WO2021179165A1 - 模型协调方法及装置 - Google Patents

模型协调方法及装置 Download PDF

Info

Publication number
WO2021179165A1
WO2021179165A1 PCT/CN2020/078609 CN2020078609W WO2021179165A1 WO 2021179165 A1 WO2021179165 A1 WO 2021179165A1 CN 2020078609 W CN2020078609 W CN 2020078609W WO 2021179165 A1 WO2021179165 A1 WO 2021179165A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
segment
model segment
identifier
stored
Prior art date
Application number
PCT/CN2020/078609
Other languages
English (en)
French (fr)
Inventor
许阳
李海涛
Original Assignee
Oppo广东移动通信有限公司
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 Oppo广东移动通信有限公司 filed Critical Oppo广东移动通信有限公司
Priority to PCT/CN2020/078609 priority Critical patent/WO2021179165A1/zh
Priority to CN202080098134.5A priority patent/CN115244512A/zh
Priority to EP20924693.3A priority patent/EP4113300A4/en
Publication of WO2021179165A1 publication Critical patent/WO2021179165A1/zh
Priority to US17/902,459 priority patent/US20230004839A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5061—Partitioning or combining of resources
    • G06F9/5066—Algorithms for mapping a plurality of inter-dependent sub-tasks onto a plurality of physical CPUs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/54—Interprogram communication
    • G06F9/545—Interprogram communication where tasks reside in different layers, e.g. user- and kernel-space
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/098—Distributed learning, e.g. federated learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/48—Indexing scheme relating to G06F9/48
    • G06F2209/483—Multiproc
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/48—Indexing scheme relating to G06F9/48
    • G06F2209/485—Resource constraint
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/50—Indexing scheme relating to G06F9/50
    • G06F2209/5017—Task decomposition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/50—Indexing scheme relating to G06F9/50
    • G06F2209/509—Offload
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a model coordination method and device.
  • AI Agent Intelligence
  • ML Machine Learning
  • the execution of the AI/ML model requires high equipment. Due to the limitations of the computing power and storage capacity of the mobile terminal, the AI/ML model can be divided into multiple model fragments, and some of the model fragments can be transferred to other devices. Multiple devices jointly complete the execution of the AI/ML model, thereby sharing the computing workload of the mobile terminal.
  • the AI/ML model divides the model fragments in a variety of ways, the model fragments cannot be coordinated among the various devices, which makes the effect of the model execution unsatisfactory.
  • the embodiments of the present application provide a model coordination method and device to solve the coordination problem between the model segments when the model is divided into multiple model segments and configured to be executed by multiple devices.
  • an embodiment of the present application provides a model coordination method, which is applied to a first device, and at least one model segment is stored in the first device, and the at least one model segment stored in the first device is used to implement pre-processing. Assuming part of the functions of the model, the method includes:
  • an embodiment of the present application provides a model coordination method, which is applied to a third device, and the method includes:
  • the first model segment being one of at least one model segment stored in the first device, and when the first model segment is executed in the first device, part of the function of the preset model is accomplish;
  • an embodiment of the present application provides a model coordination method, which is applied to a model control function, and the method includes:
  • an embodiment of the present application provides a model coordination device, including:
  • the determining module is configured to determine a first model segment from at least one model segment stored in the first device, where at least one model segment is stored in the first device for implementing part of the functions of the preset model.
  • first model fragment is executed and the second model fragment is executed in the second device, part or all of the functions of the preset model are realized, and the second model fragment is at least One of the model segments, at least one model segment stored in the second device is used to implement part of the functions of the preset model.
  • an embodiment of the present application provides a model coordination device, including:
  • the determining module is configured to determine a first model segment, where the first model segment is one of at least one model segment stored in the first device, and when the first model segment is executed in the first device, preset Part of the function of the model is realized;
  • the sending module is configured to send the first identifier of the first model fragment to the first device.
  • an embodiment of the present application provides a model coordination device, including:
  • a first processing module configured to determine a first model segment of a first device, where the first model segment is one of at least one model segment stored in the first device;
  • the second processing module is used to determine a second model segment of the second device, where the second model segment is one of at least one model segment stored in the second device, and when the first model segment is in the When the first device is executed and the second model fragment is executed in the second device, part or all of the functions of the preset model are implemented;
  • the sending module is configured to send the second identifier of the second model fragment to the second device.
  • an embodiment of the present application provides a model coordination device, including: a transceiver, a processor, and a memory;
  • the memory stores computer execution instructions
  • the processor executes the computer-executable instructions stored in the memory, so that the processor executes the model coordination method according to any one of the first aspect, or causes the processor to execute the model coordination method according to any one of the second aspects.
  • an embodiment of the present application provides a model coordination device, including: a transceiver, a processor, and a memory;
  • the memory stores computer execution instructions
  • the processor executes the computer-executable instructions stored in the memory, so that the processor executes the model coordination method according to any one of the third aspect.
  • an embodiment of the present application provides a computer-readable storage medium that stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, it is used to implement the first aspect, The model coordination method according to any one of the second aspect or the third aspect.
  • the first device determines the first model segment in at least one stored model segment.
  • the pre-processing can be realized. Set part or all of the functions of the model, thus realizing the effective coordination of model fragments among various devices.
  • FIG. 1 is a diagram of a network system architecture to which an embodiment of this application is applicable;
  • Figure 2 is a schematic diagram of model separation provided by an embodiment of the application.
  • FIG. 3 is a schematic diagram of an application scenario provided by an embodiment of the application.
  • FIG. 4 is a schematic flowchart of a model coordination method provided by an embodiment of the application.
  • FIG. 5 is a schematic diagram of a model identification provided by an embodiment of the application.
  • Fig. 6 is a signaling diagram 1 of the model coordination method provided by an embodiment of the application.
  • FIG. 7 is a schematic diagram 1 of model coordination provided by an embodiment of this application.
  • FIG. 8 is a signaling diagram 2 of the model coordination method provided by an embodiment of this application.
  • FIG. 9 is a second schematic diagram of model coordination provided by an embodiment of this application.
  • FIG. 10 is a signaling diagram 3 of the model coordination method provided by an embodiment of this application.
  • FIG. 11 is a third schematic diagram of model coordination provided by an embodiment of this application.
  • FIG. 12 is a signaling diagram 4 of the model coordination method provided by an embodiment of this application.
  • FIG. 13 is a fourth schematic diagram of model coordination provided by an embodiment of this application.
  • FIG. 14 is a schematic flowchart of a model coordination method provided by an embodiment of the application.
  • FIG. 15 is a schematic flowchart of a model coordination method provided by an embodiment of the application.
  • FIG. 16 is a first structural diagram of a model coordination device provided by an embodiment of this application.
  • FIG. 17 is a second structural diagram of a model coordination device provided by an embodiment of this application.
  • FIG. 18 is a third structural diagram of a model coordination device provided by an embodiment of this application.
  • FIG. 19 is a schematic structural diagram of a model coordination device provided by an embodiment of the application.
  • FIG. 20 is a schematic structural diagram of a model coordination device provided by an embodiment of the application.
  • Terminal equipment It can be a device that includes wireless transceiver functions and can cooperate with network equipment to provide users with communication services.
  • terminal equipment may refer to User Equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile equipment, user terminal, terminal, wireless communication equipment, User agent or user device.
  • UE User Equipment
  • the terminal device may be a cellular phone, a cordless phone, a Session Initiation Protocol (SIP) phone, a wireless local loop (Wireless Local Loop, WLL) station, a personal digital processing (Personal Digital Assistant, PDA), and a wireless Communication function handheld devices, computing devices or other processing devices connected to wireless modems, vehicle-mounted devices, wearable devices, future 5G networks or future evolution after 5G in the public land mobile network (PLMN) Terminal equipment, etc., this embodiment of the present application is not limited thereto.
  • SIP Session Initiation Protocol
  • WLL Wireless Local Loop
  • PDA Personal Digital Assistant
  • PDA Personal Digital Assistant
  • the terminal device may also be a wearable device.
  • Wearable devices can also be called wearable smart devices. It is a general term for using wearable technology to intelligently design everyday wear and develop wearable devices, such as glasses, gloves, watches, clothing and shoes.
  • a wearable device is a portable device that is directly worn on the body or integrated into the user's clothes or accessories. Wearable devices are not only a kind of hardware device, but also realize powerful functions through software support, data interaction, and cloud interaction.
  • wearable smart devices include full-featured, large-sized, complete or partial functions that can be achieved without relying on smart phones, such as smart watches or smart glasses, and only focus on a certain type of application function, and need to cooperate with other devices such as smart phones.
  • the terminal device may also be a terminal device in the Internet of Things (IoT) system.
  • IoT Internet of Things
  • Its main technical feature is to pass items through communication technology. Connect with the network to realize the intelligent network of human-machine interconnection and interconnection of things.
  • the IOT technology can achieve massive connections, deep coverage, and power saving of the terminal through, for example, narrowband NB technology.
  • Network equipment can be equipment used to communicate with terminal equipment, for example, it can be in the Global System for Mobile Communication (GSM) or Code Division Multiple Access (CDMA) communication system
  • the base station can also be the base station (NodeB, NB) in the Wideband Code Division Multiple Access (WCDMA) system, or the evolved base station (Evolutional Node) in the LTE system B, eNB or eNodeB), or the network equipment may be a relay station, access point, in-vehicle equipment, wearable equipment, and network side equipment in the future 5G network or networks after 5G, or the future evolution of the public land mobile network (Public Land Mobile Network).
  • Mobile Network, PLMN Mobile Network, etc. in the network.
  • the network equipment involved in the embodiments of the present application may also be referred to as a radio access network (Radio Access Network, RAN) equipment.
  • the RAN equipment is connected to the terminal equipment and is used to receive data from the terminal equipment and send it to the core network equipment.
  • RAN equipment corresponds to different equipment in different communication systems. For example, it corresponds to base station and base station controller in 2G system, corresponds to base station and radio network controller (RNC) in 3G system, and corresponds to evolution in 4G system.
  • Evolutional Node B (eNB) corresponds to the 5G system in the 5G system, such as the access network equipment in the NR (for example, gNB, centralized unit CU, distributed unit DU).
  • AS Access Stratum, access layer
  • NAS None Access Stratum, non-access layer
  • RRC Radio Resource Control, radio resource control.
  • AI/ML models have begun to be widely used in intelligent data analysis.
  • terminal devices are also increasingly introducing AI/ML model models to perform intelligent operations such as voice recognition, image processing, and user behavior prediction, so that terminal devices can serve more business scenarios.
  • FIG 1 is a diagram of a network system architecture to which the embodiments of this application are applicable.
  • the network architecture may be a 5G network architecture.
  • the 5G system is also called a new wireless communication system and a new access technology (New Radio, NR) Or the next generation mobile communication system.
  • New Radio NR
  • the access network in the 5G system can be a radio access network ((R)AN), and the (R)AN device in the 5G system can be composed of multiple 5G-(R)AN nodes.
  • the 5G-( R) AN nodes may include: non-3GPP access networks such as access points (APs) of WiFi networks, next-generation base stations (collectively referred to as next-generation radio access network nodes (NG-RAN nodes), among which, Next-generation base stations include new air interface base stations (NR nodeB, gNB), new generation evolved base stations (NG-eNB), central unit (CU) and distributed unit (DU) separated forms of gNB, etc.), Transmission receive point (TRP), transmission point (TP) or other nodes.
  • APs access points
  • NG-RAN nodes next-generation radio access network nodes
  • Next-generation base stations include new air interface base stations (NR nodeB, gNB), new generation evolved base stations (NG-eNB), central unit (CU) and distributed unit (DU) separated forms of
  • the 5G core network (5G core/new generation core, 5GC/NGC) includes access and mobility management function (Access and Mobility Management Function, AMF) network elements, session management function (Session Management Function, SMF) ) Network element, user plane function (UPF) network element, authentication server function (Authentication Server Function, AUSF) network element, policy control function (Policy Control Function, PCF) network element, application function (Application Function, AF) network element, unified data management function (UDM) network element, network slice selection function (Network Slice Selection Function, NSSF) network element and other functional units.
  • AMF Access and Mobility Management Function
  • SMF Session Management Function
  • UPF user plane function
  • Policy Control Function Policy Control Function
  • PCF Policy Control Function
  • UDM unified data management function
  • NSSF Network Slice Selection Function
  • the AMF network element is mainly responsible for services such as mobility management and access management.
  • the SMF network element is mainly responsible for session management, UE address management and allocation, dynamic host configuration protocol functions, selection and control of user plane functions, etc.
  • UPF is mainly responsible for data packet routing and forwarding, message filtering, and execution of quality of service (QoS) control related functions that are connected to the data network (DN) and user plane externally.
  • AUSF is mainly responsible for the authentication function of terminal equipment.
  • the PCF network element is mainly responsible for providing a unified policy framework for network behavior management, providing policy rules for control plane functions, and obtaining registration information related to policy decisions.
  • control and mobility management functions such as access authentication, security encryption, location registration for terminal equipment, and user Session management functions such as the establishment, release, and modification of the surface transmission path.
  • the model control function is a network element in the core network device, which is used to control when the model fragments are executed jointly among various devices. Coordination between model fragments.
  • the model control function can be a new network element in the core network equipment, or it can integrate the model control function’s coordinated control of model fragments into a certain network element in the current core network equipment, such as AMF network elements, SMF network elements and so on.
  • the network element that integrates the coordinated control of the model fragments is the model control function in the embodiment of this application.
  • the functional units in the 5GC can communicate through the next generation network (NG) interface.
  • the UE can transmit control plane messages with the AMF network element through the NG interface 1 (abbreviated as N1), and the RAN device can communicate through the NG interface.
  • Interface 3 (abbreviated as N3) establishes a user plane data transmission channel with UPF.
  • AN/RAN equipment can establish a control plane signaling connection with AMF network elements through NG interface 2 (abbreviated as N2), and UPF can communicate with UPF via NG interface 4 (abbreviated as N4).
  • FIG. 1 is only an exemplary architecture diagram. In addition to the functional units shown in FIG. 1, the network architecture may also include other functional units.
  • the network architecture shown in Figure 1 is a reference point-based network architecture, and the network architecture is a network architecture in a non-roaming scenario.
  • the method of this application can also be applied in a roaming scenario, and the network architecture is not limited to a reference point-based network Architecture, a network architecture based on service-oriented interfaces can also be adopted.
  • the AI/ML model has higher requirements for the equipment. If the terminal equipment is used to run the AI/ML model independently, the ability of the terminal equipment is Bigger challenge. Even if the terminal device can independently complete the execution of the AI/ML model, it may take a long time. Based on this, the AI/ML model can be divided into two or more model fragments, and two or more devices execute the corresponding fragments respectively. For example, part of the model fragments can be divided into network equipment, and the AI/ML model can be executed jointly by the network equipment and the terminal equipment.
  • Fig. 2 is a schematic diagram of model separation provided by an embodiment of the application. As shown in Fig. 2, it is aimed at the division of an AI/ML model.
  • the AI/ML model is an image recognition model that can recognize things in an image. In the embodiment of the present application, it is set that the model has been pre-trained.
  • the image recognition model is divided into three model segments, which are respectively configured on three devices for execution. As shown in FIG. 2, the three devices are the terminal device 20, the first network device 21, and the second network device 22, and each device only executes its own model segment.
  • any one or two devices can realize part of the functions of the image recognition model. After the three devices execute their respective model fragments, they can realize all the functions of the image recognition model.
  • the output of the final model is "tree", which means that after outputting an image, the model recognizes that the thing on the image is a tree.
  • an image recognition model may include multiple different parts, for example, it may include several convolutional layers, several pooling layers, several up-sampling layers, several down-sampling layers, etc., in the model
  • the various parts are arranged organically, and after data is input, each part may output intermediate results, and the output result of the previous part may be the input of the next part. Therefore, as shown in FIG. 2, the terminal device 20 executes After the corresponding model segment, the output obtained is an intermediate result, and the intermediate result is the input of the model segment executed by the first network device 21.
  • each device executes its model fragments in a sequential order. For example, in Figure 2, only the terminal device 20 executes its model fragments, and obtains intermediate results as the input of the first network device 21. Only the device 21 can execute its corresponding model segment and obtain corresponding intermediate data, and then the intermediate data output by the first network device 21 is the input of the model segment executed by the second network device 22, and so on.
  • models may be divided in different ways, and coordination between models cannot be achieved when model fragments are executed between devices.
  • the division method may be, for example:
  • model segment A1 including preprocessing layer-convolutional layer
  • model segment B1 including upsampling layer-convolutional layer-pooling layer
  • model segment C1 including convolutional layer-activation layer-fully connected Floor
  • model segment A2 including preprocessing layer
  • model segment B2 including convolutional layer-upsampling layer-convolutional layer-pooling layer
  • model segment C2 including convolutional layer-activation layer-fully connected Floor
  • model segment A3 including preprocessing layer-convolutional layer-upsampling layer
  • model segment B3 including convolutional layer-pooling layer-convolutional layer-activation layer
  • model segment C3 including fully connected Floor
  • model segment A1, model segment A2, and model segment A3 can be configured to the terminal device 20, model segment B1, model segment B2, and model segment B3 can be configured to the first network device 21, and model segment C1 , The model segment C2 and the model segment C3 are configured to the second network device 22.
  • the split and coordination between models is a problem.
  • the model segment selected to be executed by the terminal device 20 is A1
  • the execution segment that the first network device 21 may select is B2
  • the execution segment that the second network device 22 may select is C3.
  • the model executed by the three devices in total It is the preprocessing layer-convolutional layer-convolutional layer-upsampling layer-convolutional layer-pooling layer-fully connected layer, which is different from the original image recognition model.
  • model segment A (accounting for 80% of the model), model segment B (accounting for 10% of the model), and model segment C (accounting for 10% of the model), and they are respectively configured to the terminal device 20, the first The network device 21 and the second network device 22.
  • the power of the terminal device 20 is low. If the terminal device 20 executes the model segment A, the power at this time is not enough to support the terminal device 20 to complete the execution of the model segment A.
  • model segment D (accounting for 10% of the model), model segment E (accounting for 30% of the model), and model segment F (accounting for 60% of the model), respectively Configured to the terminal device 20, the first network device 21, and the second network device 22, the terminal device 20 can choose to execute the model segment D when the power is low.
  • FIG. 3 is a schematic diagram of an application scenario provided by an embodiment of the application. As shown in FIG. 3, it includes a cloud simulation server 31, a model control function 32, a first device 33, and a second device 34.
  • the AI/ML model has been trained.
  • the cloud model server 31 splits the AI/ML model in multiple ways to form a collection of different model fragments and configure them to multiple AI/ML model endpoints.
  • the AI/ML model endpoints are executed The device of the model fragment in the AI/ML model. For example, after splitting the AI/ML model endpoints into model fragments and configuring them to the first device 33 and the second device 34 respectively, the first device 33 and the second device 34 are both AI/ML model endpoints.
  • the AI/ML model has been split in multiple ways.
  • Figure 3 illustrates two split methods.
  • the AI/ML model is split into model fragment A1 + model fragment B1, and it can also be split into model fragment A2 + model fragment. B2, then configure the model segment A1 and the model segment A2 to the first device 33, and configure the model segment B1 and the model segment B2 to the second device 34.
  • the model segment selection strategy is used to ensure the above-mentioned selection scheme.
  • the model segment selection strategy is determined by the cloud simulation server 31.
  • the cloud simulation server 31 can configure the model segment selection strategy to the model control function 32, or configure the model segment selection strategy to the first device 33 or the second device 34.
  • the first device 33, the second device 34, or the model control function 32 may determine the model segment that each device determines to execute.
  • FIG. 4 is a schematic flow diagram of the model coordination method provided by an embodiment of the application, as shown in FIG. 4. The method includes:
  • the first device stores at least one model segment
  • the second device also stores at least one model segment.
  • the first device determines the first model segment from the stored at least one model segment
  • the second device stores the At least one model segment is determined to be the second model segment.
  • the first model segment and the second model segment are both part of the preset model.
  • the first device executes the first model segment and the second device executes the second model segment, part or all of the functions of the preset model are realized.
  • the first device may be a terminal device or a network device
  • the second device may also be a terminal device or a network device.
  • the model fragments stored in the first device and the second device may be pre-stored in the first device and the second device after the preset model is divided.
  • the preset model may be divided into multiple types, for example, The preset model is divided into model fragment A1 and model fragment B1.
  • the preset model can also be divided into model fragment A2 and model fragment B2.
  • model fragment A1 and model fragment A2 are model fragments on the first device side and can be stored in The first device
  • the model segment B1 and the model segment B2 are model segments on the second device side and can be stored in the second device.
  • Model fragment A1 and model fragment B1 are matched. When model fragment A1 and model fragment B1 are executed, all functions of the preset model can be realized. Model fragment A2 and model fragment B2 are matched. When model fragment A2 and model fragment B2 are matched, When the model segment B2 is executed, it can realize all the functions of the preset model.
  • the solution implemented in this application is that when the first device selects model segment A1, the model segment selected by the second device is B1, and when the first device selects model segment A2, the model segment selected by the second device is B2, so Realize the coordinated division of the model.
  • the preset model When the preset model is divided into two model fragments and executed by the first device and the second device respectively, all functions of the preset model are realized; when the preset model is divided into two or more model fragments, the first When the device and the second device execute their respective model fragments, they realize part of the functions of the preset model.
  • the first device determines the first model segment in at least one stored model segment.
  • the pre-processing can be realized. Set part or all of the functions of the model to achieve effective coordination of the model.
  • the preset model is a model that has been trained.
  • the cloud model server in the data network can transfer the AI/ML
  • the model is split in multiple ways to form a collection of different model fragments and configured to multiple AI/ML endpoints for execution.
  • the AI/ML endpoint in the embodiment of this application is the device that executes the AI/ML model.
  • the preset model is AI/ML
  • the first device and the second device are both AI/ML endpoints
  • the first device and The second device is a peer AI/ML endpoint to each other.
  • the first device is a first terminal device or a first network device
  • the second device is a second terminal device or a second network device.
  • the first network device may be a base station or a core network device
  • the second network device may It is a base station or core network equipment.
  • one AI/ML endpoint is a mobile terminal
  • the other AI/ML endpoint is a network node in a mobile network, including but not limited to core network nodes or base stations, where core network nodes include but are not limited to mobility management functions, session management functions, User plane functions or network data analysis nodes specially set up in the core network, etc.
  • the method for determining the first model segment is to obtain first information, and then determine the first model segment from at least one model segment stored in the first device according to the first information.
  • the first information includes at least one of a model segment selection strategy and a first identifier of the first model segment.
  • the model segment selection strategy includes the corresponding relationship between the model segment selection condition and the model segment stored in the first device. If the first information includes the model segment selection strategy, the model segment selection condition and the model segment stored in the first device can be selected according to the model segment selection strategy. The corresponding relationship between to determine the first model segment. If the first information includes the first identifier of the first model segment, the first model segment can be determined directly according to the first identifier.
  • the model segment selection strategy also includes the identification of the preset model. Therefore, at least the preset model stored in the first device can be determined according to the identification of the preset model.
  • FIG. 5 is a schematic diagram of a model identification provided by an embodiment of the application. As shown in FIG. 5, it includes a first preset model and a second preset model.
  • the first preset model can be divided into model fragment A1+model fragment B1, and can also be divided into model fragment A2+model fragment B2.
  • the second preset model can be divided into model fragment A1+model fragment B1, and can also be divided into model fragment A2+model fragment B2.
  • model segment A1 of the first preset model and the model segment A1 of the second preset model are configured in the terminal device 51, and the model segment B1 of the first preset model and the model segment B1 of the second preset model are configured To the network device 52.
  • the terminal device 51 may be the first device, or the network device 52 may be the first device.
  • the two stored model segments are model segments A1, but they belong to two different preset models.
  • the model segment selection strategy may include the identification of the first preset model, and then according to the identification of the first preset model, the terminal device 51 may learn the determined first model The segment is the model segment A1 under the first preset model.
  • the first device may send the first identification of the first model fragment to the second device, and the second device may send the first identification of the first model fragment to the second device according to the first identification and the model fragment.
  • the selection strategy determines the second model segment, where the model segment selection strategy includes the correspondence between the model segment selection condition and the model segment stored in the first device;
  • the first device may also send the first identification of the first model fragment to the model control function after determining the first model fragment according to the model fragment selection strategy, and the model control function may be the second device according to the first identification and the model fragment selection strategy Determine a second model segment, and send a second identifier of the second model segment to the second device, where the model segment selection strategy includes a correspondence between model segment selection conditions and model segments stored in the second device;
  • the first device may also determine the first model fragment according to the model fragment selection strategy, determine the second model fragment for the second device according to the model fragment selection strategy, and send the second identifier of the second model fragment to the second device, so that the second After receiving the second identifier, the device determines the second model segment, where the model segment selection strategy further includes the correspondence between the model segment selection condition and the model segment stored in the second device.
  • the second device may send a second identifier of the second model segment to the first device, and the first device determines the first model segment according to the second identifier and the model segment selection strategy, where:
  • the model segment selection strategy includes the correspondence between the model segment selection condition and the model segment stored in the first device;
  • the second device may also send the second identifier of the second model segment to the model control function.
  • the model control function determines the first model segment for the first device according to the model segment selection strategy and the second identifier, and sends the first model segment to the first device.
  • the first identifier of the model segment the first device determines the first model segment according to the first identifier, wherein the model segment selection strategy includes the correspondence between the model segment selection condition and the model segment stored in the first device;
  • the second device may also determine the second model segment according to the model segment selection strategy, determine the first model segment for the first device according to the model segment selection strategy, and send the first identifier of the first model segment to the first device, thereby After receiving the first identifier, the first device determines the first model segment, where the model segment selection strategy further includes a correspondence between the model segment selection condition and the model segment stored in the first device.
  • the model segment selection strategy stored in the first device is received from the model control function, and the model segment stored in the second device The selection strategy is also received from the model control function.
  • the first device may be the first terminal device or the first network device
  • the second device may be the second terminal device or the second network device
  • the network device may also be a base station or a core network device. Therefore, it needs to be based on the actual type of the device. To determine the specific transmission method.
  • the first device receives the model segment selection strategy from the model control function.
  • the model control function is a network element in the core network, and data transmission between the two requires wireless access such as a base station.
  • the network device performs the forwarding.
  • the first device receives the model segment selection strategy from the model control function. This means that the model segment selection strategy received by the first device comes from the model control function, but it does not mean it is directly controlled by the model control function. Sent to the terminal device.
  • one of the two devices, the first device and the second device is the UE, and the other is the network node as an example to illustrate the solution of the present application.
  • the UE may be the first device or the second device.
  • the device, the network node may be the first device or the second device.
  • Fig. 6 is a signaling diagram 1 of the model coordination method provided by an embodiment of the application, as shown in Fig. 6, including:
  • S61 The server sends a model segment selection strategy to the model control function.
  • the model control function in the embodiment of the present application is a network element in the core network device.
  • the server sends the model segment selection strategy to the UE and the network node through the model control function.
  • the server can also directly send the model segment to the UE or the network node. Selection strategy.
  • the server same-model control function sends a model segment selection strategy to the UE and the network device as an example.
  • the model control function sends a model segment selection strategy to the UE and the network node respectively.
  • the model control function After obtaining the model segment selection strategy, the model control function sends the model segment selection strategy to the UE and the network node respectively.
  • the model control function sends the UE-side model fragment selection strategy to the UE, and the UE-side model fragment selection strategy includes the correspondence between the model fragment selection conditions and the model fragments stored on the UE side; the model control function sends the network node to the network node
  • the end model fragment selection strategy, the network node end model fragment selection strategy includes the correspondence between the model fragment selection conditions and the model fragments stored on the network node end.
  • the model control function configures the model fragment selection strategy to the UE.
  • the model control function sends the model fragment selection strategy to the mobility management function, and then the mobility management function sends the model fragment selection strategy to the UE via the NAS message, where the model fragment selection strategy Included in the NAS message. It is understandable that when the mobility management function sends the NAS message to the UE, it is sent to the UE through the forwarding of a radio access network device (for example, a certain base station).
  • a radio access network device for example, a certain base station.
  • the UE obtains the UE-side model segment set and the model segment selection strategy.
  • FIG. 7 is a schematic diagram 1 of model coordination provided by an embodiment of this application. As shown in FIG. 7, it includes a model control function 70, a UE 71 and a base station 72.
  • the model segments configured by the server for the UE71 are model segment A1, model segment A2, and model segment A3, and the model segments configured by the server for the network node are model segment B1, model segment B2, and model segment B3.
  • the network node in FIG. 7 is a base station 72.
  • the network node may also be a core network device.
  • the model segment selection strategy includes model segment selection conditions and the corresponding relationship between the stored model segments.
  • the model segment selection conditions may be, for example, the power of the UE, the location of the UE (e.g., cell identity), and the air interface signal quality of the UE.
  • the model fragment selection condition may be not only related information of the UE, but also the load of the network node, or information that is not related to the first device and the second device, such as the current time. In the embodiment of the present application, the selection condition of the model segment is not limited.
  • the network node obtains the model segment set and the model segment selection strategy on the network node side.
  • model fragments in the network node-side model fragment collection obtained by the network node match the model fragments in the UE-side model fragment collection.
  • model fragment A1 matches model fragment B1
  • model fragment A2 matches model Fragment B2 matches
  • model fragment A3 matches model fragment B3.
  • the UE determines a model segment on the UE side.
  • the UE 71 obtains the model segment selection strategy from the model control function 70, and then determines the UE-side model segment according to the model segment selection strategy.
  • the model segment selection condition is the power of the UE.
  • the model segment A1 is selected.
  • the model segment A2 is selected.
  • the power of the UE is high
  • model segment A3 is selected. At this time, the UE learns that its own power is 25%, and then determines that the executed model segment is model segment A1.
  • the model fragment A1 is the first model fragment, and A1 is the first identifier of the first model fragment; when the UE is the second device, the model fragment A1 is the second model fragment, and A1 is the first model fragment.
  • the second identifier of the second model segment is the first model fragment.
  • the UE sends the UE-side model segment identifier to the network node.
  • the UE determines that the executed model segment is the model segment A1, it sends the identifier of the model segment A1 to the network node.
  • the UE 71 sends the identifier A1 to the base station 72.
  • the network node is a base station, and the UE can directly transmit data with the base station.
  • the network node is a core network device, the UE can also transmit data with the core network device, but data transmission between the UE and the core network device requires a radio access network device such as a base station to act as a relay node for forwarding.
  • the UE can report the identity A1 to the core network device through a NAS message.
  • the network node is a base station
  • the UE can report the identity A1 to the base station through an AS message.
  • the AS message can be, for example, an RRC message.
  • the network node determines a model segment on the network node side.
  • the network node After receiving the UE-side model segment identifier, the network node can determine the network node-side model segment according to the UE-side model segment identifier and the model segment selection strategy. For example, the model segment determined by the base station 72 in FIG. 7 is the model segment B1.
  • the model coordination scheme illustrated in Figures 6 and 7 is that the first device obtains the model segment selection strategy from the model control function, and the first device obtains the model segment selection strategy from the model control function.
  • the second device obtains the model fragment selection strategy from the model control function.
  • the first device determines the first model segment (that is, the model segment A1 in FIG. 7) according to the model segment selection strategy, and then sends the first identifier of the first model segment (that is, the identifier A1 in FIG. 7) to the second device, Then the second device determines the second model segment (that is, the model segment B1 in FIG. 7) according to the first identifier and the model segment selection strategy, thereby achieving coordination between the model segments.
  • the model coordination scheme illustrated in Figures 6 and 7 is that the second device obtains the model segment selection strategy from the model control function, and the first A device obtains a model fragment selection strategy from the model control function.
  • the second device determines the second model segment (ie, the model segment A1 in FIG. 7) according to the model segment selection strategy, and then sends the second identifier of the second model segment (ie, the identifier A1 in FIG. 7) to the first device, Then the first device determines the first model segment (that is, the model segment B1 in FIG. 7) according to the second identifier and the model segment selection strategy, thereby achieving coordination between the model segments.
  • Fig. 8 is a signaling diagram 2 of the model coordination method provided by an embodiment of the application, as shown in Fig. 8, including:
  • S81 The server sends a model segment selection strategy to the model control function.
  • the model control function in the embodiment of the present application is a network element in the core network device.
  • the server sends the model segment selection strategy to the UE and the network node through the model control function.
  • the server can also directly send the model segment to the UE or the network node. Selection strategy.
  • the server and model control function sends a model segment selection strategy to the UE and the network device as an example.
  • the model control function sends a model segment selection strategy to the network node.
  • the model control function After obtaining the model fragment selection strategy, the model control function sends the model fragment selection strategy to the network node.
  • the model control function sends a network node-side model fragment selection strategy to the network node, and the network node-side model fragment selection strategy includes the correspondence between the model fragment selection conditions and the model fragments stored on the network node.
  • S83 The UE obtains a UE-side model segment set.
  • FIG. 9 is a second schematic diagram of model coordination provided by an embodiment of this application. As shown in FIG. 9, it includes a model control function 90, a UE 91 and a base station 92.
  • the model segments configured by the server for UE91 are model segment A1, model segment A2, and model segment A3, and the model segments configured by the server for the network node are model segment B1, model segment B2, and model segment B3.
  • the model segment selection strategy includes model segment selection conditions and the corresponding relationship between the stored model segments.
  • the model segment selection conditions may be, for example, the power of the UE, the location of the UE (e.g., cell identity), and the air interface signal quality of the UE.
  • the model fragment selection condition may be not only related information of the UE, but also the load of the network node, or information that is not related to the first device and the second device, such as the current time. In the embodiment of the present application, the selection condition of the model segment is not limited.
  • S84 The network node obtains the model segment set and the model segment selection strategy on the network node side.
  • model fragments in the network node-side model fragment collection obtained by the network node match the model fragments in the UE-side model fragment collection.
  • model fragment A1 matches model fragment B1
  • model fragment A2 matches model Fragment B2 matches
  • model fragment A3 matches model fragment B3.
  • the network node determines the network node-side model segment and the UE-side model segment.
  • the base station 92 obtains the model segment selection strategy from the model control function 90, and then determines the base station-side model segment according to the model segment selection strategy.
  • the model segment selection condition is the location of the UE.
  • the base station 92 selects the model segment B1.
  • the base station 92 selects the model segment B2.
  • the base station 92 selects the model segment B3.
  • the location of the UE is within the coverage of cell A, and the base station 92 determines that the executed model segment is model segment B1.
  • the model segment B1 is the first model segment, and B1 is the first identifier of the first model segment; when the base station 92 is the second device, the model segment B1 is the second model segment, B1 Is the second identifier of the second model segment.
  • the network node determines the UE-side model segment according to the network node-side model segment. For example, after the base station 92 determines the model segment B1, it determines the model segment A1 for the UE.
  • the network node sends the UE-side model segment identifier to the UE.
  • the network node When the network node determines that the executed model segment is model segment B1 and the model segment A1 is determined for the UE, it sends the identifier of the model segment A1 to the UE. For example, in FIG. 9, the UE 91 sends the identifier A1 to the base station 92.
  • the network node in the embodiment of the present application may be a base station or a core network device.
  • the UE-side model segment identifier A1 may be included in the AS message.
  • the base station sends the UE-side model segment identifier to the UE through the AS message.
  • the AS message may be, for example, an RRC message.
  • the network node is a core network device, the UE-side model segment identifier A1 may be included in the NAS message, and the core network device may send the UE-side model segment identifier to the UE through the NAS message.
  • S87 The UE determines the UE-side model segment.
  • the UE After receiving the UE-side model segment identifier, the UE can determine the UE-side model segment. For example, the model segment determined by UE92 in FIG. 9 is model segment A1.
  • the model coordination solution illustrated in Figs. 8 and 9 is that the second device obtains the model segment selection strategy from the model control function.
  • the second device determines the second model segment (ie, the model segment B1 in FIG. 9) according to the model segment selection strategy, and then determines the first model segment (ie, the model segment A1 in FIG. 9) for the first device.
  • the second device sends the first identifier of the first model segment (ie, the identifier A1 in FIG. 9) to the first device, and the first device determines the first model segment (ie, the model segment A1 in FIG. 9) according to the first identifier. ), so as to achieve coordination between model fragments.
  • the model coordination scheme illustrated in Figs. 8 and 9 is that the second device obtains the model segment selection strategy from the model control function.
  • the second device determines the second model segment (ie, the model segment B1 in FIG. 9) according to the model segment selection strategy, and then determines the first model segment (ie, the model segment A1 in FIG. 9) for the first device.
  • the second device sends the first identifier of the first model segment (ie, the identifier A1 in FIG. 9) to the first device, and the first device determines the first model segment (ie, the model segment A1 in FIG. 9) according to the first identifier. ), so as to achieve coordination between model fragments.
  • signaling interaction is performed between two model endpoints, so as to realize the coordination of the model fragments that need to be executed on each model endpoint.
  • the following will introduce a solution for the model control function in the mobile network to coordinate the model fragments between the model endpoints.
  • Fig. 10 is a signaling diagram 3 of the model coordination method provided by an embodiment of this application, as shown in Fig. 10, including:
  • S101 The server sends a model segment selection strategy to the model control function.
  • the model control function in the embodiment of the present application is a network element in the core network device.
  • the server sends the model segment selection strategy to the UE and the network node through the model control function.
  • the server can also directly send the model segment to the UE or the network node. Selection strategy.
  • the server and model control function sends a model segment selection strategy to the UE and the network device as an example.
  • S102 The UE obtains a UE-side model segment set.
  • the network node obtains a set of model fragments on the network node side.
  • FIG. 11 is a third schematic diagram of model coordination provided by an embodiment of this application. As shown in FIG. 11, it includes a model control function 110, a UE 111 and a base station 112.
  • the model segments configured by the server for the UE111 are model segment A1, model segment A2, and model segment A3, and the model segments configured by the server for the network node are model segment B1, model segment B2, and model segment B3.
  • the model segment selection strategy includes model segment selection conditions and the corresponding relationship between the stored model segments.
  • the model segment selection conditions may be, for example, the power of the UE, the location of the UE (e.g., cell identity), and the air interface signal quality of the UE.
  • the model fragment selection condition may be not only related information of the UE, but also the load of the network node, or information that is not related to the first device and the second device, such as the current time. In the embodiment of the present application, the selection condition of the model segment is not limited.
  • the model control function determines the UE-side model segment and the network node-side model segment.
  • the model control function is responsible for determining the model segments that need to be executed on the UE and network nodes.
  • the model control function can obtain the location of the UE (such as the cell identity) or the air interface signal quality of the UE from other network elements in the mobile network, or determine to select the model segment A1 for the UE according to the current time and other information, and select the model for the corresponding network node Fragment B1, and under another condition, for example, based on conditions including but not limited to the UE's position movement or air interface signaling changes, the model fragment A2 is selected for the UE, and the model fragment B2 is selected for the corresponding network node.
  • model fragments in the network node-side model fragment collection obtained by the network node match the model fragments in the UE-side model fragment collection.
  • model fragment A1 matches model fragment B1
  • model fragment A2 matches model Fragment B2 matches
  • model fragment A3 matches model fragment B3.
  • the model control function sends the UE-side model segment identifier to the UE.
  • the model control function is a network element in the core network
  • the model control function when the model control function sends the UE-side model segment identifier to the UE, it can send the UE-side model segment identifier to the mobility management function, and then the mobility management function sends a NAS message
  • the UE-side model segment identifier is sent to the UE, where the UE-side model segment identifier is included in the NAS message.
  • the model control function sends the network node-side model segment identifier to the network node.
  • the model control function 110 determines the model segment A1 for the UE 111 and determines the model segment B1 for the base station 112, it sends an identifier A1 to the UE 111 and an identifier B2 to the base station 112.
  • the UE 111 determines the UE-side model segment A1 according to the identifier A1, and the base station 112 determines the network node-side model segment B1 according to the identifier B1.
  • the model coordination scheme illustrated in Figures 10 and 11 is that the model control function determines the first device for the first device according to the model segment selection strategy.
  • the first model segment (ie model segment A1 in Figure 11), and the second model segment (ie model segment B1 in Figure 11) is selected for the second device, and then the first identifier of the first model segment is sent to
  • the first device sends the second identifier of the second model segment to the second device, the first device determines the first model segment according to the first identifier, and the second device determines the second model segment according to the second identifier, thereby realizing the model segment Coordination between.
  • the model coordination scheme illustrated in Figures 10 and 11 is that the model control function determines the second device for the second device according to the model segment selection strategy.
  • the second model segment (ie model segment A1 in Figure 11), and the first model segment (ie model segment B1 in Figure 11) is selected for the first device, and then the second identifier of the second model segment is sent to
  • the second device sends the first identifier of the first model segment to the first device, the second device determines the second model segment according to the second identifier, and the first device determines the first model segment according to the first identifier, thereby realizing the model segment Coordination between.
  • Fig. 12 is a signaling diagram 4 of the model coordination method provided by an embodiment of this application, as shown in Fig. 12, including:
  • S121 The server sends a model segment selection strategy to the model control function.
  • the model control function in the embodiment of the present application is a network element in the core network device.
  • the server sends the model segment selection strategy to the UE and the network node through the model control function.
  • the server can also directly send the model segment to the UE or the network node. Selection strategy.
  • the server and model control function sends a model segment selection strategy to the UE and the network device as an example.
  • the model control function sends a model segment selection strategy to the UE.
  • the model control function After obtaining the model segment selection strategy, the model control function sends the model segment selection strategy to the UE.
  • the model control function sends a UE-side model fragment selection strategy to the UE, and the UE-side model fragment selection strategy includes the correspondence between the model fragment selection conditions and the model fragments stored on the UE.
  • S123 The UE obtains a model segment selection strategy and a UE-side model segment set.
  • FIG. 13 is a fourth schematic diagram of model coordination provided by an embodiment of this application. As shown in FIG. 13, it includes a model control function 130, a UE 131 and a base station 132.
  • the model segments configured by the server for the UE131 are model segment A1, model segment A2, and model segment A3, and the model segments configured by the server for the network node are model segment B1, model segment B2, and model segment B3.
  • the model segment selection strategy includes model segment selection conditions and the corresponding relationship between the stored model segments.
  • the model segment selection conditions may be, for example, the power of the UE, the location of the UE (e.g., cell identity), and the air interface signal quality of the UE.
  • the model fragment selection condition may be not only related information of the UE, but also the load of the network node, or information that is not related to the first device and the second device, such as the current time. In the embodiment of the present application, the selection condition of the model segment is not limited.
  • the network node obtains a set of model fragments on the network node side.
  • the model fragments in the model fragment set on the network node side acquired by the network node match the model fragments in the model fragment set on the UE side.
  • the model fragment A1 matches the model fragment B1
  • the model fragment A2 matches the model.
  • Fragment B2 matches
  • model fragment A3 matches model fragment B3.
  • S125 The UE determines a model segment on the UE side.
  • the UE determines the UE-side model segment according to the model segment selection strategy, where the model segment selection strategy includes the corresponding relationship between the model segment selection condition and the model segment stored in the UE.
  • the model segment selection condition is the location of the UE.
  • UE131 selects model segment A1.
  • UE131 selects model segment A2.
  • the UE is in the coverage area of cell A and cell
  • UE131 selects model segment A3.
  • the location of the UE 131 is within the coverage of the cell A, and the UE 131 determines that the executed model segment is the model segment A1.
  • S126 The UE sends the UE-side model segment identifier to the model control function.
  • the UE After the UE determines the model segment to be executed, it sends the identifier of the UE-side model segment to the model control function. For example, in FIG. 13, the UE 131 sends the identifier A1 to the model control function 130. Since the model control function is a network element of the core network, the way for the UE to send the identifier of the UE-side model segment to the model control function is to carry the identifier of the UE-side model segment in the NAS message, and then the UE sends the UE-side model segment through the NAS message The identifier is reported to the mobility management function in the core network, and the mobility management function forwards the UE-side model fragment identifier to the model control function.
  • the model control function determines the model segment on the network node side.
  • the model control function After the model control function receives the UE-side model segment identifier, it can determine the network node-side model segment according to the UE-side model segment identifier and model segment selection conditions. For example, the network node-side model segment determined by the model control function 130 in FIG. 13 is model segment B1 .
  • the model control function sends the network node-side model segment identifier to the network node.
  • the network node may determine the network node-side model segment according to the network node-side model segment identifier.
  • the model coordination scheme illustrated in Figures 12 and 13 is that the first device determines the first model segment according to the model segment selection strategy (Ie the model segment A1 in Figure 13), and then the first device sends the first identifier of the first model segment to the model control function, and the model control function determines the second model segment for the second device based on the first identifier (ie, in Figure 13
  • the model segment B1) of the second model segment is sent to the second device, and the second device determines the second model segment according to the second identifier, thereby realizing coordination between the model segments.
  • the model coordination solution illustrated in Figures 12 and 13 is that the second device determines the second model segment according to the model segment selection strategy (Ie the model segment A1 in Figure 13), and then the second device sends the second identifier of the second model segment to the model control function, and the model control function determines the first model segment for the first device based on the second identifier (ie, in Figure 13 The model segment B1), and sends the first identifier of the first model segment to the first device, and the first device determines the first model segment according to the first identifier, thereby realizing coordination between the model segments.
  • the model segment selection strategy Ie the model segment A1 in Figure 13
  • the model control function determines the first model segment for the first device based on the second identifier (ie, in Figure 13 The model segment B1), and sends the first identifier of the first model segment to the first device, and the first device determines the first model segment according to the first identifier, thereby realizing coordination between the model segments.
  • the UE and the network device are used as the first device or the second device as an example for description.
  • the first device and the second device may be both UEs, or both network devices, or among them
  • the data transmission methods between the UE and the model control function, the UE and the base station, and the UE and the core network equipment are described.
  • the model control function configures the model segment selection strategy for the UE
  • the model can be sent to the mobility management function first. Fragment selection strategy, and then the mobility management function carries the model fragment selection strategy in the NAS message to send the model fragment selection strategy to the UE.
  • the UE when the UE sends the identifier of the model segment to the base station, it can send the identifier of the model segment to the base station through an AS message. When the UE sends the identifier of the model segment to the core network device, it can send the identifier of the model segment to the core network device through a NAS message. ,and many more.
  • the data transmission between the first device, the second device, and the model control function, the specific message that can be used to carry the data to be transmitted can be based on the data transmission between the first device and the second device.
  • the specific type is determined. The foregoing embodiment is only described by taking a certain possible device type of the first device and the second device as an example. The specific data transmission can be based on the actual types of the first device and the second device. Obtained, I won't repeat it here.
  • FIG. 14 is a schematic flowchart of a model coordination method provided by an embodiment of this application. The method is applied to a third device, as shown in FIG. 14, including:
  • the third device is a device other than the first device, the second device, and the model control function, and is used to determine the first model segment for the first device.
  • the third device may obtain the model fragment selection strategy from the model control function or the cloud simulation server, and then determine the first model fragment for the first device according to the model fragment selection strategy, where the model fragment selection strategy includes the model fragment selection condition and the first device storage Correspondence between the model fragments.
  • the model segment selection strategy further includes the identifier of the preset model
  • the third device may determine at least one model segment of the preset model from the at least one model segment stored in the first device according to the identifier of the preset model, and The first model segment is determined in at least one model segment of the preset model.
  • the third device sends the first identifier to the first device, and the first device may determine that the model segment to be executed is the first model segment according to the first identifier.
  • the model segment selection strategy further includes the identification of the first device, and the third device may send the first identification to the first device according to the identification of the first device.
  • FIG. 15 is a schematic flowchart of a model coordination method provided by an embodiment of the application, as shown in FIG. 15, including:
  • S152 Determine a second model segment of the second device, where the second model segment is one of at least one model segment stored in the second device, and when the first model segment is used in the first device When executed and the second model segment is executed in the second device, part or all of the functions of the preset model are realized;
  • determining the second model segment of the second device includes:
  • the second model segment is determined according to a model segment selection strategy, where the model segment selection strategy includes a correspondence between model segment selection conditions and model segments stored in the second device.
  • model segment selection strategy further includes the correspondence between the model segment selection strategy and the model segments stored in the first device, and determining the first model segment of the first device includes:
  • the first model segment is determined according to the correspondence between the model segment selection strategy and the model segments stored in the first device.
  • determining the first model segment of the first device includes:
  • the first model segment is determined according to the first identifier.
  • the method further includes:
  • the method shown in FIG. 15 is a method on the model control function side, and the method on the model control function side has been introduced in the foregoing embodiment.
  • the foregoing embodiment please refer to the foregoing embodiment, which will not be repeated here.
  • FIG. 16 is a first structural diagram of a model coordination device provided by an embodiment of the application. As shown in FIG. 16, the model coordination device 160 includes a determining module 161, wherein:
  • the determining module 161 is configured to determine a first model segment from at least one model segment stored in the first device, where at least one model segment is stored in the first device for implementing part of the functions of the preset model.
  • first model fragment is executed and the second model fragment is executed in the second device, part or all of the functions of the preset model are realized, and the second model fragment is at least One of the model segments, at least one model segment stored in the second device is used to implement part of the functions of the preset model.
  • the determining module 161 is specifically configured to:
  • the first model segment is determined from at least one model segment stored in the first device.
  • the first information includes at least one of the following information:
  • the first identifier of the first model segment is the first identifier of the first model segment.
  • the first information includes a model segment selection strategy; the determining module 161 is specifically configured to:
  • the first model segment is determined from at least one model segment stored in the first device, and the model segment selection strategy includes a model segment selection condition and the storage of the first device Correspondence between model fragments.
  • the model segment selection strategy further includes an identifier of the preset model
  • the determining module 161 is specifically configured to:
  • the first model fragment is determined among at least one model fragment of the preset model stored in the first device .
  • the determining module 161 is specifically configured to:
  • the first model segment is determined from at least one model segment stored in the first device.
  • the determining module 161 is further configured to:
  • the model segment selection strategy further includes a correspondence between the model segment selection condition and the model segment stored in the second device, and the determining module 161 is further configured to:
  • the second model segment is determined according to the correspondence between the model segment selection condition and the model segment stored in the second device.
  • model segment selection strategy further includes the identification of the second device, and the determining module 161 is further configured to:
  • the determining module 161 is further configured to:
  • the model segment selection strategy is received from the model control function.
  • the first information includes the first identifier of the first model fragment; the determining module 161 is specifically configured to:
  • the first model segment is determined from at least one model segment stored in the first device.
  • the first information comes from the second device or model control function.
  • the first device is a first terminal device or a first network device
  • the second device is a second terminal device or a second network device.
  • the first network device is a base station or a core network device
  • the second network device is a base station or a core network device.
  • model coordination apparatus provided in the embodiments of the present application can execute the technical solutions shown in the foregoing method embodiments, and the implementation principles and beneficial effects are similar, and details are not described herein again.
  • FIG. 17 is a second structural diagram of the model coordination device provided by an embodiment of the application. As shown in FIG. 17, the model coordination device 170 includes a determining module 171 and a sending module 172, wherein:
  • the determining module 171 is configured to determine a first model segment, where the first model segment is one of at least one model segment stored in the first device, and when the first model segment is executed in the first device, preset Part of the function of the model is realized;
  • the sending module 172 is configured to send the first identifier of the first model segment to the first device.
  • the determining module 171 is specifically configured to:
  • the first model segment is determined according to a model segment selection strategy, and the model segment selection strategy includes a correspondence between model segment selection conditions and model segments stored in the first device.
  • the model segment selection strategy further includes an identifier of the preset model
  • the determining module 171 is specifically configured to:
  • the first Model fragment in at least one model segment of the preset model stored in the first device, the first Model fragment.
  • model segment selection strategy further includes the identification of the first device, and the sending module 172 is specifically configured to:
  • the first identifier is sent to the first device.
  • the determining module 171 is further configured to:
  • the model segment selection strategy is received from the model control function.
  • model coordination apparatus provided in the embodiments of the present application can execute the technical solutions shown in the foregoing method embodiments, and the implementation principles and beneficial effects are similar, and details are not described herein again.
  • FIG. 18 is the third structural diagram of the model coordination device provided by the embodiment of the application.
  • the model coordination device 180 includes a first processing module 181, a second processing module 182, and a sending module 183, in which:
  • the first processing module 181 is configured to determine a first model segment of a first device, where the first model segment is one of at least one model segment stored in the first device;
  • the second processing module 182 is configured to determine a second model segment of the second device.
  • the second model segment is one of at least one model segment stored in the second device. When the first model segment is in the When the first device is executed and the second model fragment is executed in the second device, part or all of the functions of the preset model are implemented;
  • the sending module 183 is configured to send the second identifier of the second model fragment to the second device.
  • the second processing module 182 is specifically configured to:
  • the second model segment is determined according to a model segment selection strategy, where the model segment selection strategy includes a correspondence between model segment selection conditions and model segments stored in the second device.
  • model segment selection strategy further includes a correspondence between model segment selection conditions and model segments stored in the first device, and the first processing module 181 is specifically configured to:
  • the first model segment is determined according to the correspondence between the model segment selection condition and the model segment stored in the first device.
  • the first processing module 181 is specifically configured to:
  • the first model segment is determined according to the first identifier.
  • the first processing module 181 is further configured to:
  • model coordination apparatus provided in the embodiments of the present application can execute the technical solutions shown in the foregoing method embodiments, and the implementation principles and beneficial effects are similar, and details are not described herein again.
  • FIG. 19 is a schematic structural diagram of a model coordination device provided by an embodiment of the application.
  • the model coordination device 190 may include: a transceiver 191, a memory 192, and a processor 193.
  • the transceiver 191 may include: a transmitter and/or a receiver.
  • the transmitter can also be referred to as a transmitter, a transmitter, a transmitting port, or a transmitting interface
  • the receiver can also be referred to as a receiver, a receiver, a receiving port, or a receiving interface, and similar descriptions.
  • the transceiver 191, the memory 192, and the processor 193 are connected to each other through a bus 194.
  • the memory 192 is used to store program instructions
  • the processor 193 is configured to execute the program instructions stored in the memory, so as to enable the terminal device 190 to execute any of the aforementioned model coordination methods.
  • the receiver of the transceiver 191 can be used to perform the receiving function of the model coordination device in the above model coordination method.
  • FIG. 20 is a schematic structural diagram of a model coordination device provided by an embodiment of the application.
  • the model coordination device 200 may include: a transceiver 201, a memory 202, and a processor 203.
  • the transceiver 201 may include: a transmitter and/or a receiver.
  • the transmitter can also be referred to as a transmitter, a transmitter, a transmission port or a transmission interface and other similar descriptions
  • the receiver can also be referred to as a receiver, a receiver, a reception port or a reception interface and other similar descriptions.
  • the transceiver 201, the memory 202, and the processor 203 are connected to each other through a bus 204.
  • the memory 202 is used to store program instructions
  • the processor 203 is configured to execute the program instructions stored in the memory to enable the model coordination device 200 to execute any of the model coordination methods shown above.
  • the receiver of the transceiver 201 can be used to perform the receiving function of the model coordination device in the above model coordination method.
  • An embodiment of the present application provides a computer-readable storage medium that stores a computer-executable instruction, and when the computer-executable instruction is executed by a processor, it is used to implement the aforementioned model coordination method.
  • An embodiment of the present application provides a computer-readable storage medium that stores a computer-executable instruction, and when the computer-executable instruction is executed by a processor, it is used to implement the aforementioned model coordination method.
  • the embodiments of the present application may also provide a computer program product, which can be executed by a processor, and when the computer program product is executed, it can implement the model coordination method executed by any of the model coordination devices shown above.
  • the transmission device, computer-readable storage medium, and computer program product of the embodiment of the present application can execute the model coordination method executed by the model coordination device described above.
  • the model coordination device described above For the specific implementation process and beneficial effects, refer to the foregoing, and will not be repeated here.
  • the disclosed system, device, and method can be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division, and there may be other divisions in actual implementation, for example, multiple units or components may be combined or It can be integrated into another system, or some features can be ignored or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit can be implemented in the form of hardware or software functional unit.
  • the aforementioned computer program can be stored in a computer readable storage medium.
  • the computer program When the computer program is executed by the processor, it realizes the steps including the foregoing method embodiments; and the foregoing storage medium includes: ROM, RAM, magnetic disk, or optical disk and other media that can store program codes.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

一种模型协调方法及装置，该方法应用于第一设备，第一设备中存储至少一个模型片段，用于实现预设模型的部分功能，该方法包括：在第一设备中存储的至少一个模型片段中确定第一模型片段，其中，当第一模型片段被执行且第二模型片段在第二设备中被执行时，预设模型的部分功能或者全部功能被实现，第二模型片段为第二设备中存储的至少一个模型片段中的一个，第二设备中存储的至少一个模型片段用于实现预设模型的部分功能。该方法能够在模型被划分至多个设备进行执行时，保证模型片段在各个设备之间的协调一致。

Description

模型协调方法及装置 技术领域
本申请涉及人工智能技术领域，尤其涉及一种模型协调方法及装置。
背景技术
人工智能和大数据分析技术的发展，人工智能(Artefact Intelligence，简称AI)/机器学习(Machine Learning，简称ML)在智能化数据分析领域得到广泛应用。移动通信领域也越来越多的引入AI/ML模型进行语音识别、图像处理等智能化操作，使得移动终端能够服务于更多的业务场景。
AI/ML模型的执行对设备的要求较高，由于移动终端的计算能力、存储能力等方面的限制，因此可以将AI/ML模型分成多个模型片段，将部分模型片段转移到其他设备上，多个设备共同完成AI/ML模型的执行，从而分担移动终端的计算工作量。但是，由于AI/ML模型进行模型片段划分存在多种划分方式，模型片段在各个设备之间的不能协调一致，使得模型执行的效果并不理想。
发明内容
本申请实施例提供一种模型协调方法及装置，以解决模型被划分为多个模型片段并被配置至多个设备共同执行时，模型片段在各个设备之间的协调问题。
第一方面，本申请实施例提供一种模型协调方法，应用于第一设备，所述第一设备中存储至少一个模型片段，其中所述第一设备中存储的至少一个模型片段用于实现预设模型的部分功能，所述方法包括：
在所述第一设备中存储的至少一个模型片段中确定第一模型片段，其中，当所述第一模型片段被执行且第二模型片段在第二设备中被执行时，所述预设模型的部分功能或者全部功能被实现，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，所述第二设备中存储的至少一个模型片段用于实现所述预设模型的部分功能。
第二方面，本申请实施例提供一种模型协调方法，应用于第三设备，所述方法包括：
确定第一模型片段，所述第一模型片段为第一设备中存储的至少一个模型片段中的一个，当所述第一模型片段在第一设备中被执行时，预设模型的部分功能被实现；
向第一设备发送所述第一模型片段的第一标识。
第三方面，本申请实施例提供一种模型协调方法，应用于模型控制功能，所述方法包括：
确定第一设备的第一模型片段，其中，所述第一模型片段为所述第一设备中存储的至少一个模型片段中的一个；
确定第二设备的第二模型片段，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，当所述第一模型片段在所述第一设备中被执行且所述第二模型片段在所述第二设备中被执行时，预设模型的部分功能或者全部功能被实现；
向所述第二设备发送所述第二模型片段的第二标识。
第四方面，本申请实施例提供一种模型协调装置，包括：
确定模块，用于在第一设备中存储的至少一个模型片段中确定第一模型片段，其中，所述第一设备中存储至少一个模型片段，用于实现预设模型的部分功能，当所述第一模型片段被执行且第二模型片段在第二设备中被执行时，所述预设模型的部分功能或者全部功能被实现，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，所述第二设备中存储的至少一个模型片段用于实现所述预设模型的部分功能。
第五方面，本申请实施例提供一种模型协调装置，包括：
确定模块，用于确定第一模型片段，所述第一模型片段为第一设备中存储的至少一个模型片段中的一个，当所述第一模型片段在第一设备中被执行时，预设模型的部分功能被实现；
发送模块，用于向第一设备发送所述第一模型片段的第一标识。
第六方面，本申请实施例提供一种模型协调装置，包括：
第一处理模块，用于确定第一设备的第一模型片段，其中，所述第一模型片段为所述第一设备中存储的至少一个模型片段中的一个；
第二处理模块，用于确定第二设备的第二模型片段，所述第二模型片段为所述第二设备中存储的至 少一个模型片段中的一个，当所述第一模型片段在所述第一设备中被执行且所述第二模型片段在所述第二设备中被执行时，预设模型的部分功能或者全部功能被实现；
发送模块，用于向所述第二设备发送所述第二模型片段的第二标识。
第七方面，本申请实施例提供一种模型协调设备，包括：收发器、处理器、存储器；
所述存储器存储计算机执行指令；
所述处理器执行所述存储器存储的计算机执行指令，使得所述处理器执行如第一方面任一项所述的模型协调方法，或者，使得所述处理器执行如第二方面任一项所述的模型协调方法。
第八方面，本申请实施例提供一种模型协调设备，包括：收发器、处理器、存储器；
所述存储器存储计算机执行指令；
所述处理器执行所述存储器存储的计算机执行指令，使得所述处理器执行如第三方面任一项所述的模型协调方法。
第九方面，本申请实施例提供一种计算机可读存储介质，所述计算机可读存储介质中存储有计算机执行指令，当所述计算机执行指令被处理器执行时用于实现如第一方面、第二方面或第三方面任一项所述的模型协调方法。
本申请实施例提供的模型协调方法，第一设备在存储的至少一个模型片段中确定第一模型片段，当第一模型片段被执行且第二模型片段在第二设备被执行时，能够实现预设模型的部分或者全部功能，从而实现了模型片段在各个设备之间的有效协调。
附图说明
图1为本申请实施例适用的一种网络系统架构图；
图2为本申请实施例提供的模型分离示意图；
图3为本申请实施例提供的一种应用场景示意图；
图4为本申请实施例提供的模型协调方法的流程示意图；
图5为本申请实施例提供的模型标识示意图；
图6为本申请实施例提供的模型协调方法的信令图一；
图7为本申请实施例提供的模型协调示意图一；
图8为本申请实施例提供的模型协调方法的信令图二；
图9为本申请实施例提供的模型协调示意图二；
图10为本申请实施例提供的模型协调方法的信令图三；
图11为本申请实施例提供的模型协调示意图三；
图12为本申请实施例提供的模型协调方法的信令图四；
图13为本申请实施例提供的模型协调示意图四；
图14为本申请实施例提供的模型协调方法的流程示意图；
图15为本申请实施例提供的模型协调方法的流程示意图；
图16为本申请实施例提供的模型协调装置的结构示意图一；
图17为本申请实施例提供的模型协调装置的结构示意图二；
图18为本申请实施例提供的模型协调装置的结构示意图三；
图19为本申请实施例提供的模型协调设备的结构示意图；
图20为本申请实施例提供的模型协调设备的结构示意图。
具体实施方式
为了便于理解，首先对本申请涉及的概念进行解释说明。
终端设备：可以为包含无线收发功能、且可以与网络设备配合为用户提供通讯服务的设备。具体地，终端设备可以指用户设备(User Equipment，UE)、接入终端、用户单元、用户站、移动站、移动台、远方站、远程终端、移动设备、用户终端、终端、无线通信设备、用户代理或用户装置。例如，终端设备可以是蜂窝电话、无绳电话、会话启动协议(Session Initiation Protocol，SIP)电话、无线本地环路(Wireless Local Loop，WLL)站、个人数字处理(Personal Digital Assistant，PDA)、具有无线通信功能的手持设备、计算设备或连接到无线调制解调器的其它处理设备、车载设备、可穿戴设备，未来5G网络或5G之后的未来演进的公用陆地移动通信网络(public land mobile network，PLMN)中的终端设备等，本申请实施例对此并不限定。
作为示例而非限定，在本申请实施例中，该终端设备还可以是可穿戴设备。可穿戴设备也可以称为穿戴式智能设备，是应用穿戴式技术对日常穿戴进行智能化设计、开发出可以穿戴的设备的总称，如眼镜、手套、手表、服饰及鞋等。可穿戴设备即直接穿在身上，或是整合到用户的衣服或配件的一种便携式设备。可穿戴设备不仅仅是一种硬件设备，更是通过软件支持以及数据交互、云端交互来实现强大的 功能。广义穿戴式智能设备包括功能全、尺寸大、可不依赖智能手机实现完整或者部分的功能，例如：智能手表或智能眼镜等，以及只专注于某一类应用功能，需要和其它设备如智能手机配合使用，如各类进行体征监测的智能手环、智能首饰等。
此外，在本申请实施例中，终端设备还可以是物联网(internet of things，IoT)系统中的终端设备，IoT是未来信息技术发展的重要组成部分，其主要技术特点是将物品通过通信技术与网络连接，从而实现人机互连，物物互连的智能化网络。在本申请实施例中，IOT技术可以通过例如窄带(narrow band)NB技术，做到海量连接，深度覆盖，终端省电。
网络设备：网络设备可以是用于与终端设备进行通信的设备，例如，可以是全球移动通信系统(Global System for Mobile Communication，GSM)或码分多址(Code Division Multiple Access，CDMA)通信系统中的基站(Base Transceiver Station，BTS)，也可以是宽带码分多址(Wideband Code Division Multiple Access，WCDMA)系统中的基站(NodeB，NB)，还可以是LTE系统中的演进型基站(Evolutional Node B，eNB或eNodeB)，或者该网络设备可以为中继站、接入点、车载设备、可穿戴设备以及未来5G网络或5G之后的网络中的网络侧设备或未来演进的公共陆地移动网络(Public Land Mobile Network，PLMN)网络中的网络设备等。
本申请实施例中涉及的网络设备也可称为无线接入网(Radio Access Network，RAN)设备。RAN设备与终端设备连接，用于接收终端设备的数据并发送给核心网设备。RAN设备在不同通信系统中对应不同的设备，例如，在2G系统中对应基站与基站控制器，在3G系统中对应基站与无线网络控制器(Radio Network Controller，RNC)，在4G系统中对应演进型基站(Evolutional Node B，eNB)，在5G系统中对应5G系统，如NR中的接入网设备(例如gNB，集中单元CU，分布式单元DU)。
AS：Access Stratum，接入层；
NAS：None Access Stratum，非接入层；
RRC：Radio Resource Control，无线资源控制。
下面对本申请的相关技术背景进行说明：
随着人工智能、大数据分析技术的发展，AI/ML模型开始被广泛应用于智能化数据分析。在通信领域，终端设备也越来越多的引入AI/ML模型模型进行例如语音识别、图像处理、用户行为预测等智能化操作，从而使得终端设备可以服务于更多的业务场景。
图1为本申请实施例适用的一种网络系统架构图，如图1所示，该网络架构可以为5G网络架构，5G系统也称为新无线通信系统、新接入技术(NewRadio，NR)或者下一代移动通信系统。
5G系统中的接入网可以是无线接入网(radio access network，(R)AN)，5G系统中的(R)AN设备可以由多个5G-(R)AN节点组成，该5G-(R)AN节点可以包括：非3GPP的接入网络如WiFi网络的接入点(access point，AP)、下一代基站(可统称为新一代无线接入网节点(NG-RAN node)，其中，下一代基站包括新空口基站(NR nodeB，gNB)、新一代演进型基站(NG-eNB)、中心单元(central unit，CU)和分布式单元(distributed unit，DU)分离形态的gNB等)、收发点(transmission receive point，TRP)、传输点(transmission point，TP)或其它节点。
如图1所示，5G核心网(5G core/new generation core，5GC/NGC)包括接入和移动性管理功能(Access and Mobility Management Function，AMF)网元、会话管理功能(Session Management Function，SMF)网元、用户面功能(User Plane Function，UPF)网元、鉴权服务器功能(Authentication Server Function，AUSF)网元、策略控制功能(Policy Control Function，PCF)网元、应用功能(Application Function，AF)网元、统一数据管理功能(unified data management，UDM)网元、网络切片选择功能(Network Slice Selection Function，NSSF)网元等多个功能单元。
AMF网元主要负责移动性管理、接入管理等服务。SMF网元主要负责会话管理、UE地址管理和分配、动态主机配置协议功能、用户面功能的选择和控制等。UPF主要负责对外连接到数据网络(data network，DN)以及用户面的数据包路由转发、报文过滤、执行服务质量(quality of service，QoS)控制相关功能等。AUSF主要负责对终端设备的认证功能等。PCF网元主要负责为网络行为管理提供统一的策略框架、提供控制面功能的策略规则、获取与策略决策相关的注册信息等。需要说明的是，这些功能单元可以独立工作，也可以组合在一起实现某些控制功能，如对终端设备的接入鉴权、安全加密、位置注册等接入控制和移动性管理功能，以及用户面传输路径的建立、释放和更改等会话管理功能。
进一步的，在图1示例的网络架构图中，还提供一种模型控制功能，该模型控制功能为核心网设备中的一个网元，用于对模型片段在各个设备之间共同执行时，控制模型片段之间的协调。该模型控制功能可以是核心网设备中的一个新的网元，也可以是将模型控制功能对模型片段的协调控制集成到目前的核心网设备中的某一个网元中，例如AMF网元、SMF网元等等，此时，集成了对模型片段的协调控制 的网元即为本申请实施例中的模型控制功能。
5GC中各功能单元之间可以通过下一代网络(next generation，NG)接口进行通信，如：UE可以通过NG接口1(简称N1)与AMF网元进行控制面消息的传输，RAN设备可以通过NG接口3(简称N3)与UPF建立用户面数据传输通道，AN/RAN设备可以通过NG接口2(简称N2)与AMF网元建立控制面信令连接，UPF可以通过NG接口4(简称N4)与SMF网元进行信息交互，UPF可以通过NG接口6(简称N6)与数据网络DN交互用户面数据，AMF网元可以通过NG接口11(简称N11)与SMF网元进行信息交互，SMF网元可以通过NG接口7(简称N7)与PCF网元进行信息交互，AMF网元可以通过NG接口12(简称N12)与AUSF进行信息交互。需要说明的是，图1仅为示例性架构图，除图1中所示功能单元之外，该网络架构还可以包括其他功能单元。
图1所示网络架构为基于参考点网络架构，且该网络架构为非漫游场景下的网络架构，当然本申请的方法也可以应用在漫游场景下，并且网络架构也不限于基于参考点的网络架构，也可以采用基于服务化接口的网络架构。
在上述系统架构上，由于终端设备在计算能力、存储能力、电池能力的局限，而AI/ML模型对设备要求较高，若采用终端设备独立运行AI/ML模型，对终端设备的能力是一个较大的挑战。即使终端设备能够独立完成AI/ML模型的执行，其花费的时间可能也较长。基于此，可以将AI/ML模型划分为两个或两个以上的模型片段，由两个或两个以上的设备分别执行相应的片段。例如，可以将部分模型片段划分到网络设备上，由网络设备和终端设备共同执行AI/ML模型。
图2为本申请实施例提供的模型分离示意图，如图2所示，针对的是一个AI/ML模型的划分，该AI/ML模型为一个图像识别模型，能够对图像中的事物进行识别。本申请实施例中，设定的是模型已经预先训练完成，图2中将图像识别模型划分为三个模型片段，并分别配置到三个设备上进行执行。如图2中所示，这三个设备分别是终端设备20、第一网络设备21和第二网络设备22，每个设备只执行自身的模型片段。
若三个设备执行的模型片段是协调的，则任意一个或两个设备能够实现该图像识别模型的部分功能，三个设备在执行完各自的模型片段后，能够实现该图像识别模型的全部功能，例如图2中，最后得到的模型的输出为“tree”，表示在输出一张图像后，经过该模型的处理，该模型识别出该图像上的事物为一棵树。
可以理解的是，一个图像识别模型中可能包括多个不同的部分，例如可以包括若干个卷积层、若干个池化层、若干个上采样层、若干个下采样层等等，模型中的各个部分是有机排列的，且在输入数据后，每个部分可能均会输出中间结果，上一个部分输出的结果可能是下一个部分的输入，因此，如图2中所示，终端设备20执行对应的模型片段后，其得到的输出为一个中间结果，该中间结果为第一网络设备21执行的模型片段的输入。
基于此，首先，各个设备执行各自的模型片段是有先后顺序的，例如在图2中，只有终端设备20执行完其模型片段，得到中间结果后作为第一网络设备21的输入，第一网络设备21才能执行其对应的模型片段并得到相应的中间数据，然后第一网络设备21输出的中间数据是第二网络设备22执行的模型片段的输入，等等。
其次，基于终端设备在不同的场景下的计算能力有差异，可能会对模型有不同的划分方式，而各设备之间执行模型片段时无法实现模型之间的协调。
例如，假设图2中的图像识别模型由预处理层-卷积层-上采样层-卷积层-池化层-卷积层-激活层-全连接层组成，其划分方式例如可以为：
第一种，模型片段A1(包括预处理层-卷积层)、模型片段B1(包括上采样层-卷积层-池化层)、模型片段C1(包括卷积层-激活层-全连接层)；
第二种，模型片段A2(包括预处理层)、模型片段B2(包括卷积层-上采样层-卷积层-池化层)、模型片段C2(包括卷积层-激活层-全连接层)；
第三种，模型片段A3(包括预处理层-卷积层-上采样层)、模型片段B3(包括卷积层-池化层-卷积层-激活层)、模型片段C3(包括全连接层)。
基于以上三种划分方式，可以将模型片段A1、模型片段A2和模型片段A3配置到终端设备20，将模型片段B1、模型片段B2和模型片段B3配置到第一网络设备21，将模型片段C1、模型片段C2和模型片段C3配置到第二网络设备22。
然而，由于模型的多种划分方式，模型之间的分拆和协调是一个问题。例如，当终端设备20选择执行的模型片段为A1时，第一网络设备21可能选择的执行片段为B2，第二网络设备22可能选择的执行片段为C3，此时三个设备总共执行的模型为预处理层-卷积层-卷积层-上采样层-卷积层-池化层-全 连接层，与原本的图像识别模型是不同的。
当然，如果仅对模型进行一种划分，例如仅包括第一种划分方式，则三个设备中均只包括该图像识别模型的一个模型片段，模型执行可以保证协调，但是一种划分方式无法满足多种应用场景的需求。例如，将模型划分为模型片段A(占模型的80％)、模型片段B(占模型的10％)和模型片段C(占模型的10％)，并分别依次配置至终端设备20、第一网络设备21和第二网络设备22。而此时，终端设备20的电量较低，若终端设备20执行模型片段A，此时的电量不足以支撑终端设备20完成模型片段A的执行。若此时还包括另一种划分方式，模型划分为模型片段D(占模型的10％)、模型片段E(占模型的30％)和模型片段F(占模型的60％)，并分别依次配置至终端设备20、第一网络设备21和第二网络设备22，终端设备20可在电量较低的情况下选择执行模型片段D。
因此，对模型的多种划分是必要的。但是，如何保证终端设备20在选择了模型片段A1后，第一网络设备21选择执行的模型片段为B1，第二网络设备22选择执行的模型片段为C1，是本申请需要解决的问题。
首先对本申请的应用场景进行介绍。
图3为本申请实施例提供的一种应用场景示意图，如图3所示，包括云端模拟服务器31、模型控制功能32、第一设备33和第二设备34。AI/ML模型已经训练完成，云端模型服务器31将AI/ML模型进行多种方式的拆分，形成不同的模型片段集合并配置到多个AI/ML模型端点，AI/ML模型端点即为执行AI/ML模型中的模型片段的设备。例如，将AI/ML模型端点拆分成模型片段并分别配置到第一设备33和第二设备34后，第一设备33和第二设备34均为AI/ML模型端点。
该AI/ML模型进行了多种方式的拆分，图3中示例了两种拆分方式，将AI/ML模型拆分为模型片段A1+模型片段B1，还可以拆分成模型片段A2+模型片段B2，然后将模型片段A1和模型片段A2配置到第一设备33，将模型片段B1和模型片段B2配置到第二设备34。
当第一设备33确定执行的模型片段为模型片段A1时，需要第二设备34执行的模型片段为模型片段B1，当第一设备33确定执行的模型片段为模型片段A2时，需要第二设备34执行的模型片段为模型片段B2，本申请实施例中，是通过模型片段选择策略来保证上述选择方案的。模型片段选择策略是云端模拟服务器31确定的，云端模拟服务器31可以将模型片段选择策略配置到模型控制功能32，也可以将模型片段选择策略配置到第一设备33或第二设备34。
然后，可能由第一设备33、或第二设备34、或模型控制功能32来确定各个设备确定执行的模型片段。
在上述介绍的内容的基础上，下面对本申请所提供的模型协调方法进行说明，首先结合图4进行说明，图4为本申请实施例提供的模型协调方法的流程示意图，如图4所示，该方法包括：
S41，在所述第一设备中存储的至少一个模型片段中确定第一模型片段。
其中，第一设备中存储了至少一个模型片段，第二设备中也存储了至少一个模型片段，第一设备在存储的这至少一个模型片段中确定第一模型片段时，第二设备在存储的至少一个模型片段中确定的是第二模型片段。第一模型片段和第二模型片段均为预设模型的一部分，当第一设备执行第一模型片段且第二设备执行第二模型片段时，预设模型的部分功能或全部功能被实现。第一设备可以为终端设备或网络设备，第二设备也可以为终端设备或网络设备。
第一设备和第二设备中存储的模型片段可以是对预设模型进行划分之后预先分别存储到第一设备和第二设备中的，对预设模型的划分可以有多种，例如，可以将预设模型划分为模型片段A1和模型片段B1，也可以将预设模型划分为模型片段A2和模型片段B2，其中，模型片段A1和模型片段A2为第一设备侧的模型片段，可存储至第一设备，模型片段B1和模型片段B2为第二设备侧的模型片段，可存储至第二设备。
模型片段A1和模型片段B1是相匹配的，当模型片段A1和模型片段B1被执行时，能够实现预设模型的全部功能，模型片段A2和模型片段B2是相匹配的，当模型片段A2和模型片段B2被执行时，能够实现预设模型的全部功能。本申请实现的方案是，当第一设备选择了模型片段A1时，第二设备选择的模型片段是B1，当第一设备选择了模型片段A2时，第二设备选择的模型片段是B2，从而实现模型的协调划分。当预设模型被划分为两个模型片段并分别由第一设备和第二设备执行时，实现预设模型的全部功能；当预设模型被划分为两个以上的模型片段并分别由第一设备和第二设备执行各自的模型片段时，实现预设模型的部分功能。
本申请实施例提供的模型协调方法，第一设备在存储的至少一个模型片段中确定第一模型片段，当第一模型片段被执行且第二模型片段在第二设备被执行时，能够实现预设模型的部分或者全部功能，从而实现了模型的有效协调。
本申请实施例中，预设模型是已经训练完成的模型，以预设模型为AI/ML模型为例，该训练AI/ML模型训练完成后，数据网络中的云端模型服务器可将AI/ML模型进行多种方式的拆分，形成不同的模型片段集合并配置到多个AI/ML端点执行。本申请实施例中的AI/ML端点即为执行AI/ML模型的设备，例如当预设模型为AI/ML时，第一设备和第二设备均为AI/ML端点，且第一设备和第二设备互为对端AI/ML端点。
可选的，第一设备为第一终端设备或第一网络设备，第二设备为第二终端设备或第二网络设备，其中第一网络设备可以为基站或核心网设备，第二网络设备可以为基站或核心网设备。
例如一个AI/ML端点为移动终端，另外一个AI/ML端点为移动网络中的网络节点，包括并不限于核心网节点或者基站，其中核心网节点包括并不限于移动管理功能、会话管理功能、用户面功能或者核心网中专门设置的网络数据分析节点等。
在一种可能的实现方式中，确定第一模型片段的方式是，获取第一信息，然后根据第一信息在第一设备中存储的至少一个模型片段中确定第一模型片段。其中，第一信息包括模型片段选择策略和第一模型片段的第一标识中的至少一个。模型片段选择策略中包括模型片段选择条件和第一设备存储的模型片段之间的对应关系，若第一信息中包括模型片段选择策略，则可以根据模型片段选择条件和第一设备存储的模型片段之间的对应关系来确定第一模型片段。若第一信息中包括的是第一模型片段的第一标识，则可以直接根据第一标识确定第一模型片段。
当第一信息包括模型片段选择策略时，可选的，模型片段选择策略中还包括预设模型的标识，因此，可以根据预设模型的标识确定第一设备中存储的、预设模型的至少一个模型片段。图5为本申请实施例提供的模型标识示意图，如图5所示，包括第一预设模型和第二预设模型。第一预设模型可划分为模型片段A1+模型片段B1，也可以划分为模型片段A2+模型片段B2。第二预设模型可划分为模型片段A1+模型片段B1，也可以划分为模型片段A2+模型片段B2。然后，第一预设模型的模型片段A1和第二预设模型的模型片段A1被配置到终端设备51中，第一预设模型的模型片段B1和第二预设模型的模型片段B1被配置到网络设备52中。
图5中可以是终端设备51为第一设备，也可以是网络设备52为第一设备。以第一设备为终端设备51为例，其中存储的两个模型片段均为模型片段A1，但是是属于两个不同的预设模型下想模型片段。此时，需要根据模型片段选择策略中包括的预设模型的标识来确定是哪个模型片段A1。例如，此时需要执行的是第一预设模型，则模型片段选择策略中可以包括第一预设模型的标识，然后根据第一预设模型的标识，终端设备51可以获知确定的第一模型片段为第一预设模型下的模型片段A1。
可选的，当第一设备根据模型片段选择策略确定了第一模型片段后，第一设备可以将第一模型片段的第一标识发送给第二设备，第二设备根据第一标识和模型片段选择策略确定第二模型片段，其中，模型片段选择策略中包括模型片段选择条件和第一设备存储的模型片段之间的对应关系；
第一设备也可以根据模型片段选择策略在确定了第一模型片段之后，将第一模型片段的第一标识发送给模型控制功能，模型控制功能根据第一标识和模型片段选择策略为第二设备确定第二模型片段，并向第二设备发送第二模型片段的第二标识，其中，模型片段选择策略中包括模型片段选择条件和第二设备存储的模型片段之间的对应关系；
第一设备也可以根据模型片段选择策略确了第一模型片段，根据模型片段选择策略为第二设备确定第二模型片段，并向第二设备发送第二模型片段的第二标识，从而第二设备在接收到第二标识后，确定第二模型片段，其中，模型片段选择策略中还包括模型片段选择条件和第二设备存储的模型片段之间的对应关系。
当第二设备确定了第二模型片段时，第二设备可以向第一设备发送第二模型片段的第二标识，第一设备根据第二标识和模型片段选择策略确定第一模型片段，其中，模型片段选择策略中包括模型片段选择条件和第一设备存储的模型片段之间的对应关系；
第二设备也可以向模型控制功能发送第二模型片段的第二标识，模型控制功能根据模型片段选择策略和第二标识，为第一设备确定第一模型片段，并向第一设备发送第一模型片段的第一标识，第一设备根据第一标识确定第一模型片段，其中，模型片段选择策略中包括模型片段选择条件和第一设备存储的模型片段之间的对应关系；
第二设备也可以根据模型片段选择策略在确定了第二模型片段之后，根据模型片段选择策略为第一设备确定第一模型片段，并向第一设备发送第一模型片段的第一标识，从而第一设备在接收到第一标识后，确定第一模型片段，其中，模型片段选择策略中还包括模型片段选择条件和第一设备存储的模型片段之间的对应关系。
在本申请实施例中，若第一设备或第二设备中存储了模型片段选择策略，则第一设备中存储的模型 片段选择策略是从模型控制功能接收的，第二设备中存储的模型片段选择策略也是从模型控制功能接收的。本申请所有实施例中，A向B发送某信息，或A接收B的信息，或A从B接收某信息，均是指某信息的传输是从A至B，或从B至A，但并不代表是直接由A发送给B或者由B直接发送给A。由于第一设备可能为第一终端设备或第一网络设备，第二设备可能为第二终端设备或第二网络设备，而网络设备还可能基站或核心网设备，因此，需要根据设备的实际类型来确定具体的传输方式。
例如，第一设备从模型控制功能接收模型片段选择策略，若第一设备为终端设备，模型控制功能是核心网中的一个网元，两者之间的数据传输是需要例如基站等无线接入网设备来进行转发的，此时第一设备从模型控制功能接收模型片段选择策略指的是第一设备接收的模型片段选择策略是来自于模型控制功能的，但不代表是直接由模型控制功能发送给终端设备的。
下面将以第一设备和第二设备这两个设备中其中一个为UE，另一个为网络节点为例来对本申请的方案进行举例说明，其中，UE可以为第一设备，也可以为第二设备，网络节点可以为第一设备，也可以为第二设备。
图6为本申请实施例提供的模型协调方法的信令图一，如图6所示，包括：
S61，服务器向模型控制功能发送模型片段选择策略。
本申请实施例中的模型控制功能为核心网设备中的一个网元，服务器通过模型控制功能向UE和网络节点分别发送模型片段选择策略，其中，服务器也可以直接向UE或者网络节点发送模型片段选择策略，本申请实施例中以服务器同模型控制功能向UE和网络设备发送模型片段选择策略为例。
S62，模型控制功能向UE和网络节点分别发送模型片段选择策略。
模型控制功能在获取到模型片段选择策略后，分别向UE和网络节点发送模型片段选择策略。可选的，模型控制功能向UE发送UE端模型片段选择策略，UE端模型片段选择策略包括模型片段选择条件和UE端存储的模型片段之间的对应关系；模型控制功能向网络节点发送网络节点端模型片段选择策略，网络节点端模型片段选择策略包括模型片段选择条件和网络节点端存储的模型片段之间的对应关系。
模型控制功能向UE配置模型片段选择策略，可以是模型控制功能将模型片段选择策略发送给移动管理功能，然后再由移动管理功能通过NAS消息将模型片段选择策略发送给UE，其中模型片段选择策略包括在NAS消息中。可以理解的是，移动管理功能向UE发送NAS消息时，是通过无线接入网设备(例如某个基站)的转发实现向UE发送的。
S63，UE获取UE侧模型片段集合和模型片段选择策略。
当对预设模型有多种划分时，会将这多种划分得到的模型片段分别配置到UE端和网络节点端。图7为本申请实施例提供的模型协调示意图一，如图7所示，包括模型控制功能70、UE71和基站72。服务器为UE71配置的模型片段为模型片段A1、模型片段A2和模型片段A3，服务器为网络节点配置的模型片段为模型片段B1、模型片段B2和模型片段B3。图7中的网络节点为基站72，可选的，网络节点还可以是核心网设备。
模型片段选择策略包括模型片段选择条件以及存储的模型片段之间的对应关系，模型片段选择条件例如可以为UE的电量、UE的位置(例如小区标识)、UE的空口信号质量。模型片段选择条件除了可以为UE的相关信息外，还可以是网络节点的负载，还可以是当前时间等与第一设备和第二设备无关的信息。本申请实施例中对模型片段选择条件不限。
S64，网络节点获取网络节点侧模型片段集合和模型片段选择策略。
网络节点获取的网络节点侧模型片段集合中的模型片段是与UE侧的模型片段集合中的模型片段相匹配的，例如在图7中，模型片段A1与模型片段B1匹配，模型片段A2与模型片段B2匹配，模型片段A3与模型片段B3匹配。
S65，UE确定UE侧模型片段。
在图7中，UE71从模型控制功能70获取模型片段选择策略，然后根据模型片段选择策略确定UE侧模型片段。例如模型片段选择条件为UE的电量，当UE的电量低于30％时，选择模型片段A1，当UE的电量高于30％且低于50％时，选择模型片段A2，当UE的电量高于50％且低于100％时，选择模型片段A3。此时UE获知自身电量为25％，则确定执行的模型片段为模型片段A1。当UE为第一设备时，模型片段A1即为第一模型片段，A1为第一模型片段的第一标识；当UE为第二设备时，模型片段A1即为第二模型片段，A1为第二模型片段的第二标识。
S66，UE向网络节点发送UE侧模型片段标识。
当UE确定了执行的模型片段为模型片段A1时，向网络节点发送模型片段A1的标识，例如在图7中，UE71向基站72发送标识A1。在图7的示例中，网络节点为基站，则UE可以直接和基站进行数据的传输。进一步的，若网络节点为核心网设备，则UE也可以和核心网设备进行数据的传输，但是 UE和核心网设备进行数据传输需要例如基站等无线接入网设备作为中继节点进行转发。
当网络节点为核心网设备时，UE可以通过NAS消息将标识A1上报给核心网设备，当网络节点为基站时，UE可以通过AS消息将标识A1上报给基站，AS消息例如可以为RRC消息。
S67，网络节点确定网络节点侧模型片段。
网络节点接收了UE侧模型片段标识后，根据UE侧模型片段标识和模型片段选择策略，可以确定网络节点侧模型片段，例如图7中基站72确定的模型片段为模型片段B1。
图6和图7的示例中，若UE为第一设备，网络节点为第二设备，则图6和图7示例的模型协调方案是，第一设备从模型控制功能获取模型片段选择策略，第二设备从模型控制功能获取模型片段选择策略。第一设备根据模型片段选择策略确定了第一模型片段(即图7中的模型片段A1)，然后将第一模型片段的第一标识(即图7中的标识A1)发送给第二设备，然后第二设备根据第一标识和模型片段选择策略确定了第二模型片段(即图7中的模型片段B1)，从而实现了模型片段之间的协调。
图6和图7的示例中，若网络节点为第一设备，UE为第二设备，则图6和图7示例的模型协调方案是，第二设备从模型控制功能获取模型片段选择策略，第一设备从模型控制功能获取模型片段选择策略。第二设备根据模型片段选择策略确定了第二模型片段(即图7中的模型片段A1)，然后将第二模型片段的第二标识(即图7中的标识A1)发送给第一设备，然后第一设备根据第二标识和模型片段选择策略确定了第一模型片段(即图7中的模型片段B1)，从而实现了模型片段之间的协调。
下面将介绍另一种方案。
图8为本申请实施例提供的模型协调方法的信令图二，如图8所示，包括：
S81，服务器向模型控制功能发送模型片段选择策略。
本申请实施例中的模型控制功能为核心网设备中的一个网元，服务器通过模型控制功能向UE和网络节点分别发送模型片段选择策略，其中，服务器也可以直接向UE或者网络节点发送模型片段选择策略，本申请实施例红以服务器同模型控制功能向UE和网络设备发送模型片段选择策略为例。
S82，模型控制功能向网络节点发送模型片段选择策略。
模型控制功能在获取到模型片段选择策略后，向网络节点发送模型片段选择策略。可选的，模型控制功能向网络节点发送网络节点端模型片段选择策略，网络节点端模型片段选择策略包括模型片段选择条件和网络节点端存储的模型片段之间的对应关系。
S83，UE获取UE侧模型片段集合。
当对预设模型有多种划分时，会将这多种划分得到的模型片段分别配置到UE端和网络节点端。图9为本申请实施例提供的模型协调示意图二，如图9所示，包括模型控制功能90、UE91和基站92。服务器为UE91配置的模型片段为模型片段A1、模型片段A2和模型片段A3，服务器为网络节点配置的模型片段为模型片段B1、模型片段B2和模型片段B3。
模型片段选择策略包括模型片段选择条件以及存储的模型片段之间的对应关系，模型片段选择条件例如可以为UE的电量、UE的位置(例如小区标识)、UE的空口信号质量。模型片段选择条件除了可以为UE的相关信息外，还可以是网络节点的负载，还可以是当前时间等与第一设备和第二设备无关的信息。本申请实施例中对模型片段选择条件不限。
S84，网络节点获取网络节点侧模型片段集合和模型片段选择策略。
网络节点获取的网络节点侧模型片段集合中的模型片段是与UE侧的模型片段集合中的模型片段相匹配的，例如在图9中，模型片段A1与模型片段B1匹配，模型片段A2与模型片段B2匹配，模型片段A3与模型片段B3匹配。
S85，网络节点确定网络节点侧模型片段和UE侧模型片段。
在图9中，基站92从模型控制功能90获取模型片段选择策略，然后根据模型片段选择策略确定基站侧模型片段。例如模型片段选择条件为UE的位置，当UE处于小区甲的覆盖范围时，基站92选择模型片段B1，当UE处于小区乙的覆盖范围时，基站92选择模型片段B2，当UE处于除小区甲和小区乙外的其他小区的覆盖范围时，基站92选择模型片段B3。此时UE所处位置为小区甲的覆盖范围内，则基站92确定执行的模型片段为模型片段B1。当基站92为第一设备时，模型片段B1即为第一模型片段，B1为第一模型片段的第一标识；当基站92为第二设备时，模型片段B1即为第二模型片段，B1为第二模型片段的第二标识。
然后网络节点根据网络节点侧模型片段，确定UE侧模型片段，例如基站92确定了模型片段B1后，为UE确定模型片段A1。
S86，网络节点向UE发送UE侧模型片段标识。
当网络节点确定了执行的模型片段为模型片段B1以及为UE确定了模型片段A1后，向UE发送 模型片段A1的标识，例如在图9中，UE91向基站92发送标识A1。
可选的，本申请实施例中的网络节点可以是基站，也可以是核心网设备。当网络节点为基站时，UE侧模型片段标识A1可以包括在AS消息中，基站通过AS消息将UE侧模型片段标识发送给UE，AS消息例如可以为RRC消息。当网络节点为核心网设备时，UE侧模型片段标识A1可以包括在NAS消息中，核心网设备可以通过NAS消息将UE侧模型片段标识发送给UE。
S87，UE确定UE侧模型片段。
UE接收了UE侧模型片段标识后，可以确定UE侧模型片段，例如图9中UE92确定的模型片段为模型片段A1。
图8和图9的示例中，若UE为第一设备，网络节点为第二设备，则图8和图9示例的模型协调方案是，第二设备从模型控制功能获取模型片段选择策略。第二设备根据模型片段选择策略确定了第二模型片段(即图9中的模型片段B1)，然后为第一设备确定了第一模型片段(即图9中的模型片段A1)。接着，第二设备向第一设备发送第一模型片段的第一标识(即图9中的标识A1)，第一设备根据第一标识确定了第一模型片段(即图9中的模型片段A1)，从而实现了模型片段之间的协调。
图8和图9的示例中，若网络节点为第一设备，UE为第二设备，则图8和图9示例的模型协调方案是，第二设备从模型控制功能获取模型片段选择策略。第二设备根据模型片段选择策略确定了第二模型片段(即图9中的模型片段B1)，然后为第一设备确定了第一模型片段(即图9中的模型片段A1)。接着，第二设备向第一设备发送第一模型片段的第一标识(即图9中的标识A1)，第一设备根据第一标识确定了第一模型片段(即图9中的模型片段A1)，从而实现了模型片段之间的协调。
图6-图9示意的实施例中，是两个模型端点之间进行信令交互，从而实现各个模型端点上所需要执行的模型片段的协调。下面将介绍由移动网络中的模型控制功能负责模型端点之间进行模型片段协调的方案。
图10为本申请实施例提供的模型协调方法的信令图三，如图10所示，包括：
S101，服务器向模型控制功能发送模型片段选择策略。
本申请实施例中的模型控制功能为核心网设备中的一个网元，服务器通过模型控制功能向UE和网络节点分别发送模型片段选择策略，其中，服务器也可以直接向UE或者网络节点发送模型片段选择策略，本申请实施例红以服务器同模型控制功能向UE和网络设备发送模型片段选择策略为例。
S102，UE获取UE侧模型片段集合。
S103，网络节点获取网络节点侧模型片段集合。
当对预设模型有多种划分时，会将这多种划分得到的模型片段分别配置到UE端和网络节点端。图11为本申请实施例提供的模型协调示意图三，如图11所示，包括模型控制功能110、UE111和基站112。服务器为UE111配置的模型片段为模型片段A1、模型片段A2和模型片段A3，服务器为网络节点配置的模型片段为模型片段B1、模型片段B2和模型片段B3。
模型片段选择策略包括模型片段选择条件以及存储的模型片段之间的对应关系，模型片段选择条件例如可以为UE的电量、UE的位置(例如小区标识)、UE的空口信号质量。模型片段选择条件除了可以为UE的相关信息外，还可以是网络节点的负载，还可以是当前时间等与第一设备和第二设备无关的信息。本申请实施例中对模型片段选择条件不限。
S104，模型控制功能确定UE侧模型片段和网络节点侧模型片段。
模型控制功能负责确定UE和网络节点上需要执行的模型片段。例如模型控制功能可以从移动网络中的其他网元获得UE的位置(例如小区标识)或者UE的空口信号质量或者根据当前时间等信息确定为UE选择模型片段A1，并为相应的网络节点选择模型片段B1，而在另外一种条件下，例如根据包括并不限于UE的位置移动或者空口信令的变化等条件，为UE选择模型片段A2，并为相应的网络节点选择模型片段B2。
网络节点获取的网络节点侧模型片段集合中的模型片段是与UE侧的模型片段集合中的模型片段相匹配的，例如在图11中，模型片段A1与模型片段B1匹配，模型片段A2与模型片段B2匹配，模型片段A3与模型片段B3匹配。
S105，模型控制功能向UE发送UE侧模型片段标识。
具体的，由于模型控制功能为核心网中的一个网元，因此模型控制功能向UE发送UE侧模型片段标识时，可以将UE侧模型片段标识发送给移动管理功能，然后移动管理功能通过NAS消息将UE侧模型片段标识发送给UE，其中UE侧模型片段标识包括在NAS消息中。
S106，模型控制功能向网络节点发送网络节点侧模型片段标识。
在图11中，模型控制功能110为UE111确定了模型片段A1并为基站112确定了模型片段B1后， 向UE111发送标识A1，向基站112发送标识B2。
然后UE111根据标识A1确定UE侧模型片段A1，基站112根据标识B1确定网络节点侧模型片段B1。
图10和图11的示例中，若UE为第一设备，网络节点为第二设备，则图10和图11示例的模型协调方案是，模型控制功能根据模型片段选择策略为第一设备确定了第一模型片段(即图11中的模型片段A1)，并为第二设备相应选择了第二模型片段(即图11中的模型片段B1)，然后将第一模型片段的第一标识发送给第一设备，将第二模型片段的第二标识发送给第二设备，第一设备根据第一标识确定第一模型片段，第二设备根据第二标识确定第二模型片段，从而实现了模型片段之间的协调。
图10和图11的示例中，若网络节点为第一设备，UE为第二设备，则图10和图11示例的模型协调方案是，模型控制功能根据模型片段选择策略为第二设备确定了第二模型片段(即图11中的模型片段A1)，并为第一设备相应选择了第一模型片段(即图11中的模型片段B1)，然后将第二模型片段的第二标识发送给第二设备，将第一模型片段的第一标识发送给第一设备，第二设备根据第二标识确定第二模型片段，第一设备根据第一标识确定第一模型片段，从而实现了模型片段之间的协调。
下面介绍另一种方案。
图12为本申请实施例提供的模型协调方法的信令图四，如图12所示，包括：
S121，服务器向模型控制功能发送模型片段选择策略。
本申请实施例中的模型控制功能为核心网设备中的一个网元，服务器通过模型控制功能向UE和网络节点分别发送模型片段选择策略，其中，服务器也可以直接向UE或者网络节点发送模型片段选择策略，本申请实施例红以服务器同模型控制功能向UE和网络设备发送模型片段选择策略为例。
S122，模型控制功能向UE发送模型片段选择策略。
模型控制功能在获取到模型片段选择策略后，向UE发送模型片段选择策略。可选的，模型控制功能向UE发送UE端模型片段选择策略，UE端模型片段选择策略包括模型片段选择条件和UE端存储的模型片段之间的对应关系。
S123，UE获取模型片段选择策略和UE侧模型片段集合。
当对预设模型有多种划分时，会将这多种划分得到的模型片段分别配置到UE端和网络节点端。图13为本申请实施例提供的模型协调示意图四，如图13所示，包括模型控制功能130、UE131和基站132。服务器为UE131配置的模型片段为模型片段A1、模型片段A2和模型片段A3，服务器为网络节点配置的模型片段为模型片段B1、模型片段B2和模型片段B3。
模型片段选择策略包括模型片段选择条件以及存储的模型片段之间的对应关系，模型片段选择条件例如可以为UE的电量、UE的位置(例如小区标识)、UE的空口信号质量。模型片段选择条件除了可以为UE的相关信息外，还可以是网络节点的负载，还可以是当前时间等与第一设备和第二设备无关的信息。本申请实施例中对模型片段选择条件不限。
S124，网络节点获取网络节点侧模型片段集合。
网络节点获取的网络节点侧模型片段集合中的模型片段是与UE侧的模型片段集合中的模型片段相匹配的，例如在图13中，模型片段A1与模型片段B1匹配，模型片段A2与模型片段B2匹配，模型片段A3与模型片段B3匹配。
S125，UE确定UE侧模型片段。
UE根据模型片段选择策略确定UE侧模型片段，其中模型片段选择策略中包括模型片段选择条件和UE中存储的模型片段的对应关系。例如模型片段选择条件为UE的位置，当UE处于小区甲的覆盖范围时，UE131选择模型片段A1，当UE处于小区乙的覆盖范围时，UE131选择模型片段A2，当UE处于除小区甲和小区乙外的其他小区的覆盖范围时，UE131选择模型片段A3。此时UE131所处位置为小区甲的覆盖范围内，则UE131确定执行的模型片段为模型片段A1。
S126，UE向模型控制功能发送UE侧模型片段标识。
当UE确定了执行的模型片段后，向模型控制功能发送UE侧模型片段的标识，例如在图13中，UE131向模型控制功能130发送标识A1。由于模型控制功能是核心网的网元，因此UE向模型控制功能发送UE侧模型片段的标识的方式是，在NAS消息中携带UE侧模型片段的标识，然后UE通过NAS消息将UE侧模型片段标识上报到核心网中的移动管理功能，再由移动管理功能将UE侧模型片段标识转发到模型控制功能。
S127，模型控制功能确定网络节点侧模型片段。
模型控制功能接收了UE侧模型片段标识后，可以根据UE侧模型片段标识和模型片段选择条件确定网络节点侧模型片段，例如图13中模型控制功能130确定的网络节点侧模型片段为模型片段B1。
S128，模型控制功能向网络节点发送网络节点侧模型片段标识。
网络节点可以根据网络节点侧模型片段标识确定网络节点侧模型片段。
图12和图13的示例中，若UE为第一设备，网络节点为第二设备，则图12和图13示例的模型协调方案是，第一设备根据模型片段选择策略确定了第一模型片段(即图13中的模型片段A1)，然后第一设备向模型控制功能发送第一模型片段的第一标识，模型控制功能根据第一标识为第二设备确定第二模型片段(即图13中的模型片段B1)，并向第二设备发送第二模型片段的第二标识，第二设备根据第二标识确定第二模型片段，从而实现了模型片段之间的协调。
图12和图13的示例中，若网络节点为第一设备，UE为第二设备，则图12和图13示例的模型协调方案是，第二设备根据模型片段选择策略确定了第二模型片段(即图13中的模型片段A1)，然后第二设备向模型控制功能发送第二模型片段的第二标识，模型控制功能根据第二标识为第一设备确定第一模型片段(即图13中的模型片段B1)，并向第一设备发送第一模型片段的第一标识，第一设备根据第一标识确定第一模型片段，从而实现了模型片段之间的协调。
在上述实施例中，是以UE和网络设备作为第一设备或第二设备为例进行说明的，实际上第一设备和第二设备可以均为UE，也可以均为网络设备，也可以其中一个为UE，另一个为网络设备。上述实施例中，描述了UE和模型控制功能、UE和基站、UE和核心网设备之间的数据传输方式，例如模型控制功能为UE配置模型片段选择策略时，可以先向移动管理功能发送模型片段选择策略，然后移动管理功能通过在NAS消息中携带模型片段选择策略，从而实现向UE发送模型片段选择策略。例如UE向基站发送模型片段的标识时，可以通过AS消息来向基站发送模型片段的标识，UE向核心网设备发送模型片段的标识时，可以通过NAS消息来向核心网设备发送模型片段的标识，等等。本领域技术人员可以获知的是，第一设备、第二设备和模型控制功能之间的数据传输，具体可以通过哪种消息来携带需要传输的数据，是可以根据第一设备和第二设备的具体类型来确定的，上述实施例仅仅是以第一设备和第二设备的某一种可能的设备类型为例来进行描述的，具体的数据传输可以根据第一设备和第二设备的实际类型得到，此处不再赘述。
图14为本申请实施例提供的模型协调方法的流程示意图，该方法应用于第三设备，如图14所示，包括：
S141，确定第一模型片段，所述第一模型片段为第一设备中存储的至少一个模型片段中的一个，当所述第一模型片段在第一设备中被执行时，预设模型的部分功能被实现；
S142，向第一设备发送所述第一模型片段的第一标识。
本申请实施例中，第三设备是区别于第一设备、第二设备和模型控制功能外的设备，是用于为第一设备确定第一模型片段的。第三设备可以从模型控制功能或者云端模拟服务器获取模型片段选择策略，然后根据模型片段选择策略为第一设备确定第一模型片段，其中模型片段选择策略中包括模型片段选择条件和第一设备存储的模型片段之间的对应关系。
可选的，模型片段选择策略中还包括预设模型的标识，第三设备可以根据预设模型的标识，在第一设备存储的至少一个模型片段中确定预设模型的至少一个模型片段，并在预设模型的至少一个模型片段中确定第一模型片段。
在确定了第一模型片段后，第三设备向第一设备发送第一标识，第一设备可以根据第一标识确定要执行的模型片段为第一模型片段。
可选的，模型片段选择策略中还包括第一设备的标识，第三设备可以根据第一设备的标识向第一设备发送第一标识。
图15为本申请实施例提供的模型协调方法的流程示意图，如图15所示，包括：
S151，确定第一设备的第一模型片段，其中，所述第一模型片段为所述第一设备中存储的至少一个模型片段中的一个；
S152，确定第二设备的第二模型片段，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，当所述第一模型片段在所述第一设备中被执行且所述第二模型片段在所述第二设备中被执行时，预设模型的部分功能或者全部功能被实现；
S153，向所述第二设备发送所述第二模型片段的第二标识。
可选的，确定第二设备的第二模型片段，包括：
根据模型片段选择策略，确定所述第二模型片段，其中，所述模型片段选择策略中包括模型片段选择条件和所述第二设备中存储的模型片段之间的对应关系。
可选的，所述模型片段选择策略中还包括模型片段选择策略和所述第一设备中存储的模型片段之间的对应关系，确定第一设备的第一模型片段，包括：
根据所述模型片段选择策略和所述第一设备中存储的模型片段之间的对应关系，确定所述第一模型片段。
可选的，确定第一设备的第一模型片段，包括：
从所述第一设备获取第一模型片段的第一标识；
根据所述第一标识确定所述第一模型片段。
可选的，所述方法还包括：
向所述第一设备发送所述模型片段选择策略。
图15所示的方法为模型控制功能侧的方法，模型控制功能侧的方法在上述实施例中已进行介绍，具体的方案请参见上述实施例，此处不再赘述。
图16为本申请实施例提供的模型协调装置的结构示意图一，如图16所示，该模型协调装置160包括确定模块161，其中：
确定模块161用于在第一设备中存储的至少一个模型片段中确定第一模型片段，其中，所述第一设备中存储至少一个模型片段，用于实现预设模型的部分功能，当所述第一模型片段被执行且第二模型片段在第二设备中被执行时，所述预设模型的部分功能或者全部功能被实现，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，所述第二设备中存储的至少一个模型片段用于实现所述预设模型的部分功能。
在一种可能的实现方式中，所述确定模块161具体用于：
获取第一信息；
根据所述第一信息，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
在一种可能的实现方式中，所述第一信息包括如下信息中的至少一个：
模型片段选择策略；
第一模型片段的第一标识。
在一种可能的实现方式中，所述第一信息包括模型片段选择策略；所述确定模块161具体用于：
根据所述模型片段选择策略，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段，所述模型片段选择策略中包括模型片段选择条件和所述第一设备存储的模型片段之间的对应关系。
在一种可能的实现方式中，所述模型片段选择策略中还包括所述预设模型的标识，所述确定模块161具体用于：
根据所述预设模型的标识，确定所述第一设备中存储的、所述预设模型的至少一个模型片段；
根据模型片段选择条件和所述第一设备存储的模型片段之间的对应关系，在所述第一设备中存储的、所述预设模型的至少一个模型片段中，确定所述第一模型片段。
在一种可能的实现方式中，所述确定模块161具体用于：
接收所述第二模型片段的第二标识；
根据所述第二标识和所述模型片段选择策略，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
在一种可能的实现方式中，所述确定模块161还用于：
确定所述第一模型片段的第一标识；
向所述第二设备或模型控制功能发送所述第一标识。
在一种可能的实现方式中，所述模型片段选择策略中还包括所述模型片段选择条件和所述第二设备存储的模型片段之间的对应关系，所述确定模块161还用于：
根据所述模型片段选择条件和所述第二设备存储的模型片段之间的对应关系，确定所述第二模型片段。
在一种可能的实现方式中，所述模型片段选择策略中还包括所述第二设备的标识，所述确定模块161还用于：
确定所述第二模型片段的第二标识；
根据所述第二设备的标识向所述第二设备发送所述第二标识。
在一种可能的实现方式中，所述确定模块161还用于：
从模型控制功能接收所述模型片段选择策略。
在一种可能的实现方式中，所述第一信息包括第一模型片段的第一标识；所述确定模块161具体用于：
根据所述第一标识，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
在一种可能的实现方式中，所述第一信息来自所述第二设备或模型控制功能。
在一种可能的实现方式中，所述第一设备为第一终端设备或第一网络设备，所述第二设备为第二终端设备或第二网络设备。
在一种可能的实现方式中，所述第一网络设备为基站或者核心网设备，所述第二网络设备为基站或者核心网设备。
本申请实施例提供的模型协调装置可以执行上述方法实施例所示的技术方案，其实现原理以及有益效果类似，此处不再进行赘述。
图17为本申请实施例提供的模型协调装置的结构示意图二，如图17所示，该模型协调装置170包括确定模块171和发送模块172，其中：
确定模块171用于确定第一模型片段，所述第一模型片段为第一设备中存储的至少一个模型片段中的一个，当所述第一模型片段在第一设备中被执行时，预设模型的部分功能被实现；
发送模块172用于向第一设备发送所述第一模型片段的第一标识。
在一种可能的实现方式中，所述确定模块171具体用于：
根据模型片段选择策略，确定所述第一模型片段，所述模型片段选择策略包括中包括模型片段选择条件和所述第一设备存储的模型片段之间的对应关系。
在一种可能的实现方式中，所述模型片段选择策略中还包括所述预设模型的标识，所述确定模块171具体用于：
根据所述预设模型的标识，确定所述第一设备中存储的、所述预设模型的至少一个模型片段；
根据所述模型片段选择条件和所述第一设备存储的模型片段之间的对应关系，在所述第一设备中存储的、所述预设模型的至少一个模型片段中，确定所述第一模型片段。
在一种可能的实现方式中，所述模型片段选择策略中还包括第一设备的标识，所述发送模块172具体用于：
根据所述第一设备的标识，向所述第一设备发送所述第一标识。
在一种可能的实现方式中，所述确定模块171还用于：
从模型控制功能接收所述模型片段选择策略。
本申请实施例提供的模型协调装置可以执行上述方法实施例所示的技术方案，其实现原理以及有益效果类似，此处不再进行赘述。
图18为本申请实施例提供的模型协调装置的结构示意图三，如图18所示，该模型协调装置180包括第一处理模块181、第二处理模块182和发送模块183，其中：
第一处理模块181用于确定第一设备的第一模型片段，其中，所述第一模型片段为所述第一设备中存储的至少一个模型片段中的一个；
第二处理模块182用于确定第二设备的第二模型片段，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，当所述第一模型片段在所述第一设备中被执行且所述第二模型片段在所述第二设备中被执行时，预设模型的部分功能或者全部功能被实现；
发送模块183用于向所述第二设备发送所述第二模型片段的第二标识。
在一种可能的实现方式中，所述第二处理模块182具体用于：
根据模型片段选择策略，确定所述第二模型片段，其中，所述模型片段选择策略中包括模型片段选择条件和所述第二设备中存储的模型片段之间的对应关系。
在一种可能的实现方式中，所述模型片段选择策略中还包括模型片段选择条件和所述第一设备中存储的模型片段之间的对应关系，所述第一处理模块181具体用于：
根据所述模型片段选择条件和所述第一设备中存储的模型片段之间的对应关系，确定所述第一模型片段。
在一种可能的实现方式中，所述第一处理模块181具体用于：
从所述第一设备获取第一模型片段的第一标识；
根据所述第一标识确定所述第一模型片段。
在一种可能的实现方式中，所述第一处理模块181还用于：
向所述第一设备发送所述模型片段选择策略。
本申请实施例提供的模型协调装置可以执行上述方法实施例所示的技术方案，其实现原理以及有益效果类似，此处不再进行赘述。
图19为本申请实施例提供的模型协调设备的结构示意图。请参见图19，模型协调设备190可以包括：收发器191、存储器192、处理器193。收发器191可包括：发射器和/或接收器。该发射器还可称为发送器、发射机、发送端口或发送接口等类似描述，接收器还可称为接收器、接收机、接收端口或接 收接口等类似描述。示例性地，收发器191、存储器192、处理器193，各部分之间通过总线194相互连接。
存储器192用于存储程序指令；
处理器193用于执行该存储器所存储的程序指令，用以使得终端设备190执行上述任一所示的模型协调方法。
其中，收发器191的接收器，可用于执行上述模型协调方法中模型协调设备的接收功能。
图20为本申请实施例提供的模型协调设备的结构示意图。请参见图20，模型协调设备200可以包括：收发器201、存储器202、处理器203。收发器201可包括：发射器和/或接收器。该发射器还可称为发送器、发射机、发送端口或发送接口等类似描述，接收器还可称为接收器、接收机、接收端口或接收接口等类似描述。示例性地，收发器201、存储器202、处理器203，各部分之间通过总线204相互连接。
存储器202用于存储程序指令；
处理器203用于执行该存储器所存储的程序指令，用以使得模型协调设备200执行上述任一所示的模型协调方法。
其中，收发器201的接收器，可用于执行上述模型协调方法中模型协调设备的接收功能。
本申请实施例提供一种计算机可读存储介质，所述计算机可读存储介质中存储有计算机执行指令，当所述计算机执行指令被处理器执行时用于实现上述模型协调方法。
本申请实施例提供一种计算机可读存储介质，所述计算机可读存储介质中存储有计算机执行指令，当所述计算机执行指令被处理器执行时用于实现上述模型协调方法。
本申请实施例还可提供一种计算机程序产品，该计算机程序产品可以由处理器执行，在计算机程序产品被执行时，可实现上述任一所示的模型协调设备执行的模型协调方法。
本申请实施例的传输设备、计算机可读存储介质及计算机程序产品，可执行上述模型协调设备执行的模型协调方法，其具体的实现过程及有益效果参见上述，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
本领域普通技术人员可以理解：实现上述各方法实施例的全部或部分步骤可以通过程序指令相关的硬件来完成。前述的计算机程序可以存储于一计算机可读取存储介质中。该计算机程序在被处理器执行时，实现包括上述各方法实施例的步骤；而前述的存储介质包括：ROM、RAM、磁碟或者光盘等各种可以存储程序代码的介质。
最后应说明的是：以上各实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述各实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分或者全部技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的范围。

Claims (51)

  1. 一种模型协调方法，其特征在于，应用于第一设备，所述第一设备中存储至少一个模型片段，其中所述第一设备中存储的至少一个模型片段用于实现预设模型的部分功能，所述方法包括：
    在所述第一设备中存储的至少一个模型片段中确定第一模型片段，其中，当所述第一模型片段被执行且第二模型片段在第二设备中被执行时，所述预设模型的部分功能或者全部功能被实现，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，所述第二设备中存储的至少一个模型片段用于实现所述预设模型的部分功能。
  2. 根据权利要求1所述的方法，其特征在于，在所述第一设备中存储的至少一个模型片段中确定第一模型片段，包括：
    获取第一信息；
    根据所述第一信息，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
  3. 根据权利要求2所述的方法，其特征在于，所述第一信息包括如下信息中的至少一个：
    模型片段选择策略；
    第一模型片段的第一标识。
  4. 根据权利要求2或3所述的方法，其特征在于，所述第一信息包括模型片段选择策略；根据所述第一信息，在所述第一设备中存储的至少一个模型片段中确定第一模型片段，包括：
    根据所述模型片段选择策略，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段，所述模型片段选择策略中包括模型片段选择条件和所述第一设备存储的模型片段之间的对应关系。
  5. 根据权利要求4所述的方法，其特征在于，所述模型片段选择策略中还包括所述预设模型的标识，根据所述模型片段选择策略，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段，包括：
    根据所述预设模型的标识，确定所述第一设备中存储的、所述预设模型的至少一个模型片段；
    根据模型片段选择条件和所述第一设备存储的模型片段之间的对应关系，在所述第一设备中存储的、所述预设模型的至少一个模型片段中，确定所述第一模型片段。
  6. 根据权利要求4或5所述的方法，其特征在于，根据所述模型片段选择策略，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段，包括：
    接收所述第二模型片段的第二标识；
    根据所述第二标识和所述模型片段选择策略，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
  7. 根据权利要求4或5所述的方法，其特征在于，所述方法还包括：
    确定所述第一模型片段的第一标识；
    向所述第二设备或模型控制功能发送所述第一标识。
  8. 根据权利要求4或5所述的方法，其特征在于，所述模型片段选择策略中还包括所述模型片段选择条件和所述第二设备存储的模型片段之间的对应关系，所述方法还包括：
    根据所述模型片段选择条件和所述第二设备存储的模型片段之间的对应关系，确定所述第二模型片段。
  9. 根据权利要求8所述的方法，其特征在于，所述模型片段选择策略中还包括所述第二设备的标识，所述方法还包括：
    确定所述第二模型片段的第二标识；
    根据所述第二设备的标识向所述第二设备发送所述第二标识。
  10. 根据权利要求3-9任一项所述的方法，其特征在于，所述方法还包括：
    从模型控制功能接收所述模型片段选择策略。
  11. 根据权利要求2或3所述的方法，其特征在于，所述第一信息包括第一模型片段的第一标识；在所述第一设备中存储的至少一个模型片段中确定第一模型片段，包括：
    根据所述第一标识，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
  12. 根据权利要求11所述的方法，其特征在于，所述第一信息来自所述第二设备或模型控制功能。
  13. 根据权利要求1-12任一项所述的方法，其特征在于，所述第一设备为第一终端设备或第一网络设备，所述第二设备为第二终端设备或第二网络设备。
  14. 根据权利要求13所述的方法，其特征在于，所述第一网络设备为基站或者核心网设备，所述第二网络设备为基站或者核心网设备。
  15. 一种模型协调方法，其特征在于，应用于第三设备，所述方法包括：
    确定第一模型片段，所述第一模型片段为第一设备中存储的至少一个模型片段中的一个，当所述第一模型片段在第一设备中被执行时，预设模型的部分功能被实现；
    向第一设备发送所述第一模型片段的第一标识。
  16. 根据权利要求15所述的方法，其特征在于，确定第一模型片段，包括：
    根据模型片段选择策略，确定所述第一模型片段，所述模型片段选择策略包括中包括模型片段选择条件和所述第一设备存储的模型片段之间的对应关系。
  17. 根据权利要求16所述的方法，其特征在于，所述模型片段选择策略中还包括所述预设模型的标识，根据模型片段选择策略，确定所述第一模型片段，包括：
    根据所述预设模型的标识，确定所述第一设备中存储的、所述预设模型的至少一个模型片段；
    根据所述模型片段选择条件和所述第一设备存储的模型片段之间的对应关系，在所述第一设备中存储的、所述预设模型的至少一个模型片段中，确定所述第一模型片段。
  18. 根据权利要求16或17所述的方法，其特征在于，所述模型片段选择策略中还包括第一设备的标识，向第一设备发送所述第一标识，包括：
    根据所述第一设备的标识，向所述第一设备发送所述第一标识。
  19. 根据权利要求16-18任一项所述的方法，其特征在于，所述方法还包括：
    从模型控制功能接收所述模型片段选择策略。
  20. 一种模型协调方法，其特征在于，应用于模型控制功能，所述方法包括：
    确定第一设备的第一模型片段，其中，所述第一模型片段为所述第一设备中存储的至少一个模型片段中的一个；
    确定第二设备的第二模型片段，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，当所述第一模型片段在所述第一设备中被执行且所述第二模型片段在所述第二设备中被执行时，预设模型的部分功能或者全部功能被实现；
    向所述第二设备发送所述第二模型片段的第二标识。
  21. 根据权利要求20所述的方法，其特征在于，确定第二设备的第二模型片段，包括：
    根据模型片段选择策略，确定所述第二模型片段，其中，所述模型片段选择策略中包括模型片段选择条件和所述第二设备中存储的模型片段之间的对应关系。
  22. 根据权利要求21所述的方法，其特征在于，所述模型片段选择策略中还包括模型片段选择条件和所述第一设备中存储的模型片段之间的对应关系，确定第一设备的第一模型片段，包括：
    根据所述模型片段选择条件和所述第一设备中存储的模型片段之间的对应关系，确定所述第一模型片段。
  23. 根据权利要求21所述的方法，其特征在于，确定第一设备的第一模型片段，包括：
    从所述第一设备获取第一模型片段的第一标识；
    根据所述第一标识确定所述第一模型片段。
  24. 根据权利要求23所述的方法，其特征在于，所述方法还包括：
    向所述第一设备发送所述模型片段选择策略。
  25. 一种模型协调装置，其特征在于，包括：
    确定模块，用于在第一设备中存储的至少一个模型片段中确定第一模型片段，其中，所述第一设备中存储至少一个模型片段，用于实现预设模型的部分功能，当所述第一模型片段被执行且第二模型片段在第二设备中被执行时，所述预设模型的部分功能或者全部功能被实现，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，所述第二设备中存储的至少一个模型片段用于实现所述预设模型的部分功能。
  26. 根据权利要求25所述的装置，其特征在于，所述确定模块具体用于：
    获取第一信息；
    根据所述第一信息，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
  27. 根据权利要求26所述的装置，其特征在于，所述第一信息包括如下信息中的至少一个：
    模型片段选择策略；
    第一模型片段的第一标识。
  28. 根据权利要求26或27所述的装置，其特征在于，所述第一信息包括模型片段选择策略；所述确定模块具体用于：
    根据所述模型片段选择策略，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段，所述模型片段选择策略中包括模型片段选择条件和所述第一设备存储的模型片段之间的对应关系。
  29. 根据权利要求28所述的装置，其特征在于，所述模型片段选择策略中还包括所述预设模型的标识，所述确定模块具体用于：
    根据所述预设模型的标识，确定所述第一设备中存储的、所述预设模型的至少一个模型片段；
    根据模型片段选择条件和所述第一设备存储的模型片段之间的对应关系，在所述第一设备中存储的、所述预设模型的至少一个模型片段中，确定所述第一模型片段。
  30. 根据权利要求28或29所述的装置，其特征在于，所述确定模块具体用于：
    接收所述第二模型片段的第二标识；
    根据所述第二标识和所述模型片段选择策略，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
  31. 根据权利要求28或29所述的装置，其特征在于，所述确定模块还用于：
    确定所述第一模型片段的第一标识；
    向所述第二设备或模型控制功能发送所述第一标识。
  32. 根据权利要求28或29所述的装置，其特征在于，所述模型片段选择策略中还包括所述模型片段选择条件和所述第二设备存储的模型片段之间的对应关系，所述确定模块还用于：
    根据所述模型片段选择条件和所述第二设备存储的模型片段之间的对应关系，确定所述第二模型片段。
  33. 根据权利要求32所述的装置，其特征在于，所述模型片段选择策略中还包括所述第二设备的标识，所述确定模块还用于：
    确定所述第二模型片段的第二标识；
    根据所述第二设备的标识向所述第二设备发送所述第二标识。
  34. 根据权利要求27-33任一项所述的装置，其特征在于，所述确定模块还用于：
    从模型控制功能接收所述模型片段选择策略。
  35. 根据权利要求26或27所述的装置，其特征在于，所述第一信息包括第一模型片段的第一标识；所述确定模块具体用于：
    根据所述第一标识，在所述第一设备中存储的至少一个模型片段中确定所述第一模型片段。
  36. 根据权利要求35所述的装置，其特征在于，所述第一信息来自所述第二设备或模型控制功能。
  37. 根据权利要求25-36任一项所述的装置，其特征在于，所述第一设备为第一终端设备或第一网络设备，所述第二设备为第二终端设备或第二网络设备。
  38. 根据权利要求37所述的装置，其特征在于，所述第一网络设备为基站或者核心网设备，所述第二网络设备为基站或者核心网设备。
  39. 一种模型协调装置，其特征在于，包括：
    确定模块，用于确定第一模型片段，所述第一模型片段为第一设备中存储的至少一个模型片段中的一个，当所述第一模型片段在第一设备中被执行时，预设模型的部分功能被实现；
    发送模块，用于向第一设备发送所述第一模型片段的第一标识。
  40. 根据权利要求39所述的装置，其特征在于，所述确定模块具体用于：
    根据模型片段选择策略，确定所述第一模型片段，所述模型片段选择策略包括中包括模型片段选择条件和所述第一设备存储的模型片段之间的对应关系。
  41. 根据权利要求40所述的装置，其特征在于，所述模型片段选择策略中还包括所述预设模型的标识，所述确定模块具体用于：
    根据所述预设模型的标识，确定所述第一设备中存储的、所述预设模型的至少一个模型片段；
    根据所述模型片段选择条件和所述第一设备存储的模型片段之间的对应关系，在所述第一设备中存储的、所述预设模型的至少一个模型片段中，确定所述第一模型片段。
  42. 根据权利要求40或41所述的装置，其特征在于，所述模型片段选择策略中还包括第一设备的标识，所述发送模块具体用于：
    根据所述第一设备的标识，向所述第一设备发送所述第一标识。
  43. 根据权利要求40-42任一项所述的装置，其特征在于，所述确定模块还用于：
    从模型控制功能接收所述模型片段选择策略。
  44. 一种模型协调装置，其特征在于，包括：
    第一处理模块，用于确定第一设备的第一模型片段，其中，所述第一模型片段为所述第一设备中存储的至少一个模型片段中的一个；
    第二处理模块，用于确定第二设备的第二模型片段，所述第二模型片段为所述第二设备中存储的至少一个模型片段中的一个，当所述第一模型片段在所述第一设备中被执行且所述第二模型片段在所述第 二设备中被执行时，预设模型的部分功能或者全部功能被实现；
    发送模块，用于向所述第二设备发送所述第二模型片段的第二标识。
  45. 根据权利要求44所述的装置，其特征在于，所述第二处理模块具体用于：
    根据模型片段选择策略，确定所述第二模型片段，其中，所述模型片段选择策略中包括模型片段选择条件和所述第二设备中存储的模型片段之间的对应关系。
  46. 根据权利要求45所述的装置，其特征在于，所述模型片段选择策略中还包括模型片段选择条件和所述第一设备中存储的模型片段之间的对应关系，所述第一处理模块具体用于：
    根据所述模型片段选择条件和所述第一设备中存储的模型片段之间的对应关系，确定所述第一模型片段。
  47. 根据权利要求45所述的装置，其特征在于，所述第一处理模块具体用于：
    从所述第一设备获取第一模型片段的第一标识；
    根据所述第一标识确定所述第一模型片段。
  48. 根据权利要求47所述的装置，其特征在于，所述第一处理模块还用于：
    向所述第一设备发送所述模型片段选择策略。
  49. 一种模型协调设备，其特征在于，包括：收发器、处理器、存储器；
    所述存储器存储计算机执行指令；
    所述处理器执行所述存储器存储的计算机执行指令，使得所述处理器执行如权利要求1至14任一项所述的模型协调方法，或者，使得所述处理器执行如权利要求15至19任一项所述的模型协调方法。
  50. 一种模型协调设备，其特征在于，包括：收发器、处理器、存储器；
    所述存储器存储计算机执行指令；
    所述处理器执行所述存储器存储的计算机执行指令，使得所述处理器执行如权利要求20至24任一项所述的模型协调方法。
  51. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质中存储有计算机执行指令，当所述计算机执行指令被处理器执行时用于实现如权利要求1至24任一项所述的模型协调方法。
PCT/CN2020/078609 2020-03-10 2020-03-10 模型协调方法及装置 WO2021179165A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
PCT/CN2020/078609 WO2021179165A1 (zh) 2020-03-10 2020-03-10 模型协调方法及装置
CN202080098134.5A CN115244512A (zh) 2020-03-10 2020-03-10 模型协调方法及装置
EP20924693.3A EP4113300A4 (en) 2020-03-10 2020-03-10 MODEL COORDINATION METHOD AND DEVICE
US17/902,459 US20230004839A1 (en) 2020-03-10 2022-09-02 Model coordination method and apparatus

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/078609 WO2021179165A1 (zh) 2020-03-10 2020-03-10 模型协调方法及装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/902,459 Continuation US20230004839A1 (en) 2020-03-10 2022-09-02 Model coordination method and apparatus

Publications (1)

Publication Number Publication Date
WO2021179165A1 true WO2021179165A1 (zh) 2021-09-16

Family

ID=77670376

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/078609 WO2021179165A1 (zh) 2020-03-10 2020-03-10 模型协调方法及装置

Country Status (4)

Country Link
US (1) US20230004839A1 (zh)
EP (1) EP4113300A4 (zh)
CN (1) CN115244512A (zh)
WO (1) WO2021179165A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024067245A1 (zh) * 2022-09-29 2024-04-04 华为技术有限公司 模型匹配的方法和通信装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108632880A (zh) * 2017-03-16 2018-10-09 中国移动通信有限公司研究院 一种组网工作模式选择方法、装置及终端
US20190050505A1 (en) * 2017-08-10 2019-02-14 Allstate Insurance Company Multi-platform model processing and execution management engine
CN110764905A (zh) * 2019-09-24 2020-02-07 Oppo广东移动通信有限公司 网络模型生成方法、装置、计算机设备以及存储介质
CN110809771A (zh) * 2017-07-06 2020-02-18 谷歌有限责任公司 用于机器学习模型的压缩和分发的系统和方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11144812B2 (en) * 2017-09-01 2021-10-12 Facebook, Inc. Mixed machine learning architecture

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108632880A (zh) * 2017-03-16 2018-10-09 中国移动通信有限公司研究院 一种组网工作模式选择方法、装置及终端
CN110809771A (zh) * 2017-07-06 2020-02-18 谷歌有限责任公司 用于机器学习模型的压缩和分发的系统和方法
US20190050505A1 (en) * 2017-08-10 2019-02-14 Allstate Insurance Company Multi-platform model processing and execution management engine
CN110764905A (zh) * 2019-09-24 2020-02-07 Oppo广东移动通信有限公司 网络模型生成方法、装置、计算机设备以及存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4113300A4 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024067245A1 (zh) * 2022-09-29 2024-04-04 华为技术有限公司 模型匹配的方法和通信装置

Also Published As

Publication number Publication date
EP4113300A4 (en) 2023-04-05
CN115244512A (zh) 2022-10-25
US20230004839A1 (en) 2023-01-05
EP4113300A1 (en) 2023-01-04

Similar Documents

Publication Publication Date Title
WO2022022334A1 (zh) 基于人工智能的通信方法和通信装置
WO2022016401A1 (zh) 通信方法和通信装置
WO2020200034A1 (zh) 一种网络接入的方法和装置
JP2021503845A (ja) 通信方法および通信デバイス
WO2022088826A1 (zh) 一种计算服务发现方法及通信装置
US20230337065A1 (en) Network resource selection method, and terminal device and network device
US10178689B2 (en) Adaptive resource allocation for simultaneous operation of cellular and device to device communications
CN113966592B (zh) 更新后台数据传输策略的方法和装置
CN104737592A (zh) 用于设备对设备发现的系统信息
WO2020211778A1 (zh) 小区切换方法以及装置
US20230139345A1 (en) Computing service implementation method and apparatus
CN108924926B (zh) 通信方法及通信装置
US20230004839A1 (en) Model coordination method and apparatus
WO2022056676A1 (zh) 业务识别方法、终端设备和网络设备
CN116325890A (zh) 网络切片动态拥塞控制
CN113747605A (zh) 通信方法和通信装置
WO2022126641A1 (zh) 无线通信方法、终端设备、第一接入网设备以及网元
CN104662991A (zh) 用于通信的装置和方法
WO2023125598A1 (zh) 一种通信方法及通信装置
US20220377547A1 (en) Wireless communication method, terminal device and network element
WO2023125599A1 (zh) 一种通信方法及通信装置
WO2022222748A1 (zh) 中继通信方法和装置
WO2023015573A1 (zh) 通信方法、设备及存储介质
WO2023070684A1 (zh) 无线通信的方法和设备
WO2022061829A1 (zh) 无线通信方法和设备

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: 20924693

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020924693

Country of ref document: EP

Effective date: 20220930

NENP Non-entry into the national phase

Ref country code: DE