zh) * 2021-07-01 2021-10-01 建信金融科技有限责任公司 联邦学习方法和装置
CN113961199A (zh) * 2020-12-18 2022-01-21 深圳平安智汇企业信息管理有限公司 一种模型部署系统、方法、装置及存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021163895A1 (zh) * 2020-02-18 2021-08-26 Oppo广东移动通信有限公司 网络模型的管理方法及建立或修改会话的方法、装置
CN112346870A (zh) * 2020-11-18 2021-02-09 脸萌有限公司 模型处理方法及系统
CN113961199A (zh) * 2020-12-18 2022-01-21 深圳平安智汇企业信息管理有限公司 一种模型部署系统、方法、装置及存储介质
CN113469371A (zh) * 2021-07-01 2021-10-01 建信金融科技有限责任公司 联邦学习方法和装置

Also Published As

Publication number Publication date
CN116866998A (zh) 2023-10-10

Similar Documents

Publication Publication Date Title
CN109976667A (zh) 一种镜像管理方法、装置及系统
WO2022007434A1 (zh) 可视化方法及相关设备
US10453165B1 (en) Computer vision machine learning model execution service
CN107423085B (zh) 用于部署应用的方法和装置
US20180302346A1 (en) Techniques to configure bot flow
US20150089415A1 (en) Method of processing big data, apparatus performing the same and storage media storing the same
CN111400000A (zh) 网络请求处理方法、装置、设备和存储介质
US20160259821A1 (en) Efficient storage and searching of object state and relationships at a given point of time
WO2023207980A1 (zh) 模型信息获取方法、发送方法、装置、节点和储存介质
WO2023185726A1 (zh) 模型获取方法、信息发送方法、信息接收方法、装置及网元
CN111488386A (zh) 数据查询方法和装置
US10382575B2 (en) Program execution system, method of executing program, and computer-readable storage medium
CN115563160A (zh) 数据处理方法、装置、计算机设备和计算机可读存储介质
CN106790323B (zh) 一种资源发现的方法及装置
CN113609349A (zh) 租赁车辆库存检测方法、装置及介质
CN114035861A (zh) 集群配置方法、装置、电子设备和计算机可读介质
US20220417109A1 (en) Methods for determining application of models in multi-vendor networks
US20090183172A1 (en) Middleware Bridge System And Method
CN112187916A (zh) 一种跨系统的数据同步方法与装置
CN112036418A (zh) 用于提取用户特征的方法和装置
WO2023185826A1 (zh) 网元注册方法、模型请求方法、装置、网元、通信系统及存储介质
WO2023213288A1 (zh) 模型获取方法及通信设备
WO2024178566A1 (en) Mechanism for determining energy related performance indicators for data processing entity
WO2023213246A1 (zh) 模型选择方法、装置及网络侧设备
JP7231026B2 (ja) 機器推定装置、機器推定方法、および、機器推定プログラム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23778093

Country of ref document: EP

Kind code of ref document: A1