WO2023020516A1 - 一种人工智能ai模型传输方法及装置 - Google Patents

一种人工智能ai模型传输方法及装置 Download PDF

Info

Publication number
WO2023020516A1
WO2023020516A1 PCT/CN2022/112905 CN2022112905W WO2023020516A1 WO 2023020516 A1 WO2023020516 A1 WO 2023020516A1 CN 2022112905 W CN2022112905 W CN 2022112905W WO 2023020516 A1 WO2023020516 A1 WO 2023020516A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
terminal device
information
format
communication
Prior art date
Application number
PCT/CN2022/112905
Other languages
English (en)
French (fr)
Inventor
王四海
杨锐
秦城
Original Assignee
华为技术有限公司
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
Priority claimed from CN202111576356.6A external-priority patent/CN115942298A/zh
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to EP22857827.4A priority Critical patent/EP4383774A1/en
Publication of WO2023020516A1 publication Critical patent/WO2023020516A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W8/00—Network data management
    • H04W8/22—Processing or transfer of terminal data, e.g. status or physical capabilities
    • H04W8/24—Transfer of terminal data

Definitions

  • the embodiments of the present application relate to the field of communication technologies, and in particular to an artificial intelligence AI model transmission method and device.
  • AI Artificial intelligence
  • the wireless communication system introduces AI technology, and may gradually use AI modules to replace the functional modules in the wireless communication system.
  • a possible working mode is that the network device sends the AI model to the terminal device, and the terminal device receives the AI model from the network device and applies the AI model for wireless communication.
  • the AI model will have different levels of representation in different formats. Different terminal devices have different abilities to recognize AI model formats. Therefore, the AI model delivered by the network device may not be recognized by the terminal device. This will result in the inability to apply AI models for wireless communication.
  • Embodiments of the present application provide an artificial intelligence AI model transmission method and device, in order to better apply AI technology in a wireless communication system.
  • a method for transmitting an artificial intelligence AI model is provided, and the method may be executed by a terminal device or by components of the terminal device.
  • the method may be implemented through the following steps: the terminal device sends first information to the network device, the first information includes an AI model format supported by the terminal device, and the AI model format supported by the terminal device includes a first AI model format;
  • the terminal device receives second information from the network device, where the second information is used to indicate the acquisition method information of the first AI model; wherein, the first AI model is expressed in a first AI model format, and the The format of the first AI model is determined according to the first information, and/or, the first AI model is applied to a first communication scenario.
  • the network device can know the AI model format supported by the terminal device. In this way, the network device will determine the first AI model format according to the AI model format supported by the terminal device.
  • the model format must be a model format supported by the terminal device.
  • the network device indicates to the terminal device the acquisition method information of the first AI model represented by the first AI model format, so that the model format of the first AI model acquired by the terminal device according to the acquisition method information must be supported by itself, ensuring that the terminal device and The model format of the AI model for the wireless communication service performed by the network device is agreed, and the terminal device can understand or recognize the AI model format, so as to ensure the feasibility of using the AI model for the communication service.
  • the terminal device acquires the first AI model according to the acquisition method information.
  • the terminal device can download the first AI model to the first device according to the acquisition method information.
  • the first device can be a server or a core network device.
  • the AI model can be stored and maintained by the first device.
  • the number of the first device can be more than that of network devices. The number is much smaller, reducing maintenance costs, reducing the overhead and power consumption of network equipment, and improving the performance of communication systems.
  • the terminal device sends a request message to the first device; wherein the request message is used to request the first device to judge whether the first AI model meets the hardware requirements of the terminal device. Whether the first AI model satisfies the hardware requirements of the terminal device, and when the hardware requirements are met, the first device sends the first AI model to the terminal device, and the first AI model can better adapt to the hardware requirements of the terminal device.
  • the request message may also be used to indicate the first communication scenario.
  • the request message includes an indication of the first AI model and /or the first AI model format.
  • the indication of the first AI model is used to indicate the first AI model to be acquired.
  • a model to be delivered by the first device is indicated through the indication of the first AI model and/or the format of the first AI model.
  • the indication of the first AI model may be, for example, the identification of the first AI model.
  • the first device and the terminal device pre-negotiate the correspondence between multiple AI models and multiple identifications, so that the first device can select the required AI model according to the indication of the first AI model.
  • the AI model delivered to the terminal device is used to indicate whether the first AI model meets the hardware requirements of the terminal device, in a possible design.
  • the request message is used to request the first AI model expressed in the first AI model format.
  • the request message is used to request the first AI model represented by the first AI model format, in a possible design, the request message is also used to indicate the first AI model format.
  • the first AI model format may be a model format output by a deep learning framework, an intermediate representation layer model format, or a model format represented by an executable file.
  • the request message includes one or more of the following: a first threshold, a second threshold, a hardware model, a hardware version, a storage capability, a computing capability, and a capability to support heterogeneous computing; wherein, the The first threshold is the maximum delay of the execution part of the AI model when the terminal device uses the AI model to perform communication services, and the second threshold is the maximum work delay of the execution part of the AI model when the terminal device uses the AI model to perform communication services.
  • the AI model compilation capability includes the capability of compiling and converting AI model formats.
  • the first threshold is reported by the terminal device, so that the first device can judge whether the first AI model meets the delay requirement according to the first threshold, so that the first AI model can better meet the communication delay requirement, and better apply AI in communication services technology.
  • the terminal device reports the second threshold, so that the first device can judge whether the first AI model meets the power consumption requirement according to the second threshold, so that the first AI model can better adapt to the power consumption requirement of the terminal device.
  • the terminal device sends third information to the network device, where the third information is used to indicate whether the terminal device uses the first AI model to perform a communication service.
  • the terminal device can be instructed to use the AI technology to perform communication services, and the consistency of the working modes between the terminal device and the network device can be achieved.
  • the AI model format includes one or more of the following: a model format output by a deep learning framework, an intermediate representation layer model format, or a model format represented by an executable file. Each of these includes one or more specific AI model file formats.
  • the first information further includes a first threshold and/or a second threshold
  • the first threshold is the maximum delay of the execution part of the AI model when the terminal device uses the AI model to execute communication services
  • the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to execute the communication service.
  • the first threshold is reported by the terminal device, so that the network device can judge whether the first AI model meets the delay requirement according to the first threshold, so that the first AI model can better meet the communication delay requirement, and better apply AI technology in communication services .
  • the terminal device reports the second threshold, so that the network device can judge whether the first AI model meets the power consumption requirement according to the second threshold, so that the first AI model can better adapt to the power consumption requirement of the terminal device.
  • the first information further includes one or more of the following: hardware model, hardware version, storage capability, computing capability, ability to support heterogeneous computing, or AI model compilation capability
  • the AI model Compilation capabilities include the ability to compile and convert AI model formats.
  • the software capability may include the compiling capability of the terminal device.
  • the terminal device receives working mode configuration information from the network device, the working mode configuration information is used to configure the working mode for the terminal device to perform communication services, and the working mode is AI mode , the AI model is used to instruct the terminal device to use the first AI model to perform a communication service.
  • the terminal device can be instructed to use AI technology to perform communication services, and the consistency of the working mode between the terminal device and the network device can be achieved.
  • the working mode configuration information further includes an effective time of the first AI model.
  • the network device and the terminal device can switch to the AI mode synchronously, so that the time consistency of the working modes between the terminal device and the network device is achieved.
  • an artificial intelligence AI model transmission method is provided, and the method may be executed by a network device or by components of the network device.
  • the method may be implemented through the following steps: the network device receives first information from the terminal device, the first information includes the AI model format supported by the terminal device, and the AI model format supported by the terminal device includes the first AI model format
  • the network device sends second information to the terminal device, the second information is used to indicate the format of the first AI model, and the acquisition method information of the first AI model represented by the format of the first AI model,
  • the first AI model format is determined according to the first information.
  • the acquisition method information includes a download address of the first AI model.
  • the network device receives third information from the terminal device, where the third information is used to indicate whether the terminal device uses the first AI model to perform a communication service.
  • the AI model format includes one or more of the following: a model format output by a deep learning framework, an intermediate representation layer model format, or a model format represented by an executable file.
  • the first information further includes a first threshold and/or a second threshold
  • the first threshold is the maximum delay of the execution part of the AI model when the terminal device uses the AI model to execute communication services
  • the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to execute the communication service.
  • the first AI model satisfies the first hardware requirement of the terminal device, and the first hardware requirement includes: the latency requirement indicated by the first threshold, and/or the first threshold An AI model meets the power consumption requirement indicated by the second threshold.
  • the first information further includes one or more of the following: hardware model, hardware version, storage capability, computing capability, and ability to support heterogeneous computing.
  • the first AI model meets a second hardware requirement of the terminal device, and the second hardware requirement includes: a storage space requirement indicated by the storage capability.
  • the first information further includes an AI model compilation capability
  • the AI model compilation capability includes a capability of compiling and converting an AI model format.
  • the network device determines, according to the first information, that the first AI model format is a model format output by a deep learning framework or an intermediate presentation layer model format.
  • the network device determines that the first AI model format is a model format represented by an executable file.
  • the network device sends working mode configuration information to the terminal device, the working mode configuration information is used to configure a working mode for the terminal device to perform communication services, the working mode is an AI mode,
  • the AI model is used to instruct the terminal device to use the first AI model to perform communication services.
  • the terminal device can be instructed to use AI technology to perform communication services, and the consistency of the working mode between the terminal device and the network device can be achieved.
  • the working mode configuration information further includes an effective time of the first AI model.
  • the network device and the terminal device can switch to the AI mode synchronously, so that the time consistency of the working modes between the terminal device and the network device can be achieved.
  • an artificial intelligence AI model transmission method may be executed by the first device, or may be executed by components of the first device.
  • the first device may be a core network device or an application server.
  • the method can be implemented through the following steps: the first device receives a request message from the terminal device; according to the different functions of the request message, the first device can implement the following two solutions. One is that the first device judges whether the first AI model meets the hardware requirements of the terminal device according to the request message, and the first device sends fourth information to the terminal device, and the fourth information includes the first A judgment result of whether the AI model satisfies the hardware requirements of the terminal device.
  • the first device determines the first AI model represented by the first AI model format according to the request message, and optionally, the first device simultaneously judges whether the first AI model satisfies the hardware requirements of the terminal device. Requirements; the first device sends fourth information to the terminal device, and if the first AI model meets the hardware requirements of the terminal device, the fourth information includes the first AI represented by the first AI model format model; if the first AI model does not meet the hardware requirements of the terminal device, the fourth information includes a judgment result that the first AI model does not meet the hardware requirements of the terminal device.
  • the request message includes an indication of the first AI model and /or the first AI model format.
  • the indication of the first AI model is used to indicate the first AI model to be acquired.
  • a model to be delivered by the first device is indicated through the indication of the first AI model and/or the format of the first AI model.
  • the indication of the first AI model may be, for example, the identification of the first AI model.
  • the first device and the terminal device pre-negotiate the correspondence between multiple AI models and multiple identifications, so that the first device can select the required AI model according to the indication of the first AI model.
  • the AI model delivered to the terminal device is used to indicate whether the first AI model meets the hardware requirements of the terminal device, in a possible design.
  • the request message is used to request the first AI model expressed in the first AI model format.
  • the request message is used to request the first AI model represented by the first AI model format, in a possible design, the request message is also used to indicate the first AI model format.
  • the first AI model format may be a model format output by a deep learning framework, an intermediate representation layer model format, or a model format represented by an executable file.
  • the first AI model format is a model format output by a deep learning framework, an intermediate presentation layer model format, or a model format represented by an executable file.
  • the request message carries a first threshold
  • the first threshold is a maximum delay of an execution part of the AI model when the terminal device uses the AI model to execute communication services.
  • the judging whether the first AI model satisfies the hardware requirements of the terminal device may be implemented through the following steps: when the first AI model is used to execute communication services, the time delay of the execution part of the AI model is not When the first threshold is exceeded, it is determined that the first AI model meets the delay requirement of the terminal device; and/or, when the first AI model is used to execute communication services, the delay of the execution part of the AI model exceeds the When the first threshold is reached, it is determined that the first AI model does not meet the delay requirement of the terminal device.
  • the request message carries a second threshold
  • the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to execute the communication service.
  • the judging whether the first AI model meets the hardware requirements of the terminal device may be implemented through the following steps: when using the first AI model to execute communication services, the power consumption of the execution part of the AI model is less than When the second threshold is exceeded, it is determined that the first AI model meets the power consumption requirement of the terminal device; and/or, when the first AI model is used to execute communication services, the power consumption of the execution part of the AI model exceeds the When the second threshold is reached, it is determined that the first AI model does not meet the power consumption requirement of the terminal device.
  • the request message further includes one or more of the following: hardware model, hardware version, storage capability, computing capability, or capability to support heterogeneous computing.
  • the judging whether the first AI model meets the hardware requirements of the terminal device may be implemented through the following steps: the storage space occupied by the first AI model does not exceed the storage capacity of the terminal device.
  • the indicated storage space is specified, it is determined that the first AI model meets the storage space requirement of the terminal device; and/or, when the storage space occupied by the first AI model exceeds the storage capacity indicated by the storage capacity of the terminal device When there is more space, it is determined that the first AI model does not meet the storage space requirement of the terminal device.
  • a communication device in a fourth aspect, may be a terminal device, or may be a component (for example, a chip, or a chip system, or a circuit) located in the terminal device.
  • the device has the function of implementing the first aspect and the method in any possible design of the first aspect.
  • the functions may be implemented by hardware, or may be implemented by executing corresponding software through hardware.
  • Hardware or software includes one or more modules corresponding to the above-mentioned functions.
  • the device may include a processing unit and a transceiver unit.
  • the transceiver unit is used to send first information to the network device, the first information includes the AI model format supported by the terminal device, and the AI model format supported by the terminal device includes the first AI model format; and the transceiver The unit is further configured to receive second information from the network device, where the second information is used to indicate the format of the first AI model and the acquisition method information of the first AI model represented by the first AI model format, The first AI model format is determined according to the first information. More detailed descriptions of the above processing unit and the transceiver unit can be directly obtained by referring to the relevant description in the above first aspect. For the beneficial effects of the third aspect and various possible designs, reference may be made to the description of the corresponding part of the first aspect.
  • a communication device may be a network device, or may be a component (for example, a chip, or a chip system, or a circuit) located in the network device.
  • the device has the function of realizing the above-mentioned second aspect and the method in any possible design of the second aspect. Functions can be realized by hardware, and can also be realized by executing corresponding software through hardware. Hardware or software includes one or more modules corresponding to the above-mentioned functions.
  • the device may include a processing unit and a transceiver unit.
  • the transceiver unit is configured to receive first information from the terminal device, the first information includes an AI model format supported by the terminal device, and the AI model format supported by the terminal device includes a first AI model format; and The transceiver unit is further configured to send second information to the terminal device, the second information is used to indicate the format of the first AI model, and the acquisition method information of the first AI model represented by the format of the first AI model, The first AI model format is determined according to the first information. More detailed descriptions of the processing unit and the transceiver unit can be directly obtained by referring to the relevant descriptions in the second aspect above. For the beneficial effects of the fourth aspect and various possible designs, reference may be made to the description of the corresponding part of the second aspect.
  • a communication apparatus may be a first device, or may be a component (for example, a chip, or a chip system, or a circuit) located in the first device.
  • the first device may be a core network device or an application server.
  • the device has the function of implementing the third aspect and the method in any possible design of the third aspect.
  • the functions may be implemented by hardware, or may be implemented by executing corresponding software through hardware.
  • Hardware or software includes one or more modules corresponding to the above-mentioned functions.
  • the device may include a processing unit and a transceiver unit.
  • the transceiver unit is used to receive a request message from the terminal device; the processing unit is used to judge whether the first AI model satisfies the hardware requirements of the terminal device according to the request message, and/or convert the first AI model to the terminal device.
  • the first AI model expressed in the AI model format is sent to the terminal device.
  • the transceiver unit is further configured to send fourth information to the terminal device, where the fourth information includes a judgment result of whether the first AI model satisfies the hardware requirements of the terminal device and/or the first AI model expressed in the format of the first AI model.
  • An AI model More detailed descriptions of the processing unit and the transceiver unit can be directly obtained by referring to the relevant descriptions in the third aspect above. For the beneficial effects of the fourth aspect and each possible design, reference may be made to the description of the corresponding part of the third aspect.
  • the embodiment of the present application provides a communication device, where the communication device includes an interface circuit and a processor, and the processor and the interface circuit are coupled to each other.
  • the processor implements the method described in the above first aspect and each possible design of the first aspect through a logic circuit or executing code instructions.
  • the interface circuit is used to receive signals from other communication devices other than the communication device and transmit to the processor or send signals from the processor to other communication devices other than the communication device. It can be understood that the interface circuit may be a transceiver or an input/output interface.
  • the communication device may further include a memory for storing instructions executed by the processor, or storing input data required by the processor to execute the instructions, or storing data generated after the processor executes the instructions.
  • the memory may be a physically independent unit, or may be coupled with the processor, or the processor includes the memory.
  • the embodiment of the present application provides a communication device, where the communication device includes an interface circuit and a processor, and the processor and the interface circuit are coupled to each other.
  • the processor implements the method described in the above second aspect and each possible design of the second aspect through a logic circuit or executing code instructions.
  • the interface circuit is used to receive signals from other communication devices other than the communication device and transmit to the processor or send signals from the processor to other communication devices other than the communication device. It can be understood that the interface circuit may be a transceiver or an input/output interface.
  • the communication device may further include a memory for storing instructions executed by the processor, or storing input data required by the processor to execute the instructions, or storing data generated after the processor executes the instructions.
  • the memory may be a physically independent unit, or may be coupled with the processor, or the processor includes the memory.
  • the embodiment of the present application provides a communication device, where the communication device includes an interface circuit and a processor, and the processor and the interface circuit are coupled to each other.
  • the processor implements the method described in the above third aspect and each possible design of the third aspect through a logic circuit or executing code instructions.
  • the interface circuit is used to receive signals from other communication devices other than the communication device and transmit to the processor or send signals from the processor to other communication devices other than the communication device. It can be understood that the interface circuit may be a transceiver or an input/output interface.
  • the communication device may further include a memory for storing instructions executed by the processor, or storing input data required by the processor to execute the instructions, or storing data generated after the processor executes the instructions.
  • the memory may be a physically independent unit, or may be coupled with the processor, or the processor includes the memory.
  • the embodiment of the present application provides a computer-readable storage medium, where a computer program or readable instruction is stored in the computer-readable storage medium, and when the computer program or readable instruction is executed by a communication device, the The methods described in the above aspects or in each possible design of the aspects are executed.
  • the embodiment of the present application provides a chip system, where the chip system includes a processor and may further include a memory.
  • the memory is used to store programs, instructions or codes; the processor is used to execute the programs, instructions or codes stored in the memory, so as to implement the methods described in the above aspects or possible designs of each aspect.
  • the system-on-a-chip may consist of chips, or may include chips and other discrete devices.
  • a computer program product containing instructions which, when executed by a communication device, causes the method described in the first aspect or each possible design of the aspect to be executed.
  • Fig. 1 is a schematic diagram of the system architecture in the embodiment of the present application.
  • FIG. 2 is a schematic diagram of a process of wireless communication between a network device and a terminal device using an AI model in an embodiment of the present application;
  • Fig. 3 is one of the schematic flow charts of the AI model transmission method in the embodiment of the present application.
  • FIG. 4 is a schematic flow diagram of a terminal device acquiring a first AI model according to acquisition method information in an embodiment of the present application
  • Fig. 5 is the second schematic flow diagram of the AI model transmission method in the embodiment of the present application.
  • FIG. 6 is a schematic flowchart of a method in which a network device indicates a working mode configuration to a terminal device in an embodiment of the present application
  • Fig. 7 is the third schematic flow diagram of the AI model transmission method in the embodiment of the present application.
  • FIG. 8 is a schematic flowchart of a method for assisting the server in determining the format of the first AI model in the embodiment of the present application
  • FIG. 9 is the fourth schematic flow diagram of the AI model transmission method in the embodiment of the present application.
  • FIG. 10 is one of the structural schematic diagrams of the communication device in the embodiment of the present application.
  • Fig. 11 is the second structural diagram of the communication device in the embodiment of the present application.
  • Fig. 12 is the fifth schematic flow diagram of the AI model transmission method in the embodiment of the present application.
  • Fig. 13 is the sixth schematic flow diagram of the AI model transmission method in the embodiment of the present application.
  • Fig. 14 is the seventh schematic flow diagram of the AI model transmission method in the embodiment of the present application.
  • Fig. 15 is the eighth schematic flow diagram of the AI model transmission method in the embodiment of the present application.
  • FIG. 16 is the ninth schematic flow diagram of the AI model transmission method in the embodiment of the present application.
  • the present application provides an artificial intelligence AI model transmission method and device, in order to better apply AI technology in wireless communication systems.
  • the method and the device are based on the same technical conception. Since the principle of solving the problem of the method and the device is similar, the implementation of the device and the method can be referred to each other, and the repetition will not be repeated.
  • the artificial intelligence AI model transmission method provided by the embodiment of the present application can be applied to a 5G communication system, such as a 5G new air interface (new radio, NR) system, and can be applied to various application scenarios of a 5G communication system, such as enhanced mobile broadband (enhanced mobile broadband (eMBB), ultra reliable low latency communication (URLLC) and enhanced machine-type communication (eMTC).
  • a 5G communication system such as enhanced mobile broadband (enhanced mobile broadband (eMBB), ultra reliable low latency communication (URLLC) and enhanced machine-type communication (eMTC).
  • eMBB enhanced mobile broadband
  • URLLC ultra reliable low latency communication
  • eMTC enhanced machine-type communication
  • the artificial intelligence AI model transmission method provided in the embodiment of the present application can also be applied to various communication systems that evolve in the future, such as the sixth generation (6th generation, 6G) communication system, and another example is an air-space-sea-ground integrated communication system.
  • the artificial intelligence AI model transmission method provided in the embodiment of the present application can also be applied to communication between base stations, communication between terminal equipment, communication between terminal equipment, Internet of Vehicles, Internet of Things, industrial Internet, satellite communication, etc., for example, can Applied to device-to-device (Device-to-Device, D2D), vehicle-to-everything (V2X), machine-to-machine (machine-to-machine, M2M) communication systems.
  • D2D device-to-device
  • V2X vehicle-to-everything
  • M2M machine-to-machine
  • FIG. 1 shows a system architecture applicable to this embodiment of the present application, and the communication system architecture includes a network device 101 and a terminal device 102 .
  • the network device 101 provides services for the terminal devices 102 within coverage.
  • the network device 101 provides wireless access for one or more terminal devices 102 within the coverage of the network device 101 .
  • the system architecture may further include a first device 103, and the first device 103 may be a core network device or an application server.
  • the first device 103 may also be called an AI model server, and the first device 103 may compile the AI model, and compile one AI model format into another AI model format.
  • the first device 103 may also evaluate the AI model to evaluate whether the AI model can satisfy the target hardware condition of the terminal device.
  • the first device 103 may also store or maintain the AI model, and provide the AI model to the terminal device.
  • the network device 101 is a node in a radio access network (radio access network, RAN), and may also be called a base station, and may also be called a RAN node (or device).
  • RAN radio access network
  • examples of some network devices 101 are: next generation base station (next generation nodeB, gNB), next generation evolved base station (next generation evolved nodeB, Ng-eNB), transmission reception point (transmission reception point, TRP), evolved Node B (evolved Node B, eNB), radio network controller (radio network controller, RNC), node B (Node B, NB), base station controller (base station controller, BSC), base transceiver station (base transceiver station, BTS), home base station (for example, home evolved NodeB, or home Node B, HNB), base band unit (base band unit, BBU), or wireless fidelity (wireless fidelity, Wifi) access point (access point, AP),
  • the network device 101 may also be a satellite, and the satellite
  • the network device 101 may also be other devices with network device functions, for example, the network device 101 may also be a device to device (device to device, D2D) communication, vehicle networking, or machine to machine (machine to machine, M2M) communication. A device that functions as a network device.
  • the network device 101 may also be any possible network device in the future communication system.
  • the network device 101 may include a centralized unit (CU) and a distributed unit (DU).
  • the network device may also include an active antenna unit (AAU).
  • CU implements some functions of network equipment
  • DU implements some functions of network equipment.
  • CU is responsible for processing non-real-time protocols and services, implementing radio resource control (radio resource control, RRC), packet data convergence layer protocol (packet data convergence protocol, PDCP) layer functions.
  • the DU is responsible for processing physical layer protocols and real-time services, realizing the functions of the radio link control (radio link control, RLC) layer, media access control (media access control, MAC) layer and physical (physical, PHY) layer.
  • the AAU implements some physical layer processing functions, radio frequency processing and related functions of active antennas.
  • the network device may be a device including one or more of a CU node, a DU node, and an AAU node.
  • the CU can be divided into network devices in an access network (radio access network, RAN), and the CU can also be divided into network devices in a core network (core network, CN), which is not limited in this application.
  • Terminal equipment 102 also referred to as user equipment (user equipment, UE), mobile station (mobile station, MS), mobile terminal (mobile terminal, MT), etc., is a device that provides voice and/or data connectivity to users. equipment.
  • the terminal device 102 includes a handheld device with a wireless connection function, a vehicle-mounted device, etc. If the terminal device 102 is located on the vehicle (for example, placed in the vehicle or installed in the vehicle), it can be considered as a vehicle-mounted device, and the vehicle-mounted device is also called It is an on-board unit (onBoard unit, OBU).
  • OBU onBoard unit
  • the terminal device 102 can be: mobile phone (mobile phone), tablet computer, notebook computer, palmtop computer, mobile internet device (mobile internet device, MID), wearable device (such as smart watch, smart bracelet, pedometer, etc. ), vehicle-mounted equipment (such as automobiles, bicycles, electric vehicles, airplanes, ships, trains, high-speed rail, etc.), virtual reality (virtual reality, VR) equipment, augmented reality (augmented reality, AR) equipment, industrial control (industrial control) Wireless terminals in smart home equipment (such as refrigerators, TVs, air conditioners, electric meters, etc.), intelligent robots, workshop equipment, wireless terminals in self driving, wireless terminals in remote medical surgery , wireless terminals in smart grid, wireless terminals in transportation safety, wireless terminals in smart city, or wireless terminals in smart home, flight equipment (such as , intelligent robots, hot air balloons, drones, airplanes), etc.
  • vehicle-mounted equipment such as automobiles, bicycles, electric vehicles, airplanes, ships, trains, high-speed rail, etc.
  • virtual reality virtual reality
  • the terminal device 102 may also be other devices having terminal device functions, for example, the terminal device 102 may also be a device-to-device (device to device, D2D) communication, a vehicle networking or a machine-to-machine (machine-to-machine, M2M) communication A device that functions as a terminal device in a device.
  • the network device functioning as a terminal device can also be regarded as a terminal device.
  • the terminal device 102 may also be a wearable device.
  • Wearable devices can also be called wearable smart devices or smart wearable devices, etc., which is a general term for the application of wearable technology to intelligently design daily wear and develop wearable devices, such as glasses, gloves, watches, clothing and shoes wait.
  • a wearable device is a portable device that is worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not only a hardware device, but also achieve powerful functions through software support, data interaction, and cloud interaction.
  • Generalized wearable smart devices include full-featured, large-sized, complete or partial functions without relying on smart phones, such as smart watches or smart glasses, etc., and only focus on a certain type of application functions, and need to cooperate with other devices such as smart phones Use, such as various smart bracelets, smart helmets, smart jewelry, etc. for physical sign monitoring.
  • the means for realizing the functions of the terminal device 102 are, for example, chips, wireless transceivers, and chip systems, and the means for realizing the functions of the terminal device 102 may be installed or configured or deployed in the terminal device 102 .
  • AI is a technological capability that simulates human cognition through machines.
  • the core capability of AI is to make judgments or predictions based on given input.
  • AI applications can be divided into four major categories: perception, cognition, creativity, and intelligence.
  • perception capabilities include image recognition, face recognition, speech recognition, natural language processing, machine translation, or text conversion.
  • Cognitive capabilities such as spam identification, credit risk analysis, intelligent natural disaster prediction and prevention, AI playing Go or machine learning, etc. Creativity such as AI composition, AI painting or AI design.
  • Intelligence refers to the ability to seek the real truth, distinguish right from wrong, and guide humans to live a meaningful life through a deep understanding of the truth about people, things, or things. This field involves human self-awareness, self-cognition and values, and is the most difficult for humans to imitate. a field of .
  • AI models can be implemented based on neural network models.
  • the artificial neural network (ANNs) model also known as the neural network (NNs) model or the connection model, is a typical representative of the AI model.
  • the neural network model is a mathematical calculation model that imitates the behavioral characteristics of the human brain neural network and performs distributed parallel information processing. Its main task is to learn from the principles of the human brain neural network, build a practical artificial neural network according to application requirements, realize the design of learning algorithms suitable for application requirements, simulate the intelligent activities of the human brain, and then solve practical problems technically.
  • the neural network relies on the complexity of the network structure, and realizes the design of the corresponding learning algorithm by adjusting the interconnection relationship between a large number of internal nodes.
  • the AI model format can be understood as the format used to represent the AI model.
  • AI model formats are described below.
  • the first type of format is the model format output by the deep learning framework.
  • Deep learning frameworks such as MindSpore, Tensorflow, PyTorch, PaddlePaddle.
  • the model format output by the deep learning framework has nothing to do with the running hardware of the AI model.
  • the second type of format is the intermediate representation (intermediate representation, IR) layer model format.
  • the IR layer model format has nothing to do with the deep learning framework, nor with the running hardware of the AI model.
  • the first type of format and the second type of format are AI model formats that have nothing to do with the operating hardware of the AI model, and can be collectively referred to as the high-level representation or high-level format of the AI model.
  • the third type of format is the model format represented by the executable file.
  • Executable files are files that can be run on hardware. After the high-level representation of the AI model is compiled for the running hardware of the AI model, executable files related to the specific running hardware can be obtained.
  • the third type of format is the format related to the operating hardware of the AI model, which can be called the low-level representation or low-level format of the AI model.
  • the high-level representation refers to the format of the AI model that has nothing to do with the running hardware.
  • it can also be other AI model formats that have nothing to do with the running hardware.
  • the low-level representation can be an AI model format directly related to the running hardware, and other AI model formats related to the running hardware besides the above-mentioned model format represented by the executable file. This application takes the first type format, the second type format and the third type format as examples for introduction.
  • operating hardware for terminal equipment, for example, central processing unit (central processing unit, CPU), microprocessor (graphics processing unit, GPU), embedded neural network processor (neural-network processing unit, NPU) , or field programmable gate array (field programmable gate array, FPGA), etc.
  • the terminal device supports heterogeneous computing
  • the operating hardware may also be a combination of the above types.
  • Heterogeneous computing means that AI models are distributed and executed on multiple types of computing units. For example, AI models are distributed and executed on three types of computing units: CPU, GPU, and FPGA. Different types of running hardware can correspond to different model formats represented by executable files.
  • the operating hardware of the terminal device is a CPU of a specific manufacturer or category, and the model format represented by the executable file is the model format corresponding to the CPU; for another example, the operating hardware of the terminal device is heterogeneous of a certain GPU and a certain FPGA, The model format represented by the executable file is the model format corresponding to the GPU and the model format corresponding to the FPGA.
  • Examples include AI-based channel estimation and signal detection.
  • the signal detection can be the process of extracting the received signal containing interference noise from the wireless channel;
  • the channel estimation is the process of estimating the model parameters of an assumed channel model from the received signal.
  • AI-based end-to-end communication link design is another example.
  • AI-based channel state information (CSI) feedback scheme which encodes the CSI through a neural network and feeds it back to the network device.
  • the network device sends the AI model to the terminal device.
  • the AI model may also be called an AI encoder.
  • the terminal device receives the AI coder from the network device, or the terminal device downloads the AI coder from the network device; in another possible implementation, the terminal device may also obtain the AI coder from other devices, which can be the core network device or application server.
  • the terminal device uses the AI encoder for CSI encoding, and sends the encoded data to the network device, and the network device uses the AI model to restore the received signal, wherein the AI model used by the network device can be called AI decoder.
  • an embodiment of the present application provides an AI model transmission method, and the flow of the method is as follows.
  • the terminal device sends first information to the network device, and correspondingly, the network device receives the first information from the terminal device.
  • the first information includes the AI model format supported by the terminal device, and the AI model format supported by the terminal device includes the first AI model format.
  • the network device sends the second information to the terminal device, and correspondingly, the terminal device receives the second information from the network device.
  • the second information is used to indicate the acquisition method information of the first AI model.
  • the first AI model is represented by a first AI model format, and the first AI model format is determined according to the first information, or, the first AI model is applied to the first communication scenario.
  • the terminal device reports the supported AI model format to the network device, so that the network device can know the AI model format supported by the terminal device, so that the network device will determine the first AI model according to the AI model format supported by the terminal device format, the first AI model format must be the model format supported by the terminal device.
  • the network device indicates to the terminal device the acquisition method information (or download information) of the first AI model represented by the first AI model format, so that the model format of the first AI model downloaded by the terminal device according to the acquisition method information must be supported by itself It ensures that the model format of the AI model for wireless communication services between the terminal device and the network device is consistent, and the terminal device can understand or recognize the AI model format, ensuring the feasibility of using the AI model for communication services.
  • the network device sends the acquisition method information of the first AI model to the terminal device, and the terminal device can download the first AI model to the first device according to the acquisition method information.
  • the first device may be a server or a core network device. In this way, through the first The device stores and maintains the AI model.
  • the number of devices can be less than the number of network devices, reducing maintenance costs, reducing the overhead and power consumption of network devices, and improving the performance of the communication system.
  • S303 may also be included.
  • the terminal device acquires the first AI model according to the acquisition method information.
  • the acquisition method information may be the download address of the AI model, for example, it may be the identifier of the first device, or it may be a download link.
  • the terminal device may acquire the first AI model according to the acquisition method information through the following steps.
  • the terminal device sends a request message to the first device, and correspondingly, the first device receives the request message from the terminal device.
  • the request message is used to request the first device to judge whether the first AI model satisfies the hardware requirements of the terminal device, and it can be understood that the request message is used to request the first device to evaluate the first AI model.
  • the request message may also be used to request the AI model expressed in the first AI model format.
  • the request message is used to request the first device to compile the AI model format.
  • the request message may be used to request multiple of the above types.
  • the request message may request the first device to determine whether the first AI model meets the hardware requirements of the terminal device, and may be used to request the AI model represented by the first AI model format. Model.
  • the request message may be used to indicate the first communication scenario.
  • the first device may be a core network device, and when the terminal device sends a request message to the core network device, it may send the request message to the core network device through a non-access stratum (NAS).
  • the first device may be an application server, and the terminal device may send a request message to the application server through the application layer.
  • the first device performs an operation corresponding to the request message according to the request message.
  • the first device judges whether the first AI model meets the hardware requirements of the terminal device according to the request message.
  • the first device determines the first AI model expressed in the first AI model format according to the request message.
  • the request message may carry an indication of the first AI model, where the indication of the first AI model is used to indicate the first AI model to be acquired by the terminal device.
  • the first device may store multiple AI models and determine the first AI model represented by the first AI model format according to the indication of the first AI model in the request message.
  • the request message may also carry the first AI model format, and the first device determines the first AI model in the first AI model format according to the first AI model format.
  • the first device compiles the first AI model format according to the request message, so as to obtain the low-level AI model format.
  • the first AI model format is a high-level AI model format, a model format output by a deep learning framework, or an intermediate representation IR layer model format.
  • the compiled AI model format is an AI model format represented by a low-level representation, such as an AI model format represented by an executable file.
  • the function of the first device is to compile the high-level AI model format into the low-level AI model format.
  • the first device determines the first AI model used by the terminal device in the first communication scenario according to the request message.
  • the first device sends fourth information to the terminal device, and the terminal device receives the fourth information from the first device.
  • the request message is used to request the first device to judge whether the first AI model meets the hardware requirements of the terminal device, and the first device determines that the first AI model meets the hardware requirements of the terminal device, then the fourth information indicates that the first AI model meets the terminal device's hardware requirements hardware requirements.
  • the request message is used to request the first device to determine whether the first AI model meets the hardware requirements of the terminal device, and the first device determines that the first AI model does not meet the hardware requirements of the terminal device, then the fourth information indicates that the first AI model does not meet the hardware requirements of the terminal device. Hardware requirements for end devices.
  • the first device sends the first AI model to the terminal device.
  • the fourth information includes the first AI model, or the fourth information is the first AI model, or the fourth information indicates the first AI model. It can be understood that if the request message is used to request the AI model represented by the first AI model format, the first device may also first determine whether the first AI model meets the hardware requirements of the terminal device, and if so, send the first AI model to the terminal device. If an AI model is not satisfied, the first AI model is not sent, or the terminal device may be notified of an AI model that does not meet the hardware requirements of the terminal device.
  • the first device sends the compiled AI model represented by the low layer to the terminal device.
  • the fourth information includes the first AI model used by the terminal device in the first communication scenario.
  • the terminal device sends third information to the network device, and correspondingly, the network device receives the third information from the terminal device.
  • the third information is used to indicate whether the terminal device uses the first AI model to execute the communication service.
  • the terminal device can obtain the first AI model through S401-S403, and can determine whether the first AI model matches the hardware capability of the terminal device according to the fourth information received from the first device. The terminal device determines whether to use the first AI model to execute the communication service according to the fourth information.
  • the terminal device may use the traditional communication mode to perform the communication service.
  • the third information indicates that the terminal device does not use the first AI model to perform the communication service, it may also be understood that the third information indicates that the terminal device uses the traditional communication mode to perform the communication service.
  • the network device returns confirmation information to the terminal device, and the terminal device receives the confirmation information from the network device.
  • the confirmation information may be response information to the third information.
  • the confirmation information may be used to instruct the network device to use the first AI model to perform the communication service.
  • the confirmation information may be used to instruct the network device to use the traditional communication mode to perform the communication service.
  • the AI model includes the neural network structure, the parameters (weight, bias, etc.), code, or configuration corresponding to each operator in the neural network structure.
  • the AI model may contain a large amount of parameters and calculations.
  • communication services often have requirements for delay.
  • network devices and terminal devices use AI technology for wireless communication, the resources and computing power of the terminal device need to be able to support the operation of the AI model, including the storage space of the terminal device to accommodate the storage of the AI model, and the technical capabilities to support when required.
  • the calculation of the AI model is completed within the delay, and the energy consumed by the calculation of the AI model does not exceed the limit of the terminal device, otherwise the AI technology will not be applied.
  • the first AI model may need to meet some hardware requirements of the terminal device.
  • the network device determines the first AI model format, it also needs to consider whether the AI model in the first AI model format meets some hardware requirements of the terminal device.
  • the first AI model format is determined according to the first information.
  • An optional implementation manner in which the network device determines the first AI model format according to the first information will be described below.
  • the first information may also be called AI capability information of the terminal device.
  • the first information may also include a first threshold.
  • the first threshold is: when the terminal device uses the AI model to execute the communication service, the maximum time delay of the execution part of the AI model.
  • a communication service includes multiple parts, and the terminal device uses an AI model to execute one or more of the multiple parts in the communication service.
  • the terminal device may use the AI model to encode CSI, and the network device may use the AI model to perform CSI decoding.
  • the first threshold represents the maximum delay when the terminal device performs CSI encoding using the AI model. If the delay of the terminal device using the first AI model for CSI encoding exceeds the maximum delay indicated by the first threshold, then It is considered that applying the first AI model to the communication service cannot meet the delay requirement.
  • the first information may further include a second threshold, where the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to execute the communication service.
  • the communication service includes some services executed by the AI model
  • the first threshold is the maximum power consumption of the part of the services executed by the AI model.
  • the first information also includes indication information of the AI model compilation capability, and the AI model compilation capability includes the capability of compiling and converting the format of the AI model.
  • the first information also includes hardware capability of the terminal device, and the hardware capability may be one or more combinations of the following: hardware model, hardware version, storage capability, computing capability, main frequency, or capability to support heterogeneous computing.
  • the network device may determine the first AI model format according to the first information.
  • the first AI model in the first AI model format needs to meet the hardware requirements of the terminal device.
  • the network device determines that the first AI model in the first AI model format meets the hardware requirements of the terminal device, it sends the second information to the terminal device.
  • the terminal device requests the first device to determine whether the first AI model meets the hardware requirements of the terminal device. Hardware requirements for end devices.
  • the network device may determine whether the first AI model meets the first hardware requirement of the terminal device, where the first hardware requirement is the delay requirement indicated by the first threshold.
  • the terminal device uses the first AI model to perform communication services, if the delay generated by the calculation of the first AI model does not exceed the first threshold, it means that the first AI model meets the delay requirement indicated by the first threshold; otherwise, the first The AI model does not meet the delay requirement indicated by the first threshold.
  • the terminal device supports heterogeneous computing capabilities, multiple computing units of heterogeneous computing can correspond to multiple model files of the first AI model, and one computing unit corresponds to a model file of the first AI model, or it can be heterogeneous
  • the multiple calculation units for calculation correspond to a model file of the first AI model.
  • Heterogeneous computing in this application may refer to decomposing the computing tasks of the AI model into multiple sub-tasks, and the computing of the multiple sub-tasks is executed on suitable running hardware respectively. These executions should usually be parallel, but serial ones are not excluded. . Therefore, the basis for judging the delay is that all calculations of the multiple subtasks of the AI model have been completed. In the case of heterogeneous computing, it is necessary to judge whether the terminal device has completed the calculation of the AI model on all computing units, and whether the generated delay exceeds the first threshold. If it does not exceed the first threshold, it means the first AI The model satisfies the delay requirement indicated by the first threshold; otherwise, the first AI model does not meet the delay requirement indicated by the first threshold.
  • the network device may determine whether the first AI model meets the first hardware requirement of the terminal device, and the first hardware requirement is the power consumption requirement indicated by the second threshold.
  • the terminal device uses the first AI model to perform communication services, if the power consumption generated by the calculation of the first AI model does not exceed the second threshold, it means that the first AI model meets the power consumption requirement indicated by the second threshold; otherwise, the first The AI model does not meet the power consumption requirement indicated by the second threshold.
  • the terminal device supports heterogeneous computing capabilities, it is necessary to determine whether the total power consumption generated by the terminal device when performing communication services on multiple computing units of heterogeneous computing exceeds the second threshold. When the power consumption does not exceed the second threshold, it means that the first AI model meets the power consumption requirement indicated by the second threshold; otherwise, the first AI model does not meet the power consumption requirement indicated by the second threshold.
  • the first information includes the storage capability of the terminal device, it may be determined whether the first AI model meets the second hardware requirement of the terminal device, and the second hardware requirement is the storage space requirement indicated by the storage capability.
  • the storage space occupied by the first AI model is not larger than the storage space indicated by the storage capacity, it means that the first AI model meets the storage space requirement indicated by the storage capacity; otherwise, the first AI model does not meet the storage space requirement indicated by the storage capacity .
  • the network device may The hardware capability indicated by the information determines the computing power (that is, computing power) of the terminal device, and further judges whether the computing power of the terminal device can support the operation of the first AI model, that is, whether the first AI model is used based on the computing power of the terminal device Meet latency requirements or power consumption requirements.
  • the network device can determine that the first AI model is an AI model represented by a high-level representation, such as a model output by a deep learning framework format or intermediate presentation layer model format. If the network device determines that the terminal device does not have the AI model compilation capability, the network device may determine that the first AI model is an AI model represented by a low layer, for example, a model format represented by an executable file.
  • the AI model compilation capability includes the capability of compiling and converting the format of the AI model, for example, compiling and converting the high-level AI model into the low-level AI model.
  • the network device may determine according to the first information that the terminal device has the AI model compilation capability. If the first information does not include the indication information of the AI model compilation capability or the first information indicates that the terminal device does not have the AI model compilation capability, the network device may determine that the terminal device does not have the AI model compilation capability.
  • the network device may not expect the AI model to be acquired by the terminal device, that is, the network device does not expect the AI model to be exposed.
  • the low-level AI model is not easily decompiled to obtain the original AI model, and the low-level AI model is more secure than the high-level AI model.
  • High-level AI models are more likely to be recognized or acquired by terminal devices than low-level AI models, that is, high-level AI models are more likely to be exposed than low-level AI models.
  • Network devices may have different sensitivities to AI model exposure. When the sensitivity of network devices to AI model exposure is lower than the set threshold or when the network device is not sensitive to AI model exposure, the network device determines that the first AI model is a high-level Represents the AI model. When the sensitivity of the network device to the exposure of the AI model is not lower than the set threshold or when the network device is sensitive to the exposure of the AI model, the network device determines that the first AI model is an AI model represented by a low layer.
  • the network device may determine that the first AI model is an AI model represented by a high level.
  • the network device sends the second information to the terminal device.
  • the second information is used to indicate the method information of the first AI model.
  • the network device indicates the working mode to the terminal device, and the working mode is the AI mode.
  • the AI mode is used to instruct the terminal device to use the first AI model to execute the communication service.
  • the network device may send working mode configuration information to the terminal device, where the working mode configuration information is used to configure a working mode for the terminal device to perform communication services, and the working mode is AI mode.
  • the working mode configuration information may be carried in the same signaling as the second information, or may be carried in different signalings.
  • the network device determines the format of the first AI model on the basis that the first AI model meets the hardware requirements of the terminal device. If the first AI model cannot meet the hardware requirements of the terminal device, the network device can instruct the terminal device to Communication services are performed using a legacy communication mode. If there is an error in the information communication between any two of the terminal device, the network device or the first device, the system uses the legacy communication mode to perform communication services according to the instructions of the network device, or the system works in the legacy communication mode by default. For example, if the network device cannot recognize the first information reported by the terminal device, the network device uses the traditional communication mode to perform the communication service, and the network device may instruct the terminal device to use the traditional communication mode to perform the communication service.
  • the method for the network device to indicate the working mode configuration to the terminal device can be described through the embodiment in FIG. 6 .
  • the terminal device sends first information to the network device, and correspondingly, the network device receives the first information from the terminal device.
  • the first information includes the AI model format supported by the terminal device, and the AI model format supported by the terminal device includes the first AI model format.
  • This step is the same as S301, and reference may be made to related descriptions of S301.
  • the network device sends working mode configuration information to the terminal device, and correspondingly, the terminal device receives the working mode configuration information from the network device.
  • the working mode configuration information is used to configure the working mode of the terminal device to perform communication services.
  • the working modes include AI mode or non-AI mode.
  • the AI mode is used to indicate the use of the first AI model to perform communication services
  • the non-AI mode is used to indicate the use of traditional communication modes. Carry out communication services.
  • the terminal device can use the first AI model to perform communication services; when the working mode configured in the working mode configuration information is the non-AI mode, the terminal device can use the traditional communication mode to perform communication business.
  • the working mode configuration information may also include the effective time of the first AI model.
  • the terminal device may use the first AI model to execute the communication service within the effective time according to the effective time of the first AI model.
  • the terminal device can request the first device to use the first AI model format through the embodiment of FIG. An AI model or request evaluates whether the first AI model matches its own capabilities.
  • the request message sent by the terminal device to the first device may carry the parameters carried in the above-mentioned first information.
  • the request message includes one or more of the following: first threshold, second threshold, hardware model, hardware version, storage capability, computing capability, and ability to support heterogeneous computing.
  • the method for the first device to determine whether the first AI model meets the hardware requirements of the terminal device according to the request message may refer to the method for the network device to determine whether the first AI model meets the hardware requirements of the terminal device, which will not be repeated here.
  • the request message may carry a first threshold value.
  • the first device determines that the first AI model meets the delay requirements of the terminal device; when the delay of the execution part of the AI model exceeds the first threshold when the first AI model is used to execute communication services, the first device determines that the first AI model meets the delay requirements of the terminal device; An AI model does not meet the delay requirement of the terminal equipment.
  • the request message carries a second threshold.
  • the first device judges whether the first AI model meets the hardware requirements of the terminal device, if the first AI model is used to execute communication services, the power consumption of the execution part of the AI model does not exceed the second threshold.
  • threshold the first device determines that the first AI model meets the power consumption requirements of the terminal device; It is determined that the first AI model does not meet the power consumption requirement of the terminal device.
  • the first device judges whether the first AI model meets the hardware requirements of the terminal device, if the storage space occupied by the first AI model does not exceed the storage space indicated by the storage capacity of the terminal device, then it is determined that the first AI model meets the requirements of the terminal device.
  • the storage space requirement of the device if the storage space occupied by the first AI model exceeds the storage space indicated by the storage capacity of the terminal device, it is determined that the first AI model does not meet the storage space requirement of the terminal device.
  • the first device can compile the AI model format of the high-level representation and convert it into the model format of the low-level representation .
  • the request message can carry the hardware information of the terminal device, for example, the hardware model of the terminal device, the hardware version of the terminal device, the first device compiles the high-level AI model format according to the hardware information of the terminal device, and converts it into a low-level model Format.
  • the first device may also determine whether the model format of the compiled low-level representation meets the hardware requirements of the terminal device, and if the requirements are met, send fourth information to the terminal device, and the fourth information indicates the model format of the compiled low-level representation ( That is, the first AI model format). If the first device judges whether the first AI model satisfies the storage space requirement, the fourth information may also indicate the remaining storage space.
  • the first device may send fourth information to the terminal device, the fourth information indicates that the first AI model does not meet the hardware requirements of the terminal device, and the fourth The information may also indicate that the terminal device uses a traditional communication mode to conduct communication services.
  • the embodiment of the present application also provides another AI model transmission method, and the flow of the method is as follows.
  • the terminal device sends first information to the network device, and correspondingly, the network device receives the first information from the terminal device.
  • the first information includes the AI model format supported by the terminal device, and the AI model format supported by the terminal device includes the first AI model format.
  • the network device sends the first AI model expressed in the first AI model format to the terminal device, and correspondingly, the terminal device receives the first AI model expressed in the first AI model format from the network device.
  • the first AI model format is determined according to the first information.
  • S703 may also be included.
  • the network device determines a first AI model format according to the first information received in S701.
  • the network device determines the first AI model format according to the first information
  • the network device judges whether the first AI model satisfies the hardware requirements of the terminal device according to the first information.
  • the first information includes a first threshold, and the network device judges whether the first AI model meets the delay requirement according to the first threshold.
  • the first information includes a second threshold, and the network device judges whether the first AI model meets the power consumption requirement according to the second threshold.
  • the first information includes storage capacity, and the network device determines whether the first AI model meets the storage space requirement indicated by the storage capacity according to the storage capacity.
  • the network device may request the first device to determine whether the first AI model meets the hardware requirements of the terminal device.
  • the network device may also request the first device to compile and convert the AI model format.
  • the network device can determine the first AI model format with the assistance of the first device.
  • the method requiring the assistance of the first device to determine the format of the first AI model may be implemented through the following steps.
  • the network device sends a request message to the first device, and correspondingly, the first device receives the request message from the network device.
  • the request message is used to request the first device to judge whether the first AI model meets the hardware requirements of the terminal device, or the request message is used to request the format of the first AI model.
  • the first device performs an operation corresponding to the request message according to the request message.
  • the first device judges whether the first AI model meets the hardware requirements of the terminal device according to the request message.
  • the first device determines the first AI model format according to the request message, for example, the first device compiles and converts the second AI model format into the first AI model format according to the request message.
  • the second AI model format is a high-level representation AI model format, such as a model format output by a deep learning framework, or an intermediate representation IR layer model format.
  • the first AI model format is an AI model format represented by a low layer, such as an AI model format represented by an executable file.
  • the function of the first device is to compile the high-level AI model format into the low-level AI model format.
  • the request message can include the hardware information of the terminal device, such as the hardware model and hardware version.
  • the AI model format represented at the low level is related to the hardware of the terminal device.
  • the first device compiles according to the hardware information of the terminal device to obtain more accurate information. Compile the result.
  • the first device sends fifth information to the terminal device, and the terminal device receives the fifth information from the first device.
  • the request message is used to request the first device to judge whether the first AI model meets the hardware requirements of the terminal device, and the first device determines that the first AI model meets the hardware requirements of the terminal device, then the fifth information indicates that the first AI model meets the terminal device's hardware requirements hardware requirements.
  • the request message is used to request the first device to determine whether the first AI model meets the hardware requirements of the terminal device, and the first device determines that the first AI model does not meet the hardware requirements of the terminal device, then the fifth information indicates that the first AI model does not meet the hardware requirements of the terminal device. Hardware requirements for end devices.
  • the first device sends the first AI model in the first AI model format to the terminal device, the fifth information includes the first AI model, or the fifth information is the first AI model, Alternatively, the fifth information indicates the first AI model.
  • the request message sent by the network device to the first device is used to request the first device to judge whether the first AI model meets the delay requirement, the request message may carry a first threshold, and the first device judges the first AI model according to the first threshold Whether the delay requirement is met. If the request message sent by the network device to the first device is used to request the first device to judge whether the first AI model meets the power consumption requirement, the request message may carry a second threshold, and the first device judges the first AI model based on the second threshold Whether the power consumption requirement is met.
  • the request message sent by the network device to the first device may carry the storage capacity, and the first device judges whether the first AI model meets the storage space requirements according to the first threshold. Meet storage space requirements.
  • the request message may also include one or more of the following: AI model format supported by the terminal device, hardware model, hardware version, storage capability, computing capability, or capability to support heterogeneous computing.
  • the first device judges whether the first AI model meets the hardware requirements of the terminal device according to the hardware capability of the terminal device in the request message.
  • the first device may also compile and convert the second AI model format into the first AI model format conforming to the hardware according to the request message.
  • the second AI model format is a high-level representation AI model format, such as a model format output by a deep learning framework, or an intermediate representation IR layer model format.
  • the first AI model format is an AI model format represented by a low layer, such as an AI model format represented by an executable file.
  • the function of the first device is to compile the high-level AI model format into a hardware-related low-level AI model format.
  • the network device sends the first AI model represented by the first AI model format to the terminal device, which can be understood as the network device indicates the working mode to the terminal device, and the working mode is the AI mode.
  • the network device may send working mode configuration information to the terminal device, where the working mode configuration information is used to configure a working mode for the terminal device to perform communication services, and the working mode is AI mode.
  • the working mode configuration information may be carried in the same signaling as that of the first AI model, or may be carried in different signaling.
  • the network device may instruct the terminal device to use a traditional (legacy) communication mode Carry out communication business. If there is an error in the information communication between any two of the terminal device, the network device or the first device, for example, the network device or the first device cannot recognize the first information reported by the terminal device, the network device or the first device may instruct the terminal device Communication services are performed using a legacy communication mode.
  • the method for the network device to indicate the working mode configuration to the terminal device can be described through the embodiment in FIG. 6 . Reference may be made to the description of the embodiment in FIG. 6 above, and details are not repeated here.
  • the foregoing embodiment in FIG. 8 describes that the network device determines the first AI model through assistance of the first device.
  • the first device evaluates whether the first AI model meets the hardware requirements of the terminal device, it can compile and convert the first AI model into a low-level AI model, and then make a judgment, because the low-level AI model can better reflect the hardware capabilities , so the judgment result will be more accurate.
  • the network device in S703 determines the first AI model format by itself according to the first information. In this case, if the network device evaluates whether it meets the hardware requirements of the terminal device based on the high-level AI model, it may evaluate Results will be inaccurate. As shown in Figure 9, S702, the network device sends the first AI model expressed in the first AI model format to the terminal device, and after the terminal device receives the first AI model expressed in the first AI model format from the network device, the following steps may also be performed .
  • the terminal device judges whether the first AI model matches the computing power of the terminal device. If yes, use the first AI model to perform the communication service, otherwise perform S902.
  • Judging whether the first AI model matches the computing power of the terminal device that is, judging whether the first AI model meets the hardware requirements of the terminal device. For example, when using the first AI model to perform communication services, when the time delay of the execution part of the AI model does not exceed the first threshold, it is determined that the first AI model meets the time delay requirements of the terminal device; when using the first AI model to perform communication services When the time delay of the execution part of the AI model exceeds the first threshold, it is determined that the first AI model does not meet the time delay requirement of the terminal device.
  • the first threshold is the maximum time delay of the execution part of the AI model when the terminal device uses the AI model to execute the communication service.
  • the first AI model when using the first AI model to perform communication services, when the power consumption of the execution part of the AI model does not exceed the second threshold, it is determined that the first AI model meets the power consumption requirements of the terminal device; when using the first AI model to perform communication services When the power consumption of the execution part of the AI model exceeds the second threshold, it is determined that the first AI model does not meet the power consumption requirement of the terminal device.
  • the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to execute the communication service.
  • the storage space occupied by the first AI model does not exceed the storage space indicated by the storage capacity of the terminal device, it is determined that the first AI model meets the storage space requirements of the terminal device;
  • the storage space indicated by the storage capability of the device is determined, it is determined that the first AI model does not meet the storage space requirement of the terminal device.
  • the terminal device may compile the AI model first, and convert the first AI model into a low-level AI model.
  • the terminal device When the first AI model When the model is a model format output by a deep learning framework, or an intermediate representation of a high-level AI model such as an IR layer model format, the terminal device compiles and converts the high-level AI model into a low-level AI model, such as the model format represented by an executable file .
  • the terminal device judges whether the converted low-level AI model matches the computing power of the terminal device.
  • the specific method can be to actually execute the first AI model to obtain information such as time delay and power consumption, or refer to the terminal device to determine the first AI model. The method of whether the AI model satisfies the hardware requirements of the terminal device will not be repeated here.
  • the terminal device sends a mode request message to the network device, where the mode request message is used to request to perform a communication service in a traditional communication mode.
  • the network device receives the mode request message from the terminal device.
  • S903 may also be included.
  • the network device sends a mode confirmation message to the terminal device, where the mode confirmation message is used to indicate and confirm that the traditional communication mode is used to perform the communication service.
  • the terminal device receives the mode confirmation message from the network device.
  • the terminal device judges whether the first AI model matches the computing power of the terminal device, which is more accurate than the judgment result of the network device or the first device.
  • the network device and the terminal device include hardware structures and/or software modules corresponding to each function.
  • the present application can be implemented in the form of hardware or a combination of hardware and computer software with reference to the units and method steps of the examples described in the embodiments disclosed in the present application. Whether a certain function is executed by hardware or computer software drives the hardware depends on the specific application scenario and design constraints of the technical solution.
  • FIG. 10 and FIG. 11 are schematic structural diagrams of possible communication devices provided by the embodiments of the present application. These communication apparatuses may be used to realize the functions of the terminal device, the network device, or the first device in the foregoing method embodiments, and thus also realize the beneficial effects of the foregoing method embodiments.
  • the communication device may be a terminal device, a network device or a first device, and may also be a module (such as a chip) applied to a terminal device, a network device or a first device.
  • a communication device 1000 includes a processing unit 1010 and a transceiver unit 1020 .
  • the processing unit 1010 is configured to call the transceiver unit 1020 to receive information from other communication devices or send information to other communication devices.
  • the transceiver unit 1020 may further include a receiving unit and a sending unit, the receiving unit is used to receive information from other communication devices, and the sending unit is used to send information to other communication devices.
  • the communication device 1000 is configured to realize the functions of the terminal device or the network device in the method embodiments shown in FIG. 3 , FIG. 4 , FIG. 5 , FIG. 6 , FIG. 7 , FIG. 8 , and FIG. 9 . Wherein Fig. 4 and Fig. 5 embodiment are based on Fig.
  • FIG. 3 Fig. 8 embodiment is based on Fig. 7 embodiment
  • Fig. 9 embodiment is based on Fig. 7 embodiment
  • Fig. 6 embodiment is based on Fig. 3 or Fig. 7 examples.
  • the following uses the embodiment in FIG. 3 and the embodiment in FIG. 7 as an example to illustrate the operations performed by the transceiver unit 1020 and the processing unit 1010 respectively. The operations performed by the two units in other embodiments can be obtained by referring to the method embodiment.
  • the transceiver unit 1020 is used to send the first information to the network device, the first information includes the AI model format supported by the terminal device, and the terminal device The supported AI model format includes a first AI model format; the transceiver unit 1020 is further configured to receive second information from the network device, the second information is used to indicate the first AI represented by the first AI model format Model acquisition method information, the first AI model format is determined according to the first information.
  • the communication device 1000 is used to realize the function of the network device in the method embodiment shown in FIG.
  • the transceiver unit 1020 is used to receive the first information from the terminal device, the first information includes the AI model format supported by the terminal device, and the terminal The AI model formats supported by the device include the first AI model format.
  • the transceiving unit 1020 is further configured to send second information to the terminal device, the second information is used to indicate the acquisition method information of the first AI model represented by the first AI model format, and the first AI model format is determined according to the first information.
  • the transceiver unit 1020 is configured to receive a request message from the terminal device.
  • the processing unit 1010 is used to determine whether the first AI model meets the hardware requirements of the terminal device according to the request message; or, the processing unit 1010 is used to determine the first AI model represented by the first AI model format according to the request message; the transceiver unit 1020 also uses For sending fourth information to the terminal device, the fourth information includes a judgment result of whether the first AI model satisfies the hardware requirements of the terminal device, and/or, the first AI model expressed in the format of the first AI model.
  • the transceiver unit 1020 is used to send the first information to the network device, the first information includes the AI model format supported by the terminal device, and the terminal device
  • the supported AI model formats include the first AI model format
  • the transceiver unit 1020 is further configured to receive the first AI model represented by the first AI model format from the network device, the first AI model format is determined according to the first information.
  • the transceiver unit 1020 is used to receive the first information from the terminal device, the first information includes the AI model format supported by the terminal device;
  • the unit 1020 is further configured to send the first AI model represented by the first AI model format to the terminal device, where the first AI model format is determined according to the first information.
  • the transceiver unit 1020 is used to receive the request message from the network device; the processing unit 1010 is used to judge the first AI according to the request message Whether the model meets the hardware requirements of the terminal device; or, the processing unit 1010 is used to determine the first AI model represented by the first AI model format according to the request message; the transceiver unit 1020 is also used to send fourth information to the terminal device, so The fourth information includes: a judgment result of whether the first AI model satisfies the hardware requirements of the terminal device, and/or, the first AI model represented by the first AI model format.
  • processing unit 1010 and the transceiver unit 1020 can be directly obtained by referring to the relevant descriptions in the method embodiments shown in FIG. 3 and FIG. 7 , and will not be repeated here.
  • an embodiment of the present application further provides a communication device 1100 .
  • the communication device 1100 includes a processor 1110 and an interface circuit 1120 .
  • the processor 1110 and the interface circuit 1120 are coupled to each other.
  • the interface circuit 1120 may be a transceiver or an input-output interface.
  • the communication device 1100 may further include a memory 1130 for storing instructions executed by the processor 1110 or storing input data required by the processor 1110 to execute the instructions or storing data generated by the processor 1110 after executing the instructions.
  • the terminal device chip implements the functions of the terminal device in the above method embodiment.
  • the terminal device chip receives information from other modules in the terminal device (such as radio frequency modules or antennas), and the information is sent to the terminal device by the network device; or, the terminal device chip sends information to other modules in the terminal device (such as radio frequency modules or antenna) to send information, which is sent by the terminal device to the network device.
  • the processor in the embodiments of the present application can be a central processing unit (Central Processing Unit, CPU), and can also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), Field Programmable Gate Array (Field Programmable Gate Array, FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof.
  • a general-purpose processor can be a microprocessor, or any conventional processor.
  • the method steps in the embodiments of the present application may be implemented by means of hardware, or may be implemented by means of a processor executing software instructions.
  • Software instructions can be composed of corresponding software modules, and software modules can be stored in random access memory (Random Access Memory, RAM), flash memory, read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable ROM) , PROM), erasable programmable read-only memory (Erasable PROM, EPROM), electrically erasable programmable read-only memory (Electrically EPROM, EEPROM), register, hard disk, mobile hard disk, CD-ROM or well-known in the art any other form of storage medium.
  • RAM Random Access Memory
  • ROM read-only memory
  • PROM programmable read-only memory
  • PROM erasable programmable read-only memory
  • Erasable PROM Erasable PROM
  • EPROM electrically erasable programmable read-only memory
  • register hard disk, mobile hard disk
  • An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium.
  • the storage medium may also be a component of the processor.
  • the processor and storage medium can be located in the ASIC.
  • the ASIC can be located in a network device or a terminal device.
  • the processor and the storage medium may also exist in the network device or the terminal device as discrete components.
  • the first device may be a server or a core network element.
  • the first device may be a single device, or may be a device composed of functional modules of two network elements.
  • the functions implemented by the first device are jointly completed by the first network element and the second network element.
  • the first network element can be used to realize the function of evaluating and/or compiling, and the second network element is used to store the model.
  • the first network element may be a network element with a network data analysis function (Network Data Analytics Function, NWDAF)
  • NWDAF Network Data Analytics Function
  • the second network element may be a network element with a model storage function
  • the second network element may be a core network element , which can also be an application server.
  • NWDAF Network Data Analytics Function
  • the terminal device when the terminal device acquires the first AI model from the first device, the terminal device may send a request message to the first network element, and the first network element performs an operation corresponding to the request message according to the request message, and the first network element
  • the element may send a model request message to the second network element, the second network element returns the first AI model to the first network element, and the first network element returns the first AI model to the terminal device.
  • the network device when the network device acquires the first AI model from the first device, the network device may send a request message to the first network element, and the first network element performs an operation corresponding to the request message according to the request message, A model request message is sent to the second network element, the second network element returns the first AI model to the first network element, and the first network element returns the first AI model to the network device.
  • the embodiment of the present application also provides an AI model transmission method, the method involves a network device, a terminal device and a first device, the network device may be the network device 101 in Figure 1, and the terminal The device may be the terminal device 102 in FIG. 1 , and the first device may be the first device 103 in FIG. 1 .
  • the first device 103 may include a first network element and a second network element, and for functions of the first network element and the second network element, refer to the above description.
  • the first network element can directly communicate with the network device. There may be a logical connection between the terminal device and the first network element, but the terminal device needs to communicate with the first network element through a third-party device.
  • the third-party device may include, for example, a base station (gNB) , user plane function (UPF) or one or more of application servers; another possible implementation solution is that the third-party equipment may include, for example, a base station (gNB) and/or access and mobility management functions (Access & Mobility Management Function, AMF).
  • gNB base station
  • UPF user plane function
  • AMF access and mobility management functions
  • the specific process of the AI model transmission method is as follows.
  • the network device sends sixth information to the first network element, and correspondingly, the first network element receives the sixth information from the network device.
  • the sixth information includes the AI model format supported by the network device, and the AI model format supported by the network device includes the third AI model format.
  • the sixth information may be used to indicate the first communication scenario.
  • the first network element sends seventh information to the network device, and correspondingly, the network device receives the seventh information from the first network element.
  • the seventh information includes the second AI model, or the seventh information is used to indicate the acquisition method of the second AI model, the second AI model is represented by the third AI model format, and the third AI model format is determined according to the sixth information of.
  • the network device reports the supported AI model format to the first network element, so that the first network element can know the AI model format supported by the network device, so that the first network element will use the AI model format supported by the network device
  • the third AI model format is determined, and the third AI model format must be a model format supported by the network device.
  • the first network element indicates to the network device the acquisition method information (or download information) of the second AI model represented by the third AI model format, or the first network element sends the second AI model to the network device, so that the network device according to The model format of the second AI model downloaded by the acquisition method information must be supported by itself, or the second AI model acquired by the network device must be supported by itself. It ensures that the network device can understand or recognize the format of the AI model, and ensures the feasibility of using the AI model for communication services.
  • the seventh information includes the second AI model
  • the first network element if the first network element stores the second AI model, the first network element returns the second AI model to the network device. Or the first network element may also obtain the second AI model from the second network element.
  • S1203 and S1204 may also be included before S1202.
  • the first network element sends a model request message to the second network element, and the second network element receives the model request message from the first network element.
  • the model request message is used to request the second AI model.
  • the second network element sends the second AI model to the first network element, and correspondingly, the first network element receives the second AI model from the second network element.
  • the seventh information sent by the first network element to the terminal device may include the second AI model.
  • the seventh information when used to indicate the acquisition method information of the second AI model, as shown in FIG. 14 , S1205 and S1206 may also be included after S1202.
  • the acquisition method information of the second AI model may be the address of the second network element, or the uniform resource locator (uniform resource locator, URL) of the AI model of the second network element, or the fully qualified domain name (fully qualified domain name) of the second network element. , FQDN).
  • the network device sends a model request message to the second network element, and the second network element receives the model request message from the network device.
  • the model request message may be used to request the second AI model expressed in the third AI model format, and the second network element determines the second AI model expressed in the third AI model format according to the request information.
  • the model request information may carry an indication of the second AI model, where the indication of the second AI model is used to indicate the second AI model to be acquired by the network device.
  • the second network element may store multiple AI models, and determine the second AI model represented by the third AI model format according to the indication of the second AI model in the model request message.
  • the model request message may also carry a third AI model format, and the second network element determines the second AI model in the third AI model format according to the third AI model format.
  • the second network element sends the second AI model to the network device, and correspondingly, the network device receives the second AI model from the second network element.
  • the network device may also send request information to the first network element, and the first network element performs an operation corresponding to the request information according to the request information.
  • the request information may be included in the sixth information, the request information and the sixth information are included in the same message, or the request information and the sixth information are included in different messages.
  • the request information may be used to request the first network element to judge whether the second AI model satisfies the hardware requirements of the network device. It may be understood that the request message is used to request the first network element to evaluate the second AI model. The first network element judges whether the second AI model meets the hardware requirements of the network device according to the request information. The first network element returns the judgment result to the network device, and the judgment result may be sent separately or included in the seventh information.
  • the request information can also be used to request the first network element to compile the AI model format, and then the first network element compiles the third AI model format according to the request information, so as to obtain the low-level AI model format.
  • the third AI model format is the AI model format of the high-level representation, the model format output by the deep learning framework, or the IR layer model format of the intermediate representation.
  • the compiled AI model format is an AI model format represented by a low-level representation, such as an AI model format represented by an executable file.
  • the function of the first network element is to compile the high-level AI model format into the low-level AI model format.
  • the first network element sends the compiled AI model represented by the low layer to the network device.
  • the compiled low-level AI model can be sent separately or included in the seventh information.
  • the sixth information may also be called AI capability information of the network device.
  • the parameters that may be included in the sixth information are illustrated below with examples.
  • the sixth information may include a third threshold.
  • the third threshold is the maximum time delay of the execution part of the AI model when the network device uses the AI model to execute the communication service.
  • a communication service includes multiple parts, and the network device uses an AI model to execute one or more of the multiple parts in the communication service.
  • the network device can use the AI model to decode the CSI.
  • the third threshold represents the maximum delay when the network device uses the AI model to perform CSI decoding. If the delay of the terminal device to perform CSI decoding using the second AI model exceeds the maximum delay indicated by the third threshold, Then it is considered that applying the second AI model to the communication service cannot meet the delay requirement.
  • the sixth information may further include a fourth threshold, where the fourth threshold is the maximum power consumption of the execution part of the AI model when the network device uses the AI model to execute the communication service.
  • the communication service includes some services executed by the AI model
  • the fourth threshold is the maximum power consumption of the part of the services executed by the AI model.
  • the sixth information also includes indication information of the AI model compilation capability, and the AI model compilation capability includes the capability of compiling and converting the format of the AI model.
  • the sixth information also includes the hardware capability of the network device, and the hardware capability may be one or a combination of the following: hardware model, hardware version, storage capability, computing capability, main frequency, or capability to support heterogeneous computing.
  • the first network element may determine the third AI model format according to the sixth information.
  • the second AI model in the third AI model format needs to meet the hardware requirements of the network device.
  • the network device determines that the second AI model in the third AI model format meets the hardware requirements of the network device, it sends the seventh information to the network device.
  • the first network element device may determine whether the second AI model meets the hardware requirement of the network device, and the hardware requirement is the delay requirement indicated by the third threshold.
  • the network device uses the first AI model to perform communication services, if the delay generated by the calculation of the second AI model does not exceed the third threshold, it means that the second AI model meets the delay requirement indicated by the third threshold; otherwise, the second The AI model does not meet the latency requirement indicated by the third threshold.
  • the network device supports heterogeneous computing capabilities, multiple computing units of heterogeneous computing can correspond to multiple model files of the second AI model, and one computing unit corresponds to a model file of the second AI model, or it can be heterogeneous The calculated multiple calculation units correspond to a model file of the second AI model.
  • Heterogeneous computing in this application may refer to decomposing the computing tasks of the AI model into multiple sub-tasks, and the computing of the multiple sub-tasks is executed on suitable running hardware respectively. These executions should usually be parallel, but serial ones are not excluded. . Therefore, the basis for judging the delay is that all calculations of the multiple subtasks of the AI model have been completed. In the case of heterogeneous computing, it is necessary to judge whether the network device executes the calculation of the AI model on all computing units, whether the generated delay exceeds the third threshold, and if it does not exceed the third threshold, it means the second AI The model meets the delay requirement indicated by the third threshold, otherwise the second AI model does not meet the delay requirement indicated by the third threshold.
  • the network device may determine whether the second AI model meets the hardware requirement of the network device, and the hardware requirement is the power consumption requirement indicated by the fourth threshold.
  • the network device uses the second AI model to perform communication services, if the power consumption generated by the calculation of the second AI model does not exceed the fourth threshold, it means that the second AI model meets the power consumption requirement indicated by the fourth threshold, otherwise the second The AI model does not meet the power consumption requirement indicated by the fourth threshold.
  • the network device supports heterogeneous computing capabilities, it is necessary to determine whether the total power consumption generated by the network device when performing communication services on multiple computing units of heterogeneous computing exceeds the fourth threshold. When the power consumption does not exceed the fourth threshold, it means that the second AI model meets the power consumption requirement indicated by the fourth threshold; otherwise, the second AI model does not meet the power consumption requirement indicated by the fourth threshold.
  • the sixth information includes the storage capability of the network device, it may be determined whether the second AI model meets the hardware requirement of the network device, where the hardware requirement is the storage space requirement indicated by the storage capability.
  • the storage space occupied by the second AI model is not larger than the storage space indicated by the storage capacity, it means that the second AI model meets the storage space requirement indicated by the storage capacity; otherwise, the second AI model does not meet the storage space requirement indicated by the storage capacity .
  • the first network element may determine the computing capability of the network device according to the hardware capability indicated by the sixth information (i.e. Computing power) to further determine whether the computing power of the network device can support the operation of the second AI model, that is, whether the second AI model meets the delay requirement or power consumption requirement based on the computing power of the network device.
  • the sixth information i.e. Computing power
  • the embodiment in FIG. 12 can be combined with the embodiment in FIG. 3 to form a solution that needs to be protected in this application.
  • S1201 and S1202 can also be executed.
  • the execution sequence of S301, S302, S1201 and S1202 is not limited in this application.
  • the terminal device acquires the first AI model
  • the network device acquires the second AI model
  • the terminal device uses the first AI model in the first communication scenario
  • correspondingly, the network device uses the second AI model in the first communication scenario.
  • the first communication scenario is CSI encoding and decoding
  • the terminal device performs CSI encoding using the first AI model, and sends the encoded data to the network device, and the network device restores the received signal from the terminal device using the second AI model.
  • the first AI model may be called an AI encoder
  • the second AI model may be called an AI decoder.
  • the sixth information in S1201 may further include a first identifier, where the first identifier is used to indicate one or more communication scenarios, where the communication scenario may be a scenario in which AI is used for communication.
  • the first identifier may indicate the above-mentioned first communication scenario.
  • it may be CSI codec, or AI-based channel estimation.
  • both the network device and the terminal device need to use the AI model, and they need to be used in pairs.
  • the AI model used in the pair needs to be obtained through joint training.
  • the first AI model and the second AI model are obtained through joint training. of.
  • the scenario of using AI for communication can be some communication scenarios in the wireless communication system.
  • the wireless communication system is composed of some functional modules.
  • the introduction of AI technology into the wireless communication system is to replace the modules in the traditional communication system with modules based on AI technology.
  • an AI-based CSI encoding module is used to replace a traditional CSI encoding module
  • an AI-based CSI decoding module is used to replace a traditional CSI decoding module.
  • the second information sent by the network device to the terminal device in S302 may also include the first identifier.
  • the first identifier may include or indicate a group of AI models, the group of AI models includes one or more AI models, and the group of AI models are all used in the communication scenario indicated by the first identifier (for example: the first communication scenario).
  • the network device obtains the second AI model finally used from the first network element according to the first identifier and its own AI capability.
  • the terminal device acquires the final-used first AI model from the network device or the first device according to the first identifier and its own AI capability.
  • the terminal device and the terminal device start up in the communication system, for example, when the terminal device just accesses the network, the terminal device and the network device communicate based on a traditional scheme. After startup, the terminal device and the network device negotiate a certain functional module to enable the AI technology.
  • the network device and the terminal device negotiate to start the AI communication mode.
  • the network device and the terminal device can negotiate whether to start the AI communication mode based on one or more communication scenarios, where the one or more communication scenarios may use either the original communication mode or the AI communication mode, and decide to use the AI communication mode through negotiation. Communication mode for communication.
  • the terminal device can indicate to the network device the AI model format it supports, and can also indicate to the network device which communication scenarios it supports to use the AI communication mode.
  • the network device sends first instruction information to the terminal device according to the instruction of the terminal device, the first instruction information is used to instruct the one or more communication scenarios to start the AI communication mode, and the first instruction information may include a first identifier, the first The identifier is used to indicate the one or more communication scenarios.
  • the network device sends the second information to the terminal device.
  • the second information is used to indicate the acquisition method information of the first AI model represented by the first AI model format, the first AI model format is determined according to the first information, the first information includes the AI model format supported by the terminal device, and the terminal device supports
  • the AI model formats include a first AI model format.
  • the terminal device may send the first information to the network device, and correspondingly, the network device receives the first information from the terminal device.
  • the second information may further include a first identifier, where the first identifier is used to indicate one or more communication scenarios, and the communication scenario may be a scenario of using AI for communication. It can be understood that, when the second information includes the first identifier, the second information may replace the function of the first indication information in S1501 above, that is, it is used to instruct the one or more communication scenarios to start the AI communication mode.
  • the network device sends sixth information to the first network element.
  • the sixth information includes the AI model format supported by the network device, and the AI model format supported by the network device includes the third AI model format.
  • the sixth information may further include a first identifier, where the first identifier is used to indicate one or more communication scenarios, where the communication scenario may be a scenario in which AI is used for communication.
  • the first network element sends a model request message to the second network element, and the second network element receives the model request message from the first network element.
  • the model request message is used to request the second AI model.
  • the second network element sends the second AI model to the first network element, and correspondingly, the first network element receives the second AI model from the second network element.
  • the first network element sends seventh information to the network device, and correspondingly, the network device receives the seventh information from the first network element.
  • the seventh information includes the second AI model, or the seventh information is used to indicate the acquisition method information of the second AI model, the second AI model is represented by the third AI model format, and the third AI model format is based on the sixth information definite.
  • the terminal device sends information used to indicate the AI model format supported by the terminal device to the first network element.
  • the terminal device may also send the first identifier to the first network element.
  • the first network element sends a model request message to the second network element, and the second network element receives the model request message from the first network element.
  • the model request message is used to request the first AI model.
  • the second network element sends the first AI model to the first network element, and correspondingly, the first network element receives the first AI model from the second network element.
  • the first network element sends the first AI model to the terminal device, and correspondingly, the terminal device receives the first AI model from the first network element.
  • S1503-S1506 is the process for the network device to obtain the second AI model
  • S1507-S1510 is the process for the terminal device to obtain the second AI model. These two sets of steps are not executed sequentially and can be executed in parallel.
  • the terminal device may execute S1507 after receiving the second information.
  • the two groups of steps S1503-S1506 and S1501-S1502 have no sequence and can be executed in parallel.
  • the terminal device After the terminal device receives the first AI model and the network device receives the second AI model, the terminal device and the network device communicate using the AI communication mode.
  • the network device and the terminal device negotiate to start the AI communication mode.
  • This step is the same as S1501, and repeated descriptions will not be repeated.
  • the network device sends the second information to the terminal device.
  • the second information is used to indicate the acquisition method information of the first AI model represented by the first AI model format, the first AI model format is determined according to the first information, the first information includes the AI model format supported by the terminal device, and the terminal device supports
  • the AI model formats include a first AI model format.
  • the terminal device may send the first information to the network device, and correspondingly, the network device receives the first information from the terminal device.
  • the second information may further include a first identifier, where the first identifier is used to indicate one or more communication scenarios, and the communication scenario may be a scenario of using AI for communication.
  • This step is the same as S1502, and repeated descriptions will not be repeated.
  • the network device sends sixth information to the first network element.
  • the sixth information includes the AI model format supported by the network device, and the AI model format supported by the network device includes the third AI model format.
  • the sixth information may further include a first identifier, where the first identifier is used to indicate one or more communication scenarios, where the communication scenario may be a scenario in which AI is used for communication.
  • the first network element sends the acquisition method information of the second AI model to the network device.
  • the second AI model is represented by the third AI model format, and the third AI model format is determined according to the sixth information.
  • the network device sends model request information to the second network element according to the acquisition method information of the second AI model, where the model request information is used to request the second AI model.
  • the second network element sends the second AI model to the network device, and correspondingly, the network device receives the second AI model from the second network element.
  • the terminal device sends information for indicating the AI model format supported by the terminal device to the first network element according to the second information.
  • the second information may include the address of the first network element.
  • the terminal device may also send the first identifier to the first network element.
  • the first network element sends information about the acquisition method of the first AI model to the terminal device, where the information about the acquisition method of the first AI model may include the address of the second network element.
  • the terminal device sends a model request message to the second network element according to the acquisition method information of the first AI model in S1608, and the second network element receives the model request message from the terminal device.
  • the model request message is used to request the first AI model.
  • the second network element sends the first AI model to the terminal device, and correspondingly, the terminal device receives the first AI model from the second network element.
  • S1603-S1606 is the process for the network device to obtain the second AI model
  • S1607-S1610 is the process for the terminal device to obtain the second AI model. These two sets of steps are not executed sequentially and can be executed in parallel.
  • the terminal device may execute S1607 after receiving the second information.
  • the two groups of steps S1603-S1606 and S1601-S1602 have no sequence and can be executed in parallel.
  • the terminal device After the terminal device receives the first AI model and the network device receives the second AI model, the terminal device and the network device communicate using the AI communication mode.
  • S1503-S1506, S1603-S1606 are two ways for the network device to obtain the second AI model, and any one can be selected for implementation.
  • S1507-S1510 and S1607-S1610 are two ways for the terminal device to obtain the second AI model, and any one may be selected for implementation.
  • all or part of them may be implemented by software, hardware, firmware or any combination thereof.
  • software When implemented using software, it may be implemented in whole or in part in the form of a computer program product.
  • the computer program product comprises one or more computer programs or instructions. When the computer program or instructions are loaded and executed on the computer, the processes or functions described in the embodiments of the present application are executed in whole or in part.
  • the computer may be a general purpose computer, a special purpose computer, a computer network, network equipment, user equipment, or other programmable devices.
  • the computer program or instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions may be downloaded from a website, computer, The first device or data center transmits to another website site, computer, first device or data center in a wired or wireless manner.
  • the computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a first device or a data center integrating one or more available media. Described usable medium can be magnetic medium, for example, floppy disk, hard disk, magnetic tape; It can also be optical medium, for example, digital video disc (digital video disc, DVD); It can also be semiconductor medium, for example, solid state drive (solid state drive) , SSD).
  • “at least one” means one or more, and “multiple” means two or more.
  • “And/or” describes the association relationship of associated objects, indicating that there can be three types of relationships, for example, A and/or B, which can mean: A exists alone, A and B exist at the same time, and B exists alone, where A, B can be singular or plural.
  • the character “/” generally indicates that the contextual objects are an “or” relationship; in the formulas of this application, the character “/” indicates that the contextual objects are a “division” Relationship.
  • the steps indicated by dashed lines are optional steps.
  • the first aspect includes the following embodiments 1 to 7.
  • Embodiment 1 A kind of artificial intelligence AI model transmission method is characterized in that, comprises:
  • the terminal device sends first information to the network device, where the first information includes an AI model format supported by the terminal device, and the AI model format supported by the terminal device includes a first AI model format;
  • the terminal device receives the first AI model represented by the first AI model format from the network device, where the first AI model format is determined according to the first information.
  • the network device can know the software capability of the terminal device, that is, know the AI model format supported by the terminal device. In this way, the network device will determine the first AI model format based on the AI model format supported by the terminal device.
  • An AI model format, the first AI model format must be a model format supported by the terminal device.
  • the network device indicates to the terminal device the first AI model represented by the first AI model format, so that the model format of the first AI model must be supported by itself, which ensures the model format of the AI model for wireless communication services between the terminal device and the network device An agreement is reached, and the terminal device can understand or recognize the format of the AI model to ensure the feasibility of using the AI model for communication services.
  • Embodiment 2 The method as described in embodiment 1, is characterized in that, described AI model format comprises following one or more:
  • the model format output by the deep learning framework, the intermediate representation layer model format, or the model format represented by an executable file that can run on the hardware is not limited.
  • Embodiment 3 The method according to Embodiment 1 or 2, wherein the first information further includes a first threshold and/or a second threshold, and the first threshold is that the terminal device uses an AI model to perform The maximum time delay of the execution part of the AI model during the communication service, and the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to perform the communication service.
  • the first threshold is reported by the terminal device, so that the network device can judge whether the first AI model meets the delay requirement according to the first threshold, so that the first AI model can better meet the communication delay requirement, and better apply AI technology in communication services .
  • the terminal device reports the second threshold, so that the network device can judge whether the first AI model meets the power consumption requirement according to the second threshold, so that the first AI model can better adapt to the power consumption requirement of the terminal device.
  • Embodiment 4 The method according to any one of Embodiments 1 to 3, wherein the first information further includes an AI model compilation capability, and the AI model compilation capability includes the capability of compiling and converting an AI model format.
  • the first AI model format determined according to the AI model compilation capability can be the high-level representation of the AI model, or the low-level representation of the AI model.
  • a more suitable AI model format can be selected according to the requirements of network devices. The range of options bigger.
  • Embodiment 5 The method according to any one of Embodiments 1 to 4, wherein the first information further includes one or more of the following: hardware model, hardware version, storage capability, computing capability, or supported Heterogeneous computing capabilities.
  • the first information further includes one or more of the following: hardware model, hardware version, storage capability, computing capability, or supported Heterogeneous computing capabilities.
  • Embodiment 6 The method according to any one of embodiments 1 to 5, wherein the method further comprises:
  • the terminal device receives working mode configuration information from the network device, the working mode configuration information is used to configure a working mode for the terminal device to perform communication services, the working mode is an AI mode, and the AI mode is used for Instructing the terminal device to use the first AI model to perform a communication service.
  • the terminal device can be instructed to use AI technology to perform communication services, and the consistency of the working mode between the terminal device and the network device can be achieved.
  • Embodiment 7 The method according to Embodiment 6, wherein the working mode configuration information further includes the effective time of the first AI model.
  • the second aspect includes the following Embodiment 8 to Embodiment 20.
  • the beneficial effect of the second aspect can be described in the corresponding part of the first aspect.
  • Embodiment 8 An artificial intelligence AI model transmission method is characterized in that, comprising:
  • the network device receives first information from the terminal device, where the first information includes an AI model format supported by the terminal device;
  • the network device sends the first AI model represented by the first AI model format to the terminal device, where the first AI model format is determined according to the first information.
  • Embodiment 9 The method as described in embodiment 8, wherein the AI model format includes one or more of the following:
  • Embodiment 10 The method according to Embodiment 8 or 9, wherein the first information further includes a first threshold and/or a second threshold, and the first threshold is that the terminal device uses an AI model to perform The maximum time delay of the execution part of the AI model during the communication service, and the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to perform the communication service.
  • the first threshold is that the terminal device uses an AI model to perform The maximum time delay of the execution part of the AI model during the communication service
  • the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to perform the communication service.
  • Embodiment 11 The method according to Embodiment 10, wherein the first AI model meets the hardware requirements of the terminal device, and the hardware requirements include: the delay requirement indicated by the first threshold, And/or the first AI model meets the power consumption requirement indicated by the second threshold.
  • Embodiment 12 The method according to any one of Embodiments 8 to 11, wherein the first information further includes one or more of the following: hardware model, hardware version, storage capability, computing capability, or supported Heterogeneous computing capabilities.
  • Embodiment 13 The method according to Embodiment 12, wherein the first AI model meets the hardware requirement of the terminal device, and the hardware requirement includes: the storage space requirement indicated by the storage capability.
  • Embodiment 14 The method according to Embodiment 11, further comprising: the network device sending a request message to the first device; wherein the request message is used to request the first device to determine Whether the first AI model meets the hardware requirements of the terminal device, the request message includes the AI model format supported by the terminal device, the first threshold, and/or the second threshold; or, The request message is used to request the first AI model format.
  • the first device judges whether the first AI model satisfies the hardware requirement of the terminal device, and the judgment result is more accurate.
  • the first device can be compiled and converted to obtain the low-level AI model, so that the first AI model format can more accurately match the capabilities of the terminal without causing leakage of the AI model.
  • Embodiment 15 The method according to Embodiment 13, further comprising: the network device sending a request message to the first device; wherein the request message is used to request the first device to determine Whether the first AI model meets the hardware requirements of the terminal device, the request message includes one or more of the following: AI model format supported by the terminal device, the hardware model, the hardware version, The storage capability, the computing capability, or the capability of supporting heterogeneous computing; or, the request message is used to request the first AI model format.
  • the first device judges whether the first AI model satisfies the hardware requirement of the terminal device, and the judgment result is more accurate.
  • the first device can be compiled and converted to obtain the low-level AI model, so that the first AI model format can more accurately match the capabilities of the terminal without causing leakage of the AI model.
  • Embodiment 16 The method according to any one of Embodiments 8 to 15, wherein the first information further includes an AI model compilation capability, and the AI model compilation capability includes a capability of compiling and converting an AI model format.
  • Embodiment 17 The method according to Embodiment 16, wherein the method further includes: the network device determines, according to the first information, that the first AI mode format is the model format output by the deep learning framework or The middle represents the IR layer model format.
  • the terminal When the terminal has the ability to compile, it indicates the AI model represented by the high-level. Since the AI model represented by the high-level has nothing to do with the hardware of the terminal device, the AI model represented by the high-level is easier to understand. In this case, the terminal device only needs to report the AI The model compilation capability does not need to report other hardware information (such as hardware model or hardware version). On the one hand, it protects the privacy of terminal devices, and on the other hand, it makes the solution more suitable for devices from various manufacturers.
  • Embodiment 18 The method according to any one of embodiments 8 to 15, further comprising:
  • the network device determines that the first AI model format is a model format represented by an executable file when the terminal device does not support the AI model compilation capability. The ability to make the first AI model format match the terminal more accurately.
  • Embodiment 19 The method according to any one of embodiments 8 to 18, further comprising:
  • the network device sends working mode configuration information to the terminal device, the working mode configuration information is used to configure the working mode for the terminal device to perform communication services, the working mode is an AI mode, and the AI mode is used to indicate The terminal device executes a communication service by using the first AI model.
  • Embodiment 20 The method according to Embodiment 19, wherein the working mode configuration information further includes the effective time of the first AI model.
  • the third aspect includes the following embodiment 21 to embodiment 27.
  • Embodiment 21 An artificial intelligence AI model transmission method is characterized in that, comprising:
  • the first device receives a request message from the network device
  • the first device judges whether the first AI model meets the hardware requirements of the terminal device according to the request message, and the model format of the first AI model is the first AI model format; or, the first device judges whether the first AI model meets the hardware requirements of the terminal device; A request message to compile and convert the second AI model format into the first AI model format.
  • the AI model can be stored and maintained by the first device, and the number of the first device can be less than the number of network devices, reducing maintenance costs, reducing overhead and power consumption of network devices, and improving the performance of the communication system.
  • Embodiment 22 The method as described in Embodiment 21, wherein the second AI model format is a model format output by a deep learning framework, or an intermediate representation IR layer model format, and the first AI model format can be The model format that the execution file represents.
  • Embodiment 23 The method according to Embodiment 21 or 22, wherein the request message carries a first threshold, and the first threshold is the AI model execution part when the terminal device uses the AI model to execute communication services the maximum delay.
  • Embodiment 24 The method according to Embodiment 23, wherein the judging whether the first AI model meets the hardware requirements of the terminal device includes:
  • the time delay of the AI model executing part of the AI model executing part when executing the communication service using the first AI model does not exceed a first threshold, determine that the first AI model satisfies the requirement of the terminal The delay requirement of the device; and/or, when the AI model of the AI model execution part executes the communication service when the first AI model is used to execute the communication service, when the delay of the AI model execution part exceeds the first threshold, It is determined that the first AI model does not meet the delay requirement of the terminal device.
  • Embodiment 25 The method according to any one of Embodiments 21 to 24, wherein the request message carries a second threshold, and the second threshold is the AI value when the terminal device uses the AI model to perform communication services.
  • Embodiment 26 The method according to Embodiment 25, wherein the judging whether the first AI model meets the hardware requirements of the terminal device includes:
  • Embodiment 27 The method according to any one of embodiments 21-26, wherein the judging whether the first AI model meets the hardware requirements of the terminal device includes:
  • the storage space occupied by the first AI model does not exceed the storage space indicated by the storage capacity of the terminal device, it is determined that the first AI model meets the storage space requirement of the terminal device; and/or, in the When the storage space occupied by the first AI model exceeds the storage space indicated by the storage capacity of the terminal device, it is determined that the first AI model does not meet the storage space requirement of the terminal device.
  • the fourth aspect includes the following embodiment 28 to embodiment 36.
  • Embodiment 28 An artificial intelligence AI model transmission method is characterized in that, comprising:
  • the terminal device sends first information to the network device, where the first information includes an AI model format supported by the terminal device, and the AI model format supported by the terminal device includes a first AI model format;
  • the terminal device receives the first AI model represented by the first AI model format from the network device, where the first AI model format is determined according to the first information.
  • Embodiment 29 The method according to embodiment 28, wherein the AI model format includes one or more of the following:
  • the model format output by the deep learning framework, the intermediate representation layer model format, or the model format represented by an executable file that can run on the hardware is not limited.
  • Embodiment 30 the method as described in embodiment 28 or 29, is characterized in that, described method also comprises:
  • the terminal device judges whether the first AI model meets hardware requirements of the terminal device.
  • the terminal device judges whether the first AI model meets the hardware requirements of the terminal device, which is closer to the actual running hardware environment, so that the judgment result is more accurate.
  • Embodiment 31 The method according to Embodiment 30, wherein the terminal device determines whether the first AI model meets the hardware requirements of the terminal device, including:
  • the time delay of the execution part of the AI model does not exceed the first threshold when the first AI model is used to execute the communication service, it is determined that the first AI model meets the time delay requirement of the terminal device; and/or, when using determining that the first AI model does not meet the delay requirement of the terminal device when the time delay of the execution part of the AI model exceeds the first threshold when the first AI model executes the communication service;
  • the first threshold is the maximum time delay of the execution part of the AI model when the terminal device uses the AI model to execute the communication service.
  • Embodiment 32 The method according to Embodiment 30 or 31, wherein the terminal device determines whether the first AI model meets the hardware requirements of the terminal device, including:
  • the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to execute the communication service.
  • Embodiment 33 The method according to any one of Embodiments 30-32, wherein the terminal device determines whether the first AI model meets the hardware requirements of the terminal device, including:
  • the storage space occupied by the first AI model does not exceed the storage space indicated by the storage capacity of the terminal device, it is determined that the first AI model meets the storage space requirement of the terminal device; and/or, in the When the storage space occupied by the first AI model exceeds the storage space indicated by the storage capacity of the terminal device, it is determined that the first AI model does not meet the storage space requirement of the terminal device.
  • Embodiment 34 The method according to any one of embodiments 28 to 33, further comprising:
  • the terminal device compiles and converts the first AI model format into a second AI model format.
  • Embodiment 35 The method as described in Embodiment 34, wherein the first AI model format is a model format output by a deep learning framework, or an intermediate representation IR layer model format, and the second AI model format is an optional The model format that the execution file represents.
  • the low-level representation AI model is an AI model related to the running hardware, and using the low-level representation AI model for evaluation can make the evaluation result more accurate.
  • Embodiment 36 The method according to any one of embodiments 28 to 35, further comprising:
  • the terminal device sends a mode request message to the network device, where the mode request message is used to request to use a traditional communication mode to perform communication services.
  • the fifth aspect includes the following Example 37 to Example 48.
  • Example 37 to Example 48 For the beneficial effects of the fifth aspect, reference may also be made to the description of the corresponding embodiment of the second aspect.
  • Embodiment 37 An artificial intelligence AI model transmission method, characterized in that it comprises:
  • the network device receives first information from the terminal device, where the first information includes an AI model format supported by the terminal device;
  • the network device sends the first AI model represented by the first AI model format to the terminal device, where the first AI model format is determined according to the first information.
  • Embodiment 38 The method according to embodiment 37, wherein the AI model format includes one or more of the following:
  • the model format output by the deep learning framework, the intermediate representation layer model format, or the model format represented by an executable file that can run on the hardware is not limited.
  • Embodiment 39 The method according to Embodiment 37 or 38, wherein the first information further includes a first threshold and/or a second threshold, and the first threshold is that the terminal device uses an AI model to perform The maximum time delay of the execution part of the AI model during the communication service, and the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to perform the communication service.
  • Embodiment 40 The method according to Embodiment 39, wherein the first AI model meets the hardware requirements of the terminal device, and the hardware requirements include: the delay requirement indicated by the first threshold, And/or the first AI model meets the power consumption requirement indicated by the second threshold.
  • Embodiment 41 The method according to any one of Embodiments 37 to 40, wherein the first information further includes one or more of the following: hardware model, hardware version, storage capability, computing capability, or supported Heterogeneous computing capabilities.
  • Embodiment 42 The method according to Embodiment 41, wherein the first AI model meets the hardware requirement of the terminal device, and the hardware requirement includes: the storage space requirement indicated by the storage capability.
  • Embodiment 43 The method according to any one of Embodiments 37 to 42, wherein the first information further includes AI model compilation capability, and the AI model compilation capability includes the capability of compiling and converting the AI model format.
  • Embodiment 44 The method according to Embodiment 43, further comprising: the network device determines, according to the first information, that the first AI model format is the model format output by the deep learning framework or The middle represents the IR layer model format.
  • Embodiment 45 The method according to any one of embodiments 37-42, further comprising:
  • the network device determines that the first AI model format is a model format represented by an executable file when the terminal device does not support the AI model compilation capability.
  • Embodiment 46 The method according to any one of embodiments 37-45, further comprising:
  • the network device sends working mode configuration information to the terminal device, the working mode configuration information is used to configure the working mode for the terminal device to perform communication services, the working mode is an AI mode, and the AI mode is used to indicate The terminal device executes a communication service by using the first AI model.
  • Embodiment 47 The method according to Embodiment 46, wherein the working mode configuration information further includes the effective time of the first AI model.
  • Embodiment 48 The method according to any one of embodiments 37-47, further comprising:
  • the network device receives a mode request message from the terminal device, where the mode request message is used to request to use a traditional communication mode to perform communication services;
  • the network device sends a mode confirmation message to the terminal device, where the mode confirmation message is used to indicate confirmation of using the traditional communication mode to perform communication services.
  • the sixth aspect includes the following Example 49 to Example 54.
  • Embodiment 49 provides an artificial intelligence AI model transmission method, which can be executed by a terminal device or by components of the terminal device. This method can be implemented through the following steps:
  • the terminal device sends first information to the network device, where the first information includes an AI model format supported by the terminal device, and the AI model format supported by the terminal device includes a first AI model format; the terminal device receives information from the Working mode configuration information of the network device, the working mode configuration information is used to configure the working mode of the terminal device to perform communication services, the working mode includes AI mode or non-AI mode, and the AI mode is used to indicate the use of the
  • the first AI model executes the communication service, and the non-AI mode is used to indicate that the traditional communication mode is used to execute the communication service.
  • the terminal device can be instructed to use AI technology to perform communication services, and the consistency of the working mode between the terminal device and the network device can be achieved.
  • Embodiment 50 The method as described in embodiment 49, wherein the AI model format includes one or more of the following: a model format output by a deep learning framework, an intermediate representation layer model format, or a model that can run on hardware The format of the model represented by the executable.
  • Embodiment 51 The method according to Embodiment 49 or 50, wherein the first information further includes a first threshold and/or a second threshold, and the first threshold is that the terminal device uses an AI model to perform The maximum time delay of the execution part of the AI model during the communication service, and the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to perform the communication service.
  • the first information further includes a first threshold and/or a second threshold
  • the first threshold is that the terminal device uses an AI model to perform The maximum time delay of the execution part of the AI model during the communication service
  • the second threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to perform the communication service.
  • Embodiment 52 The method according to any one of Embodiments 49-51, wherein the first information further includes AI model compilation capability, and the AI model compilation capability includes the capability of compiling and converting AI model formats.
  • Embodiment 53 The method according to any one of Embodiments 49-52, wherein the first information further includes one or more of the following: hardware model, hardware version, storage capability, computing capability, or supported Heterogeneous computing capabilities.
  • Embodiment 54 The method according to any one of Embodiments 49 to 53, wherein the working mode configuration information is used to configure the terminal device to perform communication services in the AI mode, and the working mode configuration information also includes Effective time of the first AI model.
  • embodiment 55 provides an artificial intelligence AI model transmission method, and the method can be executed by a network device or by a component of the network device. This method can be implemented through the following steps:
  • the network device receives first information from the terminal device, the first information includes an AI model format supported by the terminal device, and the AI model format supported by the terminal device includes a first AI model format; the network device sends the network device Sending work mode configuration information, the work mode configuration information is used to configure the work mode of the terminal device to perform communication services, the work mode includes AI mode or non-AI mode, and the AI mode is used to indicate the use of the first
  • the AI model performs the communication service, and the non-AI mode is used to indicate that the traditional communication mode is used to perform the communication service.
  • the seventh aspect may also include the solutions of the above-mentioned embodiments 50-54.
  • embodiment 56 provides an artificial intelligence AI model transmission method, and the method may be executed by a network device or by a component of the network device. This method can be implemented through the following steps:
  • the network device sends sixth information to the first network element, where the sixth information includes an AI model format supported by the network device, where the AI model format supported by the network device includes a third AI model format, and/or, the The sixth message is used to indicate the first communication scenario;
  • the network device receives seventh information from the first network element, where the seventh information is used to indicate the acquisition method information of the second AI model represented by the third AI model format, or the seventh information includes The second AI model represented by the third AI model format, where the third AI model format is determined according to the sixth information.
  • Embodiment 57 The method of embodiment 56, further comprising:
  • the network device sends a first request message to a second network element according to the acquisition method information, where the first request message is used to request the second AI model;
  • the network device receives the second AI model from the second network element.
  • Embodiment 58 The method according to Embodiment 56 or 57, wherein the sixth information further includes a third threshold and/or a fourth threshold, and the third threshold is that the network device uses an AI model to perform The maximum time delay of the execution part of the AI model during the communication service, the fourth threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to perform the communication service.
  • Embodiment 59 The method according to any one of Embodiments 55 to 58, wherein the sixth information further includes one or more of the following information of the network device: hardware model, hardware version , storage capability, computing capability, ability to support heterogeneous computing, and AI model compilation capability, where the AI model compilation capability includes the ability to compile and convert AI model formats.
  • Embodiment 60 The method according to any one of embodiments 55 to 59, wherein the AI model format includes one or more of the following:
  • the model format output by the deep learning framework the intermediate presentation layer model format, or the model format represented by the executable file.
  • Embodiment 61 The method according to any one of embodiments 55 to 60, further comprising:
  • the network device sends first indication information to the terminal device, where the first indication information is used to instruct one or more communication scenarios to enable the AI communication mode, where the one or more communication scenarios include the first communication scenario.
  • embodiment 62 provides an artificial intelligence AI model transmission method, the method may be executed by a network element, or may be executed by components of the network element, and the network element may be recorded as the first network element.
  • the method may be implemented through the following steps: the first network element receives sixth information from the network device, the sixth information includes the AI model format supported by the network device, and the AI model format supported by the network device includes a third AI A model format, and/or, the sixth message is used to indicate the first communication scenario;
  • the first network element sends seventh information to the network device, where the seventh information is used to indicate the acquisition method information of the second AI model represented by the third AI model format, or the seventh information includes the A second AI model represented by the third AI model format, where the third AI model format is determined according to the sixth information.
  • Embodiment 63 The method of embodiment 62, further comprising:
  • the first network element sends a request message to the second network element, where the request message is used to request the second AI model;
  • the first network element receives the second AI model from the second network element.
  • Embodiment 64 The method according to Embodiment 62 or 63, wherein the sixth information further includes a third threshold and/or a fourth threshold, and the third threshold is that the network device uses an AI model to perform The maximum time delay of the execution part of the AI model during the communication service, the fourth threshold is the maximum power consumption of the execution part of the AI model when the terminal device uses the AI model to perform the communication service.
  • Embodiment 65 The method according to any one of Embodiments 62 to 64, wherein the sixth information further includes one or more of the following information of the network device: hardware model, hardware version , storage capability, computing capability, ability to support heterogeneous computing, and AI model compilation capability, where the AI model compilation capability includes the ability to compile and convert AI model formats.
  • Embodiment 66 The method according to any one of Embodiment 62 to Embodiment 65, wherein the AI model format includes one or more of the following:
  • the model format output by the deep learning framework the intermediate representation layer model format, or the model format represented by the executable file.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

一种人工智能AI模型传输方法及装置，该方法为：终端设备向网络设备发送第一信息，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式；第一信息还包括终端设备的硬件和/或软件要求；终端设备接收来自网络设备的第二信息，第二信息用于指示第一AI模型格式表示的第一AI模型的获取方法信息，第一AI模型格式是根据第一信息确定的。终端设备获取的第一AI模型的模型格式是自身支持的，且满足自身的硬件和/或软件要求，保证了终端设备能够识别该AI模型格式，且能够在自身的硬件和/或软件条件下执行该AI模型，保证使用AI模型进行通信业务的可行性。

Description

一种人工智能AI模型传输方法及装置
相关申请的交叉引用
本申请要求在2021年08月17日提交中国专利局、申请号为202110945160.3、申请名称为“一种AI模型传输方法和装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中；本申请要求在2021年09月26日提交中国专利局、申请号为202111128825.8、申请名称为“一种人工智能AI模型传输方法及装置”的中国专利申请中除所述申请号为202110945160.3的申请文件中的全部内容以外的其他部分内容的优先权，该部分内容通过引用结合在本申请中；本申请要求在2021年12月22日提交中国专利局、申请号为202111576356.6、申请名称为“一种人工智能AI模型传输方法及装置”的中国专利申请中除所述申请号为202110945160.3和202111128825.8的申请文件中的全部内容以外的其他部分内容的优先权，该部分内容通过引用结合在本申请中。
技术领域
本申请实施例涉及通信技术领域，尤其涉及一种人工智能AI模型传输方法及装置。
背景技术
人工智能(artificial intelligence，AI)技术是计算机科学的一个分支，贯穿在计算机发展的历史过程中，是信息科技业界的一个重要的发展方向。随着通信技术的发展，越来越多的应用将通过AI实现智能化。目前，无线通信系统引入AI技术，可能会逐步使用AI模块代替无线通信系统中的功能模块。无线通信系统引入AI技术后，一种可能的工作模式是，网络设备向终端设备发送AI模型，终端设备从网络设备接收AI模型并应用AI模型进行无线通信。
AI模型从设计到硬件可执行的文件，会有不同层次的不同格式的表示方法。不同的终端设备识别AI模型格式的能力也是不同的。因此，网络设备下发的AI模型可能无法被终端设备所识别。这样就会导致无法应用AI模型进行无线通信。
发明内容
本申请实施例提供一种人工智能AI模型传输方法及装置，以期更好地在无线通信系统中应用AI技术。
本申请实施例提供的具体技术方案如下：
第一方面，提供一种人工智能AI模型传输方法，该方法可以由终端设备执行，也可以由终端设备的部件执行。该方法可以通过以下步骤实现：终端设备向网络设备发送第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；所述终端设备接收来自所述网络设备的第二信息，所述第二信息用于指示第一AI模型的获取方法信息；其中，所述第一AI模型是第一AI模型格式表示的，所述第一AI模型格式是根据所述第一信息确定的，和/或，所述第一AI模型应用于第一通信场景。通过终端设备向网络设备上报支持的AI模型格式，能够使得网络设备 获知终端设备支持的AI模型格式，这样，网络设备会根据终端设备支持的AI模型格式确定出第一AI模型格式，第一AI模型格式一定是终端设备支持的模型格式。网络设备向终端设备指示第一AI模型格式表示的第一AI模型的获取方法信息，这样，终端设备根据获取方法信息获取的第一AI模型的模型格式肯定是自身支持的，保证了终端设备和网络设备进行无线通信业务的AI模型的模型格式达成一致，且终端设备能够理解或者能够识别该AI模型格式，保证使用AI模型进行通信业务的可行性。
在一个可能的设计中，所述终端设备根据所述获取方法信息，获取所述第一AI模型。终端设备能够根据获取方法信息向第一设备下载第一AI模型，第一设备可以是服务器或核心网设备，这样，可以通过第一设备存储和维护AI模型，第一设备的数量可以比网络设备的数量少得多，减少维护成本，降低网络设备的开销和功耗，提高通信系统的性能。
在一个可能的设计中，所述终端设备向第一设备发送请求消息；其中，所述请求消息用于请求所述第一设备判断所述第一AI模型是否满足终端设备的硬件要求通过判断所述第一AI模型是否满足终端设备的硬件要求，在满足硬件要求时第一设备向终端设备发送第一AI模型，第一AI模型能够更好的适配终端设备的硬件要求。所述请求消息也可以用于指示所述第一通信场景。
在请求消息用于请求所述第一设备判断所述第一AI模型是否满足终端设备的硬件要求的基础上，在一个可能的设计中，所述请求消息包括所述第一AI模型的指示和/或第一AI模型格式。其中，所述第一AI模型的指示用于指示所要获取的所述第一AI模型。通过所述第一AI模型的指示和/或所述第一AI模型格式来指示第一设备需要下发的模型。第一AI模型的指示例如可以是第一AI模型的标识，第一设备和终端设备预先协商多个AI模型与多个标识的对应关系，这样第一设备可以根据第一AI模型的指示选择需要下发给终端设备的AI模型。
在一个可能的设计中，所述请求消息用于请求所述第一AI模型格式表示的所述第一AI模型。
在请求消息用于请求所述第一AI模型格式表示的所述第一AI模型的基础上，在一个可能的设计中，该请求消息还用于指示所述第一AI模型格式。其中，该第一AI模型格式可以是深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。在一个可能的设计中，所述请求消息包括以下一种或多种：第一阈值、第二阈值、硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力；其中，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗，所述AI模型编译能力包括编译转换AI模型格式的能力。通过终端设备上报第一阈值，这样第一设备能够根据第一阈值判断第一AI模型是否满足时延要求，使得第一AI模型更好的满足通信时延要求，更好在通信业务中应用AI技术。通过终端设备上报第二阈值，这样第一设备能够根据第二阈值判断第一AI模型是否满足功耗要求，使得第一AI模型更好的适配终端设备的功耗要求。
在一个可能的设计中，所述终端设备向所述网络设备发送第三信息，所述第三信息用于指示所述终端设备是否使用所述第一AI模型执行通信业务。通过第三信息，能够指示终端设备使用AI技术执行通信业务，能够达成终端设备与网络设备之间工作模式的一致。
在一个可能的设计中，所述AI模型格式包括以下一项或多项：深度学习框架输出的 模型格式、中间表示层模型格式、或可执行文件表示的模型格式。其中的每一项都包含一种或多种具体的AI模型文件格式。
在一个可能的设计中，所述第一信息还包括第一阈值和/或第二阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。通过终端设备上报第一阈值，这样网络设备能够根据第一阈值判断第一AI模型是否满足时延要求，使得第一AI模型更好的满足通信时延要求，更好在通信业务中应用AI技术。通过终端设备上报第二阈值，这样网络设备能够根据第二阈值判断第一AI模型是否满足功耗要求，使得第一AI模型更好的适配终端设备的功耗要求。
在一个可能的设计中，所述第一信息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力、或AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。这样能够根据第一信息的各种参数能够获得更匹配终端设备硬件能力和软件能力的AI模型，使得在通信业务中更好的应用AI技术。其中，软件能力可以包括终端设备的编译能力。
在一个可能的设计中，所述终端设备接收来自所述网络设备的工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式为AI模式，所述AI模式用于指示所述终端设备使用所述第一AI模型执行通信业务。通过工作模式配置信息，能够指示终端设备使用AI技术执行通信业务，能够达成终端设备与网络设备之间工作模式的一致。
在一个可能的设计中，所述工作模式配置信息还包括所述第一AI模型的生效时间。通过配置第一AI模型的生效时间，网络设备和终端设备能够同步切换到AI模式，达成终端设备与网络设备之间工作模式在时间上的一致。
第二方面，提供一种人工智能AI模型传输方法，该方法可以由网络设备执行，也可以由网络设备的部件执行。该方法可以通过以下步骤实现：网络设备接收来自终端设备的第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；所述网络设备向所述终端设备发送第二信息，所述第二信息用于指示所述第一AI模型格式，及所述第一AI模型格式表示的第一AI模型的获取方法信息，所述第一AI模型格式是根据所述第一信息确定的。
在一个可能的设计中，所述获取方法信息包括所述第一AI模型的下载地址。
在一个可能的设计中，所述网络设备接收来自所述终端设备的第三信息，所述第三信息用于指示所述终端设备是否使用所述第一AI模型执行通信业务。
在一个可能的设计中，所述AI模型格式包括以下一项或多项：深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
在一个可能的设计中，所述第一信息还包括第一阈值和/或第二阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
在一个可能的设计中，所述第一AI模型满足所述终端设备的第一硬件要求，所述第一硬件要求包括：所述第一阈值所指示的时延要求，和/或所述第一AI模型满足所述第二阈值所指示的功耗要求。
在一个可能的设计中，所述第一信息还包括以下一种或多种：硬件型号、硬件版本、 存储能力、计算能力、支持异构计算的能力。
在一个可能的设计中，所述第一AI模型满足所述终端设备的第二硬件要求，所述第二硬件要求包括：所述存储能力所指示的存储空间要求。
在一个可能的设计中，所述第一信息还包括AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
在一个可能的设计中，所述网络设备根据所述第一信息确定所述第一AI模式格式为深度学习框架输出的模型格式或中间表示层模型格式。
在一个可能的设计中，所述网络设备在所述终端设备不支持AI模型编译能力时，确定所述第一AI模式格式为可执行文件表示的模型格式。
在一个可能的设计中，所述网络设备向所述终端设备发送工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式为AI模式，所述AI模式用于指示所述终端设备使用所述第一AI模型执行通信业务。通过工作模式配置信息，能够指示终端设备使用AI技术执行通信业务，能够达成终端设备与网络设备之间工作模式的一致。
在所述网络设备向所述终端设备发送工作模式配置信息的基础上，可选的，所述工作模式配置信息还包括所述第一AI模型的生效时间。这样，网络设备和终端设备能够同步切换到AI模式，达成终端设备与网络设备之间工作模式在时间上的一致。
第二方面及各个可能的设计的有益效果可以参考第一方面相关的描述，在此不予赘述。
第三方面，提供一种人工智能AI模型传输方法，该方法可以由第一设备执行，也可以由第一设备的部件执行。第一设备可以是核心网设备也可以是应用服务器。该方法可以通过以下步骤实现：第一设备接收来自终端设备的请求消息；根据请求消息的作用不同，第一设备可以执行下述两种方案。一种是，第一设备根据所述请求消息，判断第一AI模型是否满足所述终端设备的硬件要求，所述第一设备向所述终端设备发送第四信息，所述第四信息包括第一AI模型是否满足所述终端设备的硬件要求的判断结果。另一种是，第一设备根据所述请求消息，确定第一AI模型格式表示的第一AI模型，可选的，所述第一设备同时判断第一AI模型是否满足所述终端设备的硬件要求；所述第一设备向所述终端设备发送第四信息，如果第一AI模型满足所述终端设备的硬件要求，则所述第四信息包括所述第一AI模型格式表示的第一AI模型；如果第一AI模型不满足所述终端设备的硬件要求，则所述第四信息包括第一AI模型不满足所述终端设备的硬件要求的判断结果。
在请求消息用于请求所述第一设备判断所述第一AI模型是否满足终端设备的硬件要求的基础上，在一个可能的设计中，所述请求消息包括所述第一AI模型的指示和/或第一AI模型格式。其中，所述第一AI模型的指示用于指示所要获取的所述第一AI模型。通过所述第一AI模型的指示和/或所述第一AI模型格式来指示第一设备需要下发的模型。第一AI模型的指示例如可以是第一AI模型的标识，第一设备和终端设备预先协商多个AI模型与多个标识的对应关系，这样第一设备可以根据第一AI模型的指示选择需要下发给终端设备的AI模型。
在一个可能的设计中，所述请求消息用于请求所述第一AI模型格式表示的所述第一AI模型。
在请求消息用于请求所述第一AI模型格式表示的所述第一AI模型的基础上，在一个可能的设计中，该请求消息还用于指示所述第一AI模型格式。其中，该第一AI模型格式 可以是深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
在一个可能的设计中，所述第一AI模型格式为深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
在一个可能的设计中，所述请求消息携带第一阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延。
在一个可能的设计中，所述判断第一AI模型是否满足终端设备的硬件要求，可以通过下述步骤实现：在使用所述第一AI模型执行通信业务时该AI模型执行部分的时延不超过第一阈值时，确定所述第一AI模型满足所述终端设备的时延要求；和/或，在使用所述第一AI模型执行通信业务时该AI模型执行部分的时延超过所述第一阈值时，确定所述第一AI模型不满足所述终端设备的时延要求。
在一个可能的设计中，所述请求消息携带第二阈值，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
在一个可能的设计中，所述判断第一AI模型是否满足终端设备的硬件要求，可以通过下述步骤实现：在使用所述第一AI模型执行通信业务时该AI模型执行部分的功耗不超过第二阈值时，确定所述第一AI模型满足所述终端设备的功耗要求；和/或，在使用所述第一AI模型执行通信业务时该AI模型执行部分的功耗超过所述第二阈值时，确定所述第一AI模型不满足所述终端设备的功耗要求。
在一个可能的设计中，所述请求消息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力或支持异构计算的能力。
在一个可能的设计中，所述判断第一AI模型是否满足终端设备的硬件要求，可以通过下述步骤实现：在所述第一AI模型占用的存储空间不超过所述终端设备的存储能力所指示的存储空间时，确定所述第一AI模型满足所述终端设备的存储空间要求；和/或，在所述第一AI模型占用的存储空间超过所述终端设备的存储能力所指示的存储空间时，确定所述第一AI模型不满足所述终端设备的存储空间要求。
第三方面及各个可能的设计的有益效果可以参考第一方面相关的描述，在此不予赘述。
第四方面，提供一种通信装置，该装置可以是终端设备，也可以是位于终端设备中的部件(例如，芯片，或者芯片系统，或者电路)。该装置具有实现上述第一方面和第一方面的任一种可能的设计中的方法的功能。功能可以通过硬件实现，也可以通过硬件执行相应的软件实现。硬件或软件包括一个或多个与上述功能相对应的模块。一种设计中，该装置可以包括处理单元和收发单元。示例性地：收发单元用于向网络设备发送第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；以及收发单元还用于接收来自所述网络设备的第二信息，所述第二信息用于指示所述第一AI模型格式，及所述第一AI模型格式表示的第一AI模型的获取方法信息，所述第一AI模型格式是根据所述第一信息确定的。上述处理单元和收发单元更详细的描述可以参考上述第一方面中相关描述直接得到。第三方面以及各个可能的设计的有益效果可以参考第一方面对应部分的描述。
第五方面，提供一种通信装置，该装置可以是网络设备，也可以是位于网络设备中的部件(例如，芯片，或者芯片系统，或者电路)。该装置具有实现上述第二方面和第二方面的任一种可能的设计中的方法的功能。功能可以通过硬件实现，也可以通过硬件执行相 应的软件实现。硬件或软件包括一个或多个与上述功能相对应的模块。一种设计中，该装置可以包括处理单元和收发单元。示例性地：收发单元用于接收来自终端设备的第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；以及收发单元还用于向所述终端设备发送第二信息，所述第二信息用于指示所述第一AI模型格式，及所述第一AI模型格式表示的第一AI模型的获取方法信息，所述第一AI模型格式是根据所述第一信息确定的。上述处理单元和收发单元更详细的描述可以参考上述第二方面中相关描述直接得到。第四方面以及各个可能的设计的有益效果可以参考第二方面对应部分的描述。
第六方面，提供一种通信装置，该装置可以是第一设备，也可以是位于第一设备中的部件(例如，芯片，或者芯片系统，或者电路)。该第一设备可以是核心网设备也可以是应用服务器。该装置具有实现上述第三方面和第三方面的任一种可能的设计中的方法的功能。功能可以通过硬件实现，也可以通过硬件执行相应的软件实现。硬件或软件包括一个或多个与上述功能相对应的模块。一种设计中，该装置可以包括处理单元和收发单元。示例性地：收发单元用于接收来自终端设备的请求消息；处理单元用于根据所述请求消息，判断第一AI模型是否满足所述终端设备的硬件要求，和/或，将所述第一AI模型格式表示的所述第一AI模型发送给终端设备。收发单元还用于向所述终端设备发送第四信息，所述第四信息包括第一AI模型是否满足所述终端设备的硬件要求的判断结果和/或所述第一AI模型格式表示的第一AI模型。上述处理单元和收发单元更详细的描述可以参考上述第三方面中相关描述直接得到。第四方面以及各个可能的设计的有益效果可以参考第三方面对应部分的描述。
第七方面，本申请实施例提供一种通信装置，该通信装置包括接口电路和处理器，处理器和接口电路之间相互耦合。处理器通过逻辑电路或执行代码指令用于实现上述第一方面、第一方面各个可能的设计所描述的方法。接口电路用于接收来自所述通信装置之外的其它通信装置的信号并传输至所述处理器或将来自所述处理器的信号发送给所述通信装置之外的其它通信装置。可以理解的是，接口电路可以为收发器或输入输出接口。
可选的，通信装置还可以包括存储器，用于存储处理器执行的指令或存储处理器运行指令所需要的输入数据或存储处理器运行指令后产生的数据。所述存储器可以是物理上独立的单元，也可以与所述处理器耦合，或者所述处理器包括所述存储器。
第八方面，本申请实施例提供一种通信装置，该通信装置包括接口电路和处理器，处理器和接口电路之间相互耦合。处理器通过逻辑电路或执行代码指令用于实现上述第二方面、第二方面各个可能的设计所描述的方法。接口电路用于接收来自所述通信装置之外的其它通信装置的信号并传输至所述处理器或将来自所述处理器的信号发送给所述通信装置之外的其它通信装置。可以理解的是，接口电路可以为收发器或输入输出接口。
可选的，通信装置还可以包括存储器，用于存储处理器执行的指令或存储处理器运行指令所需要的输入数据或存储处理器运行指令后产生的数据。所述存储器可以是物理上独立的单元，也可以与所述处理器耦合，或者所述处理器包括所述存储器。
第九方面，本申请实施例提供一种通信装置，该通信装置包括接口电路和处理器，处理器和接口电路之间相互耦合。处理器通过逻辑电路或执行代码指令用于实现上述第三方面、第三方面各个可能的设计所描述的方法。接口电路用于接收来自所述通信装置之外的其它通信装置的信号并传输至所述处理器或将来自所述处理器的信号发送给所述通信装 置之外的其它通信装置。可以理解的是，接口电路可以为收发器或输入输出接口。
可选的，通信装置还可以包括存储器，用于存储处理器执行的指令或存储处理器运行指令所需要的输入数据或存储处理器运行指令后产生的数据。所述存储器可以是物理上独立的单元，也可以与所述处理器耦合，或者所述处理器包括所述存储器。
第十方面，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程序或可读指令，当所述计算机序或可读指令被通信装置执行时，使得如上述各方面或各方面各个可能的设计中所述的方法被执行。
第十一方面，本申请实施例提供了一种芯片系统，该芯片系统包括处理器，还可以包括存储器。存储器用于存储程序、指令或代码；处理器用于执行存储器存储的程序、指令或代码，以实现上述各方面或各方面各个可能的设计中所述的方法。该芯片系统可以由芯片构成，也可以包含芯片和其他分立器件。
第十二方面，提供了一种包含指令的计算机程序产品，当其被通信装置执行时，使得如第各方面或各方面各个可能的设计中所述的方法被执行。
附图说明
图1为本申请实施例中系统架构示意图；
图2为本申请实施例中网络设备与终端设备利用AI模型进行无线通信的过程示意图；
图3为本申请实施例中AI模型传输方法流程示意图之一；
图4为本申请实施例中终端设备根据获取方法信息获取第一AI模型的流程示意图；
图5为本申请实施例中AI模型传输方法流程示意图之二；
图6为本申请实施例中网络设备向终端设备指示工作模式配置的方法流程示意图；
图7为本申请实施例中AI模型传输方法流程示意图之三；
图8为本申请实施例中服务器协助确定第一AI模型格式的方法流程示意图；
图9为本申请实施例中AI模型传输方法流程示意图之四；
图10为本申请实施例中通信装置结构示意图之一；
图11为本申请实施例中通信装置结构示意图之二；
图12为本申请实施例中AI模型传输方法流程示意图之五；
图13为本申请实施例中AI模型传输方法流程示意图之六；
图14为本申请实施例中AI模型传输方法流程示意图之七；
图15为本申请实施例中AI模型传输方法流程示意图之八；
图16为本申请实施例中AI模型传输方法流程示意图之九。
具体实施方式
下面将结合附图，对本申请实施例进行详细描述。
本申请提供一种人工智能AI模型传输方法及装置，以期更好地在无线通信系统中应用AI技术。其中，方法和装置是基于同一技术构思的，由于方法及装置解决问题的原理相似，因此装置与方法的实施可以相互参见，重复之处不再赘述。
本申请实施例的描述中，“和/或”，描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。字 符“/”一般表示前后关联对象是一种“或”的关系。在本申请的描述中，“第一”、“第二”等词汇，仅用于区分描述的目的，而不能理解为指示或暗示相对重要性，也不能理解为指示或暗示顺序。
下面将结合附图，对本申请实施例进行详细描述。
本申请实施例提供的人工智能AI模型传输方法可以应用于5G通信系统，例如5G新空口(new radio，NR)系统，可以应用于5G通信系统的各种应用场景中，如增强移动宽带(enhanced mobile broadband，eMBB)，超高可靠超低时延通信(ultra reliable low latency communication，URLLC)和增强型机器类型通信(enhanced machine-type communication，eMTC)。本申请实施例提供的人工智能AI模型传输方法也可以应用于未来演进的各种通信系统，例如第六代(6th generation，6G)通信系统，又例如空天海地一体化通信系统。本申请实施例提供的人工智能AI模型传输方法还可以应用于基站和基站之间的通信、终端设备和终端设备的通信、车联网、物联网、工业互联网、卫星通信等的通信，例如，可以应用于设备到设备(Device-to-Device，D2D)、车辆外联(vehicle-to-everything，V2X)、机器到机器(machine-to-machine，M2M)通信系统。
图1示出了本申请实施例适用的一种系统架构，通信系统架构中包括网络设备101和终端设备102。网络设备101为覆盖范围内的终端设备102提供服务。网络设备101为网络设备101覆盖范围内的一个或多个终端设备102提供无线接入。可选的，系统架构还可以包括第一设备103，第一设备103可以是核心网设备或应用服务器。第一设备103也可以称为AI模型服务器，第一设备103可以对AI模型进行编译，将一种AI模型格式编译成另一种AI模型格式。第一设备103还可以对AI模型进行评估，评估AI模型是否能够满足终端设备的目标硬件条件。第一设备103还可以存储或维护AI模型，向终端设备提供AI模型。
网络设备101为无线接入网(radio access network，RAN)中的节点，又可以称为基站，还可以称为RAN节点(或设备)。目前，一些网络设备101的举例为：下一代基站(next generation nodeB，gNB)、下一代演进的基站(next generation evolved nodeB，Ng-eNB)、传输接收点(transmission reception point，TRP)、演进型节点B(evolved Node B，eNB)、无线网络控制器(radio network controller，RNC)、节点B(Node B，NB)、基站控制器(base station controller，BSC)、基站收发台(base transceiver station,BTS)、家庭基站(例如，home evolved NodeB，或home Node B，HNB)、基带单元(base band unit，BBU)，或无线保真(wireless fidelity，Wifi)接入点(access point，AP)，网络设备101还可以是卫星，卫星还可以称为高空平台、高空飞行器、或卫星基站。网络设备101还可以是其他具有网络设备功能的设备，例如，网络设备101还可以是设备到设备(device to device，D2D)通信、车联网或机器到机器(machine to machine，M2M)通信中担任网络设备功能的设备。网络设备101还可以是未来通信系统中任何可能的网络设备。在一些部署中，网络设备101可以包括集中式单元(centralized unit，CU)和(distributed unit，DU)。网络设备还可以包括有源天线单元(active antenna unit，AAU)。CU实现网络设备的部分功能，DU实现网络设备的部分功能，比如，CU负责处理非实时协议和服务，实现无线资源控制(radio resource control，RRC)，分组数据汇聚层协议(packet data convergence protocol，PDCP)层的功能。DU负责处理物理层协议和实时服务，实现无线链路控制(radio link control，RLC)层、媒体接入控制(media access control，MAC)层和物理(physical，PHY)层的 功能。AAU实现部分物理层处理功能、射频处理及有源天线的相关功能。由于RRC层的信息最终会变成PHY层的信息，或者，由PHY层的信息转变而来，因而，在这种架构下，高层信令，如RRC层信令，也可以认为是由DU发送的，或者，由DU+AAU发送的。可以理解的是，网络设备可以为包括CU节点、DU节点、AAU节点中一项或多项的设备。此外，可以将CU划分为接入网(radio access network，RAN)中的网络设备，也可以将CU划分为核心网(core network，CN)中的网络设备，本申请对此不做限定。
终端设备102，又称之为用户设备(user equipment，UE)、移动台(mobile station，MS)、移动终端(mobile terminal，MT)等，是一种向用户提供语音和/或数据连通性的设备。例如，终端设备102包括具有无线连接功能的手持式设备、车载设备等，终端设备102如果位于车辆上(例如放置在车辆内或安装在车辆内)，都可以认为是车载设备，车载设备也称为车载单元(onBoard unit，OBU)。目前，终端设备102可以是：手机(mobile phone)、平板电脑、笔记本电脑、掌上电脑、移动互联网设备(mobile internet device，MID)、可穿戴设备(例如智能手表、智能手环、计步器等)，车载设备(例如，汽车、自行车、电动车、飞机、船舶、火车、高铁等)、虚拟现实(virtual reality，VR)设备、增强现实(augmented reality，AR)设备、工业控制(industrial control)中的无线终端、智能家居设备(例如，冰箱、电视、空调、电表等)、智能机器人、车间设备、无人驾驶(self driving)中的无线终端、远程手术(remote medical surgery)中的无线终端、智能电网(smart grid)中的无线终端、运输安全(transportation safety)中的无线终端、智慧城市(smart city)中的无线终端，或智慧家庭(smart home)中的无线终端、飞行设备(例如，智能机器人、热气球、无人机、飞机)等。终端设备102还可以是其他具有终端设备功能的设备，例如，终端设备102还可以是设备到设备(device to device，D2D)通信、车联网或机器到机器(machine-to-machine，M2M)通信中担任终端设备功能的设备。特别地，在网络设备间进行通信的时候，担任终端设备功能的网络设备也可以看作是终端设备。
作为示例而非限定，在本申请实施例中，终端设备102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或智能穿戴式设备等，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、智能头盔、智能首饰等。
本申请实施例中，用于实现终端设备102的功能的装置例如是芯片、无线收发器、芯片系统，用于实现终端设备102的功能的装置可以被安装或设置或部署在终端设备102中。
为了本领域技术人员更好的理解本申请实施例提供的方案，首先对本申请涉及的几个概念或术语进行解释说明。
1)AI技术
AI是通过机器来模拟人类认识能力的一种科技能力。AI最核心的能力就是根据给定的输入做出判断或预测。AI应用可分为四大部分：感知能力、认知能力、创造力和智能。其中，感知能力例如图像识别、人脸识别、语音识别、自然语言处理、机器翻译、或文本转换等。认知能力例如垃圾邮件识别、信用风险分析、智能天然灾害预测与防治、AI下围 棋或机器学习等。创造力例如AI作曲、AI绘画或AI设计。智能是指通过深刻了解人、事或物的真相，探求真实真理、明辨是非，指导人类过着有意义生活的一种能力，这个领域涉及人类自我意识、自我认知与价值观，是人类最难以模仿的一个领域。
2)AI模型
AI模型多种多样，不同的应用场景可采用不同的AI模型。
AI模型可以基于神经网络模型实现。人工神经网络(artificial neural networks，ANNs)模型，也称为神经网络(NNs)模型或连接模型(connection model)，神经网络模型是AI模型的一种典型代表。神经网络模型是一种模仿人脑神经网络的行为特征，进行分布式并行信息处理的数学计算模型。它的主要任务是借鉴人脑神经网络的原理，根据应用需求建造实用的人工神经网络，实现适用于应用需求的学习算法设计，模拟出人脑的智能活动，然后在技术上解决实际问题。神经网络是依靠网络结构的复杂程度，通过调整内部大量节点之间相互连接的关系，实现相应学习算法的设计的。
3)AI模型格式
AI模型从设计到硬件可执行文件，会有不同层次的不同格式的表示方法，AI模型格式可以理解为用于表示AI模型的格式。AI模型格式的几种举例如下所述。
第一类格式为深度学习框架输出的模型格式。深度学习框架例如MindSpore、Tensorflow、PyTorch、飞桨(PaddlePaddle)。深度学习框架输出的模型格式与AI模型的运行硬件无关。
第二类格式为中间表示(intermediate representation，IR)层模型格式。IR层模型格式与深度学习框架无关，也与AI模型的运行硬件无关。
第一类格式和第二类格式是与AI模型的运行硬件无关的AI模型格式，可以统称为AI模型的高层表示或高层格式。
第三类格式为可执行文件表示的模型格式。可执行文件为可在硬件上运行的文件。AI模型的高层表示经过面向AI模型运行硬件的编译后，可以得到具体运行硬件相关的可执行文件。第三类格式是与AI模型的运行硬件相关的格式，可以称为AI模型的低层表示或低层格式。
高层表示是指AI模型格式与运行硬件无关的格式，除了上述第一类格式和第二类格式，还可以是其它与运行硬件无关的AI模型格式。低层表示可以是与运行硬件直接相关的AI模型格式，除了上述可执行文件表示的模型格式之外，还可以是其它与运行硬件相关的AI模型格式。本申请以第一类格式、第二类格式和第三类格式为例进行介绍。
终端设备的运行硬件可能有多种类型，例如，中央控制单元(central processing unit，CPU)、微型处理器(graphics processing unit，GPU)、嵌入式神经网络处理器(neural-network processing unit，NPU)、或现场可编程门阵列(field programmable gate array，FPGA)等，如果终端设备支持异构计算，运行硬件还可能是上述多种类型的组合。异构计算指的是AI模型分布在多种类型的计算单元上执行，例如，AI模型分布在CPU、GPU和FPGA三种的计算单元上执行。不同类型的运行硬件可以分别对应不同的可执行文件表示的模型格式。例如，终端设备的运行硬件为特定厂商或类别的CPU，可执行文件表示的模型格式为该CPU对应的模型格式；又例如，终端设备的运行硬件为某种GPU和某种FPGA的异构，可执行文件表示的模型格式为该GPU对应的模型格式及该FPGA对应的模型格式。
4)无线通信和AI
针对无线空口特点，考虑将AI与无线空口相结合以提升无线网络性能。应用场景的举例如下所述。例如基于AI的信道估计和信号检测。其中，信号检测可以是从无线信道中把所接收到的含干扰噪声的信号提取出来的过程；信道估计是从接收到的信号中将假定的某个信道模型的模型参数估计出来的过程。又例如，基于AI的端到端的通信链路设计。再例如，基于AI的信道状态信息(channel state information，CSI)反馈方案，即通过神经网络对CSI进行编码并反馈给网络设备。
以基于AI的CSI反馈方案为例，介绍一下网络设备与终端设备利用AI模型进行无线通信的过程。
如图2所示，在图2的(a)中，网络设备向终端设备发送AI模型，在基于AI的CSI反馈方案中，AI模型也可以称为AI编码器。终端设备从网络设备接收AI编码器，或者说终端设备从网络设备下载AI编码器；在另一种可能的实现方案中，终端设备还可能从其它设备获取AI编码器，该其它设备可以是核心网设备或应用服务器。在图2的(b)中，终端设备使用AI编码器用于CSI编码，将编码数据发送给网络设备，网络设备使用AI模型对接收信号进行恢复，其中，网络设备使用的AI模型可以是称为AI解码器。
表示AI模型的AI模型格式有多种，网络设备向终端设备发送的AI模型能够被终端设备识别，终端设备才能够使用该AI模型执行通信业务。这就意味着，终端设备和网络设备需要就AI模型格式达成一致。基于此，如图3所示，本申请实施例提供一种AI模型传输方法，该方法的流程如下所述。
S301.终端设备向网络设备发送第一信息，对应地，网络设备接收来自终端设备的第一信息。
其中，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式。
S302.网络设备向终端设备发送第二信息，对应地，终端设备接收来自网络设备的第二信息。
第二信息用于指示第一AI模型的获取方法信息。其中，第一AI模型是第一AI模型格式表示的，第一AI模型格式是根据第一信息确定的，或者，第一AI模型应用于第一通信场景。
图3实施例，通过终端设备向网络设备上报支持的AI模型格式，能够使得网络设备获知终端设备支持的AI模型格式，这样，网络设备会根据终端设备支持的AI模型格式确定出第一AI模型格式，第一AI模型格式一定是终端设备支持的模型格式。网络设备向终端设备指示第一AI模型格式表示的第一AI模型的获取方法信息(或称，下载信息)，这样，终端设备根据获取方法信息下载的第一AI模型的模型格式肯定是自身支持的，保证了终端设备和网络设备进行无线通信业务的AI模型的模型格式达成一致，且终端设备能够理解或者能够识别该AI模型格式，保证使用AI模型进行通信业务的可行性。通过网络设备向终端设备发送第一AI模型的获取方法信息，终端设备能够根据获取方法信息向第一设备下载第一AI模型，第一设备可以是服务器或核心网设备，这样，可以通过第一设备存储和维护AI模型，第一设备的数量可以比网络设备的数量少，减少维护成本，降低网络设备的开销和功耗，提高通信系统的性能。
下面结合图3实施例，提供一些可选的实现方式。
可选的，在S302之后，还可以包括S303。
S303.终端设备根据获取方法信息，获取第一AI模型。
获取方法信息可以是AI模型的下载地址，例如，可以是第一设备的标识，也可以是下载链接。
基于图3实施例，如图4所示，终端设备根据获取方法信息获取第一AI模型可以通过以下步骤实现。
S401.终端设备向第一设备发送请求消息，对应地，第一设备接收来自终端设备的请求消息。
其中，请求消息用于请求第一设备判断第一AI模型是否满足终端设备的硬件要求，可以理解为请求消息用于请求第一设备对第一AI模型进行评估。
或者，请求消息还可以用于请求第一AI模型格式表示的AI模型。
或者，请求消息用于请求第一设备编译AI模型格式。
或者，请求消息可以用于请求上述几种中的多种，例如，请求消息可以请求第一设备判断第一AI模型是否满足终端设备的硬件要求，以及用于请求第一AI模型格式表示的AI模型。
或者，请求消息可以用于指示第一通信场景。
第一设备可以是核心网设备，终端设备向核心网设备发送请求消息时，可以通过非接入层(NAS)向核心网设备发送请求消息。第一设备可以是应用服务器，终端设备可以通过应用层向应用服务器发送请求消息。
S402.第一设备根据请求消息，执行与请求消息对应的操作。
若请求消息用于请求第一设备判断第一AI模型是否满足终端设备的硬件要求，则第一设备根据请求消息判断第一AI模型是否满足终端设备的硬件要求。
若请求消息用于请求第一AI模型格式表示的AI模型，则第一设备根据请求消息确定第一AI模型格式表示的第一AI模型。请求消息中可以携带第一AI模型的指示，该第一AI模型的指示用于指示终端设备所要获取的第一AI模型。第一设备可能从存储多个AI模型，根据请求消息中的第一AI模型的指示，确定第一AI模型格式表示的第一AI模型。请求消息也可以携带第一AI模型格式，第一设备根据第一AI模型格式，确定第一AI模型格式的第一AI模型。
若请求消息用于请求第一设备编译AI模型格式，则第一设备根据请求消息对第一AI模型格式进行编译，从而获得低层表示的AI模型格式。例如，第一AI模型格式为高层表示的AI模型格式，深度学习框架输出的模型格式、或中间表示IR层模型格式。编译后AI模型格式为低层表示的AI模型格式，例如可执行文件表示的AI模型格式。第一设备的作用为将高层表示的AI模型格式编译成低层表示的AI模型格式。
若请求消息用于指示第一通信场景，则第一设备根据请求消息，确定终端设备用于所述第一通信场景的第一AI模型。
S403.第一设备向终端设备发送第四信息，终端设备接收来自第一设备的第四信息。
若请求消息用于请求第一设备判断第一AI模型是否满足终端设备的硬件要求，且第一设备确定第一AI模型满足终端设备的硬件要求，则第四信息指示第一AI模型满足终端设备的硬件要求。
若请求消息用于请求第一设备判断第一AI模型是否满足终端设备的硬件要求，且第一设备确定第一AI模型不满足终端设备的硬件要求，则第四信息指示第一AI模型不满足 终端设备的硬件要求。
若请求消息用于请求第一AI模型格式表示的AI模型，则第一设备向终端设备发送第一AI模型。第四信息包括第一AI模型，或者第四信息为第一AI模型，或者，第四信息指示第一AI模型。可以理解的是，若请求消息用于请求第一AI模型格式表示的AI模型，则第一设备也可以先判断第一AI模型是否满足终端设备的硬件要求，若满足，则向终端设备发送第一AI模型，若不满足，则不发送第一AI模型，或者也可以向终端设备通知没有满足终端设备的硬件要求的AI模型。
若请求消息用于请求第一设备编译AI模型格式，则第一设备向终端设备发送编译后的低层表示的AI模型。
若请求消息用于指示第一通信场景，第四信息包括终端设备用于所述第一通信场景的第一AI模型。
可选的，如图5所示，基于图3实施例，在S303之后，还可以包括以下步骤。下面所述步骤还可以与图4实施例结合，即在S403之后执行。
S304.终端设备向网络设备发送第三信息，对应地，网络设备接收来自终端设备的第三信息。
第三信息用于指示终端设备是否使用第一AI模型执行通信业务。
当与图4实施例结合时，终端设备可以通过S401～S403获取第一AI模型，并且可以根据从第一设备接收的第四信息确定第一AI模型是否与终端设备的硬件能力匹配。终端设备根据第四信息确定是否使用第一AI模型执行通信业务。
终端设备如果不使用第一AI模型执行通信业务，则终端设备可以使用传统通信模式执行通信业务。第三信息指示终端设备不使用第一AI模型执行通信业务时，也可以理解为，第三信息指示终端设备使用传统通信模式执行通信业务。
S305.网络设备向终端设备返回确认信息，终端设备接收网络设备的确认信息。
该确认信息可以为第三信息的响应信息，当第三信息用于指示终端设备使用第一AI模型执行通信业务，确认信息可以用于指示网络设备使用第一AI模型执行通信业务。当第三信息用于指示终端设备使用传统通信模式执行通信业务，确认信息可以用于指示网络设备使用传统通信模式执行通信业务。
AI模型包括神经网络结构、神经网络结构中各算子所对应的参数(权重、偏置等)、代码、或配置，AI模型可能包含的参数量和计算量都很大。另外通信业务往往对时延是有要求的。网络设备与终端设备在使用AI技术进行无线通信时，终端设备的资源和计算能力需要能够支撑AI模型的运行，包括终端设备的存储空间能够容纳AI模型的存储、技术能力能够支持在要求的时延内完成AI模型的计算、AI模型计算所消耗的能量没有超出终端设备的限定，否则将无法应用AI技术。基于此，第一AI模型可能需要满足终端设备的一些硬件要求。网络设备在确定第一AI模型格式时，也需要考虑第一AI模型格式的AI模型是否满足终端设备的一些硬件要求。
图3实施例中，第一AI模型格式是根据第一信息确定的，以下对网络设备根据第一信息确定第一AI模型格式的可选实现方式进行说明。
首先对终端设备上报的第一信息可能包含的参数进行举例说明，第一信息也可以称为终端设备的AI能力信息。
第一信息还可以包括第一阈值。第一阈值为：终端设备使用AI模型执行通信业务时， 该AI模型执行部分的最大时延。例如，一个通信业务包括多个部分，终端设备使用AI模型执行该通信业务中多个部分中的一个或多个。在终端设备与网络设备进行通信时，终端设备可以使用AI模型进行CSI的编码，网络设备可以使用AI模型进行CSI的解码。这种通信场景中，第一阈值表示终端设备使用AI模型进行CSI编码时的最大时延，如果终端设备使用第一AI模型进行CSI编码的时延超过第一阈值指示的该最大时延，则认为应用第一AI模型进行通信业务不能满足时延要求。
第一信息还可以包括第二阈值，第二阈值为终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。如上所述，通信业务中包括AI模型执行的部分业务，第一阈值为AI模型执行的部分业务的最大功耗。
第一信息还包括AI模型编译能力的指示信息，AI模型编译能力包括编译转换AI模型格式的能力。
第一信息还包括终端设备的硬件能力，硬件能力可以是以下一种或多种组合：硬件型号、硬件版本、存储能力、计算能力、主频或支持异构计算的能力。
根据第一信息携带的参数，网络设备可以根据第一信息确定第一AI模型格式。首先第一AI模型格式的第一AI模型需要满足终端设备的硬件要求。网络设备在判定第一AI模型格式的第一AI模型满足终端设备的硬件要求时，向终端设备发送第二信息。或者，如果网络设备无法判断第一AI模型格式的第一AI模型是否满足终端设备的硬件要求，则向终端设备发送第二信息后，由终端设备向第一设备请求判断第一AI模型是否满足终端设备的硬件要求。
若第一信息包括第一阈值，则网络设备可以判断第一AI模型是否满足终端设备的第一硬件要求，该第一硬件要求即第一阈值所指示的时延要求。当终端设备使用第一AI模型执行通信业务时，如果第一AI模型计算所产生的时延不超过第一阈值，则表示第一AI模型满足第一阈值所指示的时延要求，否则第一AI模型不满足第一阈值所指示的时延要求。如果终端设备支持异构计算的能力，则异构计算的多个计算单元可以对应多个第一AI模型的模型文件，其中一个计算单元对应一个第一AI模型的模型文件，也可以是异构计算的多个计算单元对应一个第一AI模型的模型文件。本申请中异构计算可以是指将AI模型的计算任务分解成多个子任务，多个子任务的计算分别在合适的运行硬件中执行，这些执行通常应该是并行的，但不排除也有串行的。因此，判断时延的依据是，该AI模型的多个子任务的所有计算已完成。在异构计算情况下，需要判断终端设备在所有计算单元上执行该AI模型的计算都已完成时，产生的时延是否超过第一阈值，如果不超过第一阈值时，则表示第一AI模型满足第一阈值所指示的时延要求，否则第一AI模型不满足第一阈值所指示的时延要求。
若第一信息包括第二阈值，则网络设备可以判断第一AI模型是否满足终端设备的第一硬件要求，该第一硬件要求即第二阈值所指示的功耗要求。当终端设备使用第一AI模型执行通信业务时，如果第一AI模型计算所产生的功耗不超过第二阈值，则表示第一AI模型满足第二阈值所指示的功耗要求，否则第一AI模型不满足第二阈值所指示的功耗要求。如果终端设备支持异构计算的能力，则需要判断终端设备在异构计算的多个计算单元上分别执行通信业务时，产生的总功耗是否超过第二阈值，在多个计算单元产生的总功耗不超过第二阈值时，则表示第一AI模型满足第二阈值所指示的功耗要求，否则第一AI模型不满足第二阈值所指示的功耗要求。
若第一信息包括终端设备的存储能力，则可以判断第一AI模型是否满足终端设备的第二硬件要求，第二硬件要求为存储能力所指示的存储空间要求。当第一AI模型占用的存储空间不大于存储能力所指示的存储空间时，表示第一AI模型满足存储能力所指示的存储空间要求，否则第一AI模型不满足存储能力所指示的存储空间要求。
当判断第一AI模型是否满足终端设备的第一硬件要求时，即判断第一AI模型是否满足第一阈值指示的时延要求和第二阈值指示的功耗要求时，网络设备可以根据第一信息指示的硬件能力确定终端设备的计算能力(即算力)，进一步判断终端设备的算力是否能够支持第一AI模型的运行，即在终端设备的算力基础上，使用第一AI模型是否满足时延要求或功耗要求。
在第一AI模型满足终端设备的硬件要求的基础上，如果网络设备确定终端设备具有AI模型编译能力，则网络设备可以确定第一AI模型为高层表示的AI模型，例如深度学习框架输出的模型格式或中间表示层模型格式。如果网络设备确定终端设备不具有AI模型编译能力时，网络设备可以确定第一AI模型为低层表示的AI模型，例如可执行文件表示的模型格式。其中，AI模型编译能力包括编译转换AI模型格式的能力，例如，将高层表示的AI模型编译转换为低层表示的AI模型。当第一信息包括AI模型编译能力的指示信息时，网络设备可以根据第一信息确定终端设备具有AI模型编译能力。如果第一信息不包括AI模型编译能力的指示信息或者第一信息指示终端设备不具有AI模型编译能力，则网络设备可以确定终端设备不具有AI模型编译能力。
网络设备可能不期望AI模型被终端设备获取，即网络设备不期望AI模型暴露。低层表示的AI模型不容易被反编译得出原始的AI模型，低层表示的AI模型比高层表示的AI模型的安全性更高。高层表示的AI模型相比低层表示的AI模型更容易被终端设备识别或获取，即高层表示的AI模型相比低层表示的AI模型更容易被暴露。网络设备可能对AI模型暴露有不同的敏感性，当网络设备对AI模型暴露的敏感度低于设定阈值时或者说网络设备对AI模型暴露不敏感时，网络设备确定第一AI模型为高层表示的AI模型。当网络设备对AI模型暴露的敏感度不低于设定阈值时或者说网络设备对AI模型暴露敏感时，网络设备确定第一AI模型为低层表示的AI模型。
上述判断依据也可以结合使用。例如，在第一AI模型满足终端设备的硬件要求的基础上，如果网络设备对AI模型暴露的敏感度低于设定阈值时或者说网络设备对AI模型暴露不敏感、并且终端设备具有AI模型编译能力，则网络设备可以确定第一AI模型为高层表示的AI模型。
S302中网络设备向终端设备发送第二信息，第二信息用于指示第一AI模型的获取方法信息，可以理解为网络设备向终端设备指示了工作模式，工作模式为AI模式。AI模式用于指示终端设备使用第一AI模型执行通信业务。可选的，网络设备可以向终端设备发送工作模式配置信息，该工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，工作模式为AI模式。该工作模式配置信息可以与第二信息在同一个信令中携带，也可以通过不同的信令携带。
上述描述中，网络设备确定第一AI模型格式是在第一AI模型满足终端设备的硬件要求的基础上判断的，如果第一AI模型不能满足终端设备的硬件要求，则网络设备可以指示终端设备使用传统(legacy)通信模式执行通信业务。若终端设备、网络设备或第一设备中任意两者之间的信息沟通出现错误，则系统按照网络设备指示使用传统(legacy)通 信模式执行通信业务，或者系统默认工作在传统通信模式下。例如网络设备无法识别终端设备上报的第一信息，则网络设备使用传统通信模式执行通信业务，网络设备可以指示终端设备使用传统通信模式执行通信业务。
网络设备向终端设备指示工作模式配置的方法可以通过图6实施例来描述。
S601.终端设备向网络设备发送第一信息，对应地，网络设备接收来自终端设备的第一信息。
其中，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式。
该步骤同S301，可以参照S301相关描述。
S602.网络设备向终端设备发送工作模式配置信息，对应地，终端设备接收来自网络设备的工作模式配置信息。
工作模式配置信息用于配置终端设备执行通信业务的工作模式，工作模式包括AI模式或非AI模式，AI模式用于指示使用第一AI模型执行通信业务，非AI模式用于指示使用传统通信模式执行通信业务。
当工作模式配置信息配置的工作模式为AI模式时，终端设备可以使用第一AI模型执行通信业务；当工作模式配置信息配置的工作模式为非AI模式时，终端设备可以使用传统通信模式执行通信业务。
工作模式为AI模式时，工作模式配置信息还可以包括第一AI模型的生效时间。终端设备可以根据第一AI模型的生效时间，在该生效时间内使用第一AI模型执行通信业务。
当网络设备的能力不能确定第一AI模型格式是否能够与终端设备的能力匹配，或者网络设备不具有第一AI模型格式的AI模型时，终端设备可以通过图4实施例向第一设备请求第一AI模型或请求评估第一AI模型是否匹配自身的能力。这种情况下，终端设备向第一设备发送的请求消息中可以携带上述第一信息携带的参数，例如，请求消息包括以下一种或多种：第一阈值、第二阈值、硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力。第一设备根据请求消息判断第一AI模型是否满足终端设备的硬件要求的方法，可以参考网络设备判断第一AI模型是否满足终端设备的硬件要求的方法，在此不予赘述。例如，请求消息中可以携带第一阈值，第一设备判断第一AI模型是否满足终端设备的硬件要求时，若使用第一AI模型执行通信业务时该AI模型执行部分的时延不超过第一阈值，则第一设备确定第一AI模型满足所述终端设备的时延要求；在使用第一AI模型执行通信业务时该AI模型执行部分的时延超过第一阈值时，第一设备确定第一AI模型不满足终端设备的时延要求。又例如，请求消息携带第二阈值，第一设备在判断第一AI模型是否满足终端设备的硬件要求时，若使用第一AI模型执行通信业务时该AI模型执行部分的功耗不超过第二阈值时，则第一设备确定第一AI模型满足所述终端设备的功耗要求；若使用第一AI模型执行通信业务时该AI模型执行部分的功耗超过第二阈值时，则第一设备确定第一AI模型不满足终端设备的功耗要求。又例如，第一设备判断第一AI模型是否满足终端设备的硬件要求时，若第一AI模型占用的存储空间不超过终端设备的存储能力所指示的存储空间，则确定第一AI模型满足终端设备的存储空间要求；若第一AI模型占用的存储空间超过终端设备的存储能力所指示的存储空间，则确定第一AI模型不满足终端设备的存储空间要求。
当请求消息用于请求第一AI模型格式表示的AI模型，第一AI模型格式为低层表示 的模型格式，则第一设备可以将高层表示的AI模型格式进行编译，转换成低层表示的模型格式。请求消息中可以携带终端设备的硬件信息，例如，终端设备的硬件型号，终端设备的硬件版本，第一设备根据终端设备的硬件信息将高层表示的AI模型格式进行编译，转换成低层表示的模型格式。第一设备还可以判断编译后的低层表示的模型格式是否满足终端设备的硬件要求，在满足要求的情况下，向终端设备发送第四信息，第四信息指示编译后的低层表示的模型格式(即第一AI模型格式)。如果第一设备判断了第一AI模型是否满足存储空间要求，第四信息还可以指示存储空间的余量。在编译后的低层表示的模型格式不满足终端设备的硬件要求的情况下，第一设备可以向终端设备发送第四信息，第四信息指示第一AI模型不满足终端设备的硬件要求，第四信息也可以指示终端设备使用传统通信模式进行通信业务。
基于与图3实施例同一技术构思，如图7所示，本申请实施例还提供另一种AI模型传输方法，该方法的流程如下所述。
S701.终端设备向网络设备发送第一信息，对应地，网络设备接收来自终端设备的第一信息。
其中，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式。
S701的步骤与S301的步骤相同，可以参考S301的描述。
S702.网络设备向终端设备发送第一AI模型格式表示的第一AI模型，对应地，终端设备接收来自网络设备的第一AI模型格式表示的第一AI模型。
其中，第一AI模型格式是根据第一信息确定的。
在S701之后，在S702之前，还可以包括S703。
S703.网络设备根据S701接收的第一信息确定第一AI模型格式。
网络设备根据第一信息确定第一AI模型格式的具体方式可以参考图3实施例中网络设备根据第一信息确定第一AI模型格式的描述。网络设备根据第一信息判断第一AI模型是否满足终端设备的硬件要求，具体方式可以参考图3实施例相关的描述。例如，第一信息包括第一阈值，网络设备根据第一阈值判断第一AI模型是否满足时延要求。第一信息包括第二阈值，网络设备根据第二阈值判断第一AI模型是否满足功耗要求。第一信息包括存储能力，网络设备根据存储能力判断第一AI模型是否符合存储能力指示的存储空间要求。
如果网络设备无法判断第一AI模型格式的第一AI模型是否满足终端设备的硬件要求，可以向第一设备请求判断第一AI模型是否满足终端设备的硬件要求。或者，网络设备也可以向第一设备请求编译转换AI模型格式。总之，网络设备可以在第一设备的协助下确定第一AI模型格式。可选的，如图8所示，需要第一设备协助确定第一AI模型格式的方法可以通过以下步骤实现。
S703-1.网络设备向第一设备发送请求消息，对应地，第一设备接收来自网络设备的请求消息。
其中，请求消息用于请求第一设备判断第一AI模型是否满足终端设备的硬件要求，或者，请求消息用于请求第一AI模型格式。
S703-2.第一设备根据请求消息，执行与请求消息对应的操作。
若请求消息用于请求第一设备判断第一AI模型是否满足终端设备的硬件要求，则第 一设备根据请求消息判断第一AI模型是否满足终端设备的硬件要求。
若请求消息用于请求第一AI模型格式，则第一设备根据请求消息确定第一AI模型格式，例如第一设备根据请求消息将第二AI模型格式编译转换成第一AI模型格式。第二AI模型格式为高层表示的AI模型格式，例如深度学习框架输出的模型格式、或中间表示IR层模型格式。第一AI模型格式为低层表示的AI模型格式，例如可执行文件表示的AI模型格式。第一设备的作用为将高层表示的AI模型格式编译成低层表示的AI模型格式。此时请求消息中可以包含终端设备的硬件信息，例如硬件型号和硬件版本，低层表示的AI模型格式与终端设备的硬件相关，第一设备根据终端设备的硬件信息进行编译，能够获得更准确的编译结果。
S703-3.第一设备向终端设备发送第五信息，终端设备接收来自第一设备的第五信息。
若请求消息用于请求第一设备判断第一AI模型是否满足终端设备的硬件要求，且第一设备确定第一AI模型满足终端设备的硬件要求，则第五信息指示第一AI模型满足终端设备的硬件要求。
若请求消息用于请求第一设备判断第一AI模型是否满足终端设备的硬件要求，且第一设备确定第一AI模型不满足终端设备的硬件要求，则第五信息指示第一AI模型不满足终端设备的硬件要求。
若请求消息用于请求第一AI模型格式，则第一设备向终端设备发送第一AI模型格式的第一AI模型，第五信息包括第一AI模型，或者第五信息为第一AI模型，或者，第五信息指示第一AI模型。
第一设备判断第一AI模型是否满足终端设备的硬件要求的方法，可以参考图3实施例中第一设备的判断方法，具体细节在此不予赘述。如果网络设备向第一设备发送的请求消息用于请求第一设备判断第一AI模型是否满足时延要求，则请求消息中可以携带第一阈值，第一设备根据第一阈值判断第一AI模型是否满足时延要求。如果网络设备向第一设备发送的请求消息用于请求第一设备判断第一AI模型是否满足功耗要求，则请求消息中可以携带第二阈值，第一设备根据第二阈值判断第一AI模型是否满足功耗要求。如果网络设备向第一设备发送的请求消息用于请求第一设备判断第一AI模型是否满足存储空间要求，则请求消息中可以携带存储能力，第一设备根据第一阈值判断第一AI模型是否满足存储空间要求。请求消息还可以包括以下一种或多种：终端设备支持的AI模型格式、硬件型号、硬件版本、存储能力、计算能力、或支持异构计算的能力。第一设备根据请求消息中的终端设备的硬件能力，判断第一AI模型是否满足终端设备硬件要求。第一设备还可以根据请求消息将第二AI模型格式编译转换成符合该硬件的第一AI模型格式。第二AI模型格式为高层表示的AI模型格式，例如深度学习框架输出的模型格式、或中间表示IR层模型格式。第一AI模型格式为低层表示的AI模型格式，例如可执行文件表示的AI模型格式。第一设备的作用为将高层表示的AI模型格式编译成硬件相关的低层表示的AI模型格式。
S702中网络设备向终端设备发送第一AI模型格式表示的第一AI模型，可以理解为网络设备向终端设备指示了工作模式，工作模式为AI模式。可选的，网络设备可以向终端设备发送工作模式配置信息，该工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，工作模式为AI模式。该工作模式配置信息可以与第一AI模型在同一个信令中携带，也可以通过不同的信令携带。
如果网络设备判断第一AI模型不能满足终端设备的硬件要求，或者网络设备通过第一设备判断第一AI模型不能满足终端设备的硬件要求，则网络设备可以指示终端设备使用传统(legacy)通信模式执行通信业务。若终端设备、网络设备或第一设备中任意两者之间的信息沟通出现错误，例如网络设备或第一设备无法识别终端设备上报的第一信息，则网络设备或第一设备可以指示终端设备使用传统(legacy)通信模式执行通信业务。网络设备向终端设备指示工作模式配置的方法可以通过图6实施例来描述。可以参考上述图6实施例的描述，在此不予赘述。
上述图8实施例描述了网络设备通过第一设备协助确定第一AI模型。第一设备在评估第一AI模型是否满足终端设备的硬件要求时，可以将第一AI模型编译转换为低层表示的AI模型，再进行判断，因为低层表示的AI模型能够更好的体现硬件能力，因此判断结果会更加准确。
在一种可能的设计中，S703中网络设备自行根据第一信息确定第一AI模型格式，这种情况下，网络设备如果基于高层表示的AI模型评估是否满足终端设备的硬件要求，则可能评估结果会不准确。如图9所示，S702网络设备向终端设备发送第一AI模型格式表示的第一AI模型，终端设备接收来自网络设备的第一AI模型格式表示的第一AI模型之后，还可以执行以下步骤。
S901.终端设备判断第一AI模型是否与终端设备的算力匹配。若是，则使用第一AI模型进行通信业务，否则执行S902。
判断第一AI模型是否与终端设备的算力匹配，即判断第一AI模型是否满足终端设备的硬件要求。例如，在使用第一AI模型执行通信业务时该AI模型执行部分的时延不超过第一阈值时，确定第一AI模型满足终端设备的时延要求；在使用第一AI模型执行通信业务时该AI模型执行部分的时延超过第一阈值时，确定第一AI模型不满足终端设备的时延要求。第一阈值为终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延。
又例如，在使用第一AI模型执行通信业务时该AI模型执行部分的功耗不超过第二阈值时，确定第一AI模型满足终端设备的功耗要求；在使用第一AI模型执行通信业务时该AI模型执行部分的功耗超过第二阈值时，确定第一AI模型不满足终端设备的功耗要求。第二阈值为终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
再例如，在第一AI模型占用的存储空间不超过终端设备的存储能力所指示的存储空间时，确定第一AI模型满足终端设备的存储空间要求；在第一AI模型占用的存储空间超过终端设备的存储能力所指示的存储空间时，确定第一AI模型不满足终端设备的存储空间要求。
当然，在终端设备支持异构计算能力时，需要综合考虑多个计算单元计算时延要求、功耗要求或存储要求，具体细节可以参照上文中网络设备针对终端设备支持异构计算能力时的判断方法。
可选的，终端设备在判断第一AI模型是否与终端设备的算力匹配时，可以先进行AI模型的编译，将第一AI模型编译转换为低层表示的AI模型，例如，当第一AI模型为深度学习框架输出的模型格式、或中间表示IR层模型格式等高层表示的AI模型时，终端设备将高层表示的AI模型编译转换为低层表示的AI模型，例如可执行文件表示的模型格式。终端设备判断转换后的低层表示的AI模型是否与终端设备的算力匹配，具体方式可以是实际执行第一AI模型，即可获得时延、功耗等信息，也可以参照终端设备判断第一AI模 型是否满足终端设备的硬件要求的方法，在此不予赘述。
S902.终端设备在第一AI模型不满足终端设备的硬件要求时，向网络设备发送模式请求消息，模式请求消息用于请求使用传统通信模式执行通信业务。对应地，网络设备接收来自终端设备的模式请求消息。
可选的，还可以包括S903。
S903.网络设备向终端设备发送模式确认消息，模式确认消息用于指示确认使用传统通信模式执行通信业务。对应地，终端设备接收来自网络设备的模式确认消息。
终端设备判断第一AI模型是否与终端设备的算力匹配，相比网络设备或第一设备的判断结果更加准确。
可以理解的是，为了实现上述实施例中功能，网络设备和终端设备包括了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该很容易意识到，结合本申请中所公开的实施例描述的各示例的单元及方法步骤，本申请能够以硬件或硬件和计算机软件相结合的形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用场景和设计约束条件。
图10和图11为本申请的实施例提供的可能的通信装置的结构示意图。这些通信装置可以用于实现上述方法实施例中终端设备、网络设备或第一设备的功能，因此也能实现上述方法实施例所具备的有益效果。在本申请的实施例中，该通信装置可以是终端设备、网络设备或第一设备，还可以是应用于终端设备、网络设备或第一设备的模块(如芯片)。
如图10所示，通信装置1000包括处理单元1010和收发单元1020。处理单元1010用于调用收发单元1020接收其他通信装置的信息或者向其他通信装置发送信息。收发单元1020还可以进一步包括接收单元和发送单元，接收单元用于接收其他通信装置的信息，发送单元用于向其他通信装置发送信息。通信装置1000用于实现上述图3、图4、图5、图6、图7、图8、图9中所示的方法实施例中终端设备或网络设备的功能。其中图4和图5实施例是基于图3实施例的，图8实施例是基于图7实施例的，图9实施例是基于图7实施例的，图6实施例是基于图3或图7实施例的。以下以图3实施例和图7实施例进行举例，说明收发单元1020和处理单元1010分别执行的操作，其它实施例中两个单元执行的操作可以参考方法实施例得到。
当通信装置1000用于实现图3所示的方法实施例中终端设备的功能时：收发单元1020，用于向网络设备发送第一信息，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式；收发单元1020，还用于接收来自所述网络设备的第二信息，所述第二信息用于指示所述第一AI模型格式表示的第一AI模型的获取方法信息，所述第一AI模型格式是根据所述第一信息确定的。当通信装置1000用于实现图3所示的方法实施例中网络设备的功能时：收发单元1020，用于接收来自终端设备的第一信息，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式。收发单元1020，还用于向终端设备发送第二信息，第二信息用于指示第一AI模型格式表示的第一AI模型的获取方法信息，第一AI模型格式是根据第一信息确定的。当通信装置1000用于实现图3所示的方法实施例中第一设备的功能时：收发单元1020，用于接收来自终端设备的请求消息。处理单元1010用于根据请求消息，判断第一AI模型是否满足终端设备的硬件要求；或者，处理单元1010用于根据请求消息确定第一AI模型格式表示的第一AI模型；收发单元1020还用于向终端设备发送第四信息，第四信息包括第 一AI模型是否满足所述终端设备的硬件要求的判断结果，和/或，第一AI模型格式表示的第一AI模型。
当通信装置1000用于实现图7所示的方法实施例中终端设备的功能时：收发单元1020，用于向网络设备发送第一信息，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式；收发单元1020，还用于接收来自网络设备的第一AI模型格式表示的第一AI模型，第一AI模型格式是根据第一信息确定的。当通信装置1000用于实现图7所示的方法实施例中网络设备的功能时：收发单元1020，用于接收来自终端设备的第一信息，第一信息包括终端设备支持的AI模型格式；收发单元1020，还用于向终端设备发送第一AI模型格式表示的第一AI模型，第一AI模型格式是根据第一信息确定的。当通信装置1000用于实现图3所示的方法实施例中第一设备的功能时：收发单元1020，用于接收来自网络设备的请求消息；处理单元1010用于根据请求消息，判断第一AI模型是否满足终端设备的硬件要求；或者，处理单元1010用于根据请求消息，确定第一AI模型格式表示的第一AI模型；收发单元1020还用于向所述终端设备发送第四信息，所述第四信息包括：第一AI模型是否满足所述终端设备的硬件要求的判断结果，和/或，所述第一AI模型格式表示的第一AI模型。
有关上述处理单元1010和收发单元1020更详细的描述可以直接参考图3、图7所示的方法实施例中相关描述直接得到，这里不加赘述。
基于同一技术构思，如图11所示，本申请实施例还提供一种通信装置1100。通信装置1100包括处理器1110和接口电路1120。处理器1110和接口电路1120之间相互耦合。可以理解的是，接口电路1120可以为收发器或输入输出接口。可选的，通信装置1100还可以包括存储器1130，用于存储处理器1110执行的指令或存储处理器1110运行指令所需要的输入数据或存储处理器1110运行指令后产生的数据。
当通信装置1100用于实现图3、图4、图5、图6、图7、图8、图9所示的方法时，处理器1110用于实现上述处理单元1010的功能，接口电路1120用于实现上述收发单元1020的功能。
当上述通信装置为应用于终端设备的芯片时，该终端设备芯片实现上述方法实施例中终端设备的功能。该终端设备芯片从终端设备中的其它模块(如射频模块或天线)接收信息，该信息是网络设备发送给终端设备的；或者，该终端设备芯片向终端设备中的其它模块(如射频模块或天线)发送信息，该信息是终端设备发送给网络设备的。
可以理解的是，本申请的实施例中的处理器可以是中央处理单元(Central Processing Unit，CPU)，还可以是其它通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field Programmable Gate Array，FPGA)或者其它可编程逻辑器件、晶体管逻辑器件，硬件部件或者其任意组合。通用处理器可以是微处理器，也可以是任何常规的处理器。
本申请的实施例中的方法步骤可以通过硬件的方式来实现，也可以由处理器执行软件指令的方式来实现。软件指令可以由相应的软件模块组成，软件模块可以被存放于随机存取存储器(Random Access Memory，RAM)、闪存、只读存储器(Read-Only Memory，ROM)、可编程只读存储器(Programmable ROM，PROM)、可擦除可编程只读存储器(Erasable PROM，EPROM)、电可擦除可编程只读存储器(Electrically EPROM，EEPROM)、寄存器、硬盘、移动硬盘、CD-ROM或者本领域熟知的任何其它形式的存储介质中。一种示例 性的存储介质耦合至处理器，从而使处理器能够从该存储介质读取信息，且可向该存储介质写入信息。当然，存储介质也可以是处理器的组成部分。处理器和存储介质可以位于ASIC中。另外，该ASIC可以位于网络设备或终端设备中。当然，处理器和存储介质也可以作为分立组件存在于网络设备或终端设备中。
本申请实施例中，第一设备可以是服务器或核心网网元。第一设备可以是单独的一个设备，也可以是由两个网元的功能模块组合成的设备。第一设备实现的功能由第一网元和第二网元共同完成。第一网元可以用于实现评估和/或编译的功能，第二网元用于存储模型。例如，第一网元可以是具有网络数据分析功能(Network Data Analytics Function，NWDAF)的网元，第二网元可以是具有模型存储功能的网元，该第二网元可以是核心网网元，也可以是应用服务器。
基于图3实施例，当终端设备从第一设备获取第一AI模型时，终端设备可以向第一网元发送请求消息，第一网元根据请求消息执行与请求消息对应的操作，第一网元可以向第二网元发送模型请求消息，第二网元向第一网元返回第一AI模型，第一网元向终端设备返回第一AI模型。
类似地，基于图8实施例，当网络设备从第一设备获取第一AI模型时，网络设备可以向第一网元发送请求消息，第一网元根据请求消息执行与请求消息对应的操作，向第二网元发送模型请求消息，第二网元向第一网元返回第一AI模型，第一网元向网络设备返回第一AI模型。
基于同一技术构思，如图12所示，本申请实施例还提供一种AI模型传输方法，该方法涉及网络设备、终端设备和第一设备，网络设备可以是图1中的网络设备101，终端设备可以是图1中的终端设备102，第一设备可以是图1中的第一设备103。在一个可能的设计中，第一设备103可以包括第一网元和第二网元，第一网元和第二网元的功能介绍参考上文的描述。第一网元可以和网络设备直接通信。终端设备可以与第一网元之间存在逻辑连接，但是终端设备需要通过第三方设备与第一网元之间进行通信，一种可能的实现方案是，第三方设备例如可以包括基站(gNB)、用户面功能(user plane function，UPF)或应用服务器中的一种或多种；另一种可能的实现方案是，第三方设备例如可以包括基站(gNB)和/或接入和移动管理功能(Access&Mobility Management Function，AMF)。
该AI模型传输方法的具体过程如下所述。
S1201.网络设备向第一网元发送第六信息，对应地，第一网元接收来自网络设备的第六信息。
其中，第六信息包括网络设备支持的AI模型格式，网络设备支持的AI模型格式包括第三AI模型格式。第六信息可以用于指示第一通信场景。
S1202.第一网元向网络设备发送第七信息，对应地，网络设备接收来自第一网元的第七信息。
其中，第七信息包括第二AI模型，或者第七信息用于指示第二AI模型的获取方法，第二AI模型是第三AI模型格式表示的，第三AI模型格式是根据第六信息确定的。
图12实施例，通过网络设备向第一网元上报支持的AI模型格式，能够使得第一网元获知网络设备支持的AI模型格式，这样，第一网元会根据网络设备支持的AI模型格式确定出第三AI模型格式，第三AI模型格式一定是网络设备支持的模型格式。第一网元向网络设备指示第三AI模型格式表示的第二AI模型的获取方法信息(或称，下载信息)，或 者第一网元向网络设备发送第二AI模型，这样，网络设备根据获取方法信息下载的第二AI模型的模型格式肯定是自身支持的，或者网络设备获取的第二AI模型肯定是自身支持的。保证了网络设备能够理解或者能够识别该AI模型格式，保证使用AI模型进行通信业务的可行性。
下面结合图12实施例，提供一些可选的实现方式。
在一种可能的实现方式中，第七信息包括第二AI模型时，若第一网元存储有第二AI模型，则第一网元向网络设备返回第二AI模型。或者第一网元也可以从第二网元获取第二AI模型。
当第一网元可以从第二网元获取第二AI模型时，如图13所示，在S1202之前还可以包括S1203和S1204。
S1203.第一网元向第二网元发送模型请求消息，第二网元接收来自第一网元的该模型请求消息。
该模型请求消息用于请求第二AI模型。
S1204.第二网元向第一网元发送第二AI模型，对应地，第一网元接收来自第二网元的该第二AI模型。
这种情况下，第一网元向终端设备发送的第七信息中可以包括该第二AI模型。
在一种可能的实现方式中，当第七信息用于指示第二AI模型的获取方法信息时，如图14所示，在S1202之后还可以包括S1205和S1206。第二AI模型的获取方法信息可以是第二网元的地址，或第二网元的AI模型的统一资源定位符(uniform resource locator，URL)或第二网元的全域名(fully qualified domain name，FQDN)。
S1205.网络设备向第二网元发送模型请求消息，第二网元接收来自网络设备的该模型请求消息。
该模型请求消息可以用于请求第三AI模型格式表示的第二AI模型，则第二网元根据请求信息确定第三AI模型格式表示的第二AI模型。该模型请求信息中可以携带第二AI模型的指示，该第二AI模型的指示用于指示网络设备所要获取的第二AI模型。第二网元可能存储多个AI模型，根据该模型请求消息中的第二AI模型的指示，确定第三AI模型格式表示的第二AI模型。该模型请求消息也可以携带第三AI模型格式，第二网元根据第三AI模型格式确定第三AI模型格式的第二AI模型。
S1206.第二网元向网络设备发送第二AI模型，对应地，网络设备接收来自第二网元的该第二AI模型。
图12实施例、图13实施例和图14实施例，通过第二网元存储和维护AI模型，可以使用较少数量的设备存储和维护AI模型，减少维护成本，提高通信系统的性能。
可选的，网络设备还可以向第一网元发送请求信息，第一网元根据该请求信息执行与该请求信息对应的操作。可以理解的是，该请求信息可以包含于该第六信息中，该请求信息与第六信息包含与同一条消息中，或者该请求信息与第六信息包含于不同的消息中。
该请求信息可以用于请求第一网元判断第二AI模型是否满足网络设备的硬件要求，可以理解为请求消息用于请求第一网元对第二AI模型进行评估。第一网元根据该请求信息判断第二AI模型是否满足网络设备的硬件要求。第一网元向网络设备返回判断结果，该判断结果可以单独发送，也可以包含于第七信息中。
该请求信息还可以用于请求第一网元编译AI模型格式，则第一网元根据该请求信息 对第三AI模型格式进行编译，从而获得低层表示的AI模型格式。例如，第三AI模型格式为高层表示的AI模型格式，深度学习框架输出的模型格式、或中间表示IR层模型格式。编译后AI模型格式为低层表示的AI模型格式，例如可执行文件表示的AI模型格式。第一网元的作用为将高层表示的AI模型格式编译成低层表示的AI模型格式。第一网元向网络设备发送编译后的低层表示的AI模型。编译后的低层表示的AI模型可以单独发送，也可以包含于第七信息中。
第六信息也可以称为网络设备的AI能力信息。以下对第六信息可能包含的参数进行举例说明。
第六信息可以包括第三阈值。该第三阈值为网络设备使用AI模型执行通信业务时，该AI模型执行部分的最大时延。例如，一个通信业务包括多个部分，网络设备使用AI模型执行该通信业务中多个部分中的一个或多个。例如，在终端设备与网络设备进行通信时，网络设备可以使用AI模型进行CSI的解码。这种通信场景中，该第三阈值表示网络设备使用AI模型进行CSI解码时的最大时延，如果终端设备使用第二AI模型进行CSI解码的时延超过第三阈值指示的该最大时延，则认为应用第二AI模型进行通信业务不能满足时延要求。
第六信息还可以包括第四阈值，第四阈值为网络设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。如上所述，通信业务中包括AI模型执行的部分业务，第四阈值为AI模型执行的部分业务的最大功耗。
第六信息还包括AI模型编译能力的指示信息，AI模型编译能力包括编译转换AI模型格式的能力。
第六信息还包括网络设备的硬件能力，硬件能力可以是以下一种或多种组合：硬件型号、硬件版本、存储能力、计算能力、主频或支持异构计算的能力。
根据第六信息携带的参数，第一网元可以根据第六信息确定第三AI模型格式。首先第三AI模型格式的第二AI模型需要满足网络设备的硬件要求。网络设备在判定第三AI模型格式的第二AI模型满足网络设备的硬件要求时，向网络设备发送第七信息。
若第六信息包括第三阈值，则第一网元设备可以判断第二AI模型是否满足网络设备的硬件要求，该硬件要求即第三阈值所指示的时延要求。当网络设备使用第一AI模型执行通信业务时，如果第二AI模型计算所产生的时延不超过第三阈值，则表示第二AI模型满足第三阈值所指示的时延要求，否则第二AI模型不满足第三阈值所指示的时延要求。如果网络设备支持异构计算的能力，则异构计算的多个计算单元可以对应多个第二AI模型的模型文件，其中一个计算单元对应一个第二AI模型的模型文件，也可以是异构计算的多个计算单元对应一个第二AI模型的模型文件。本申请中异构计算可以是指将AI模型的计算任务分解成多个子任务，多个子任务的计算分别在合适的运行硬件中执行，这些执行通常应该是并行的，但不排除也有串行的。因此，判断时延的依据是，该AI模型的多个子任务的所有计算已完成。在异构计算情况下，需要判断网络设备在所有计算单元上执行该AI模型的计算都已完成时，产生的时延是否超过第三阈值，如果不超过第三阈值时，则表示第二AI模型满足第三阈值所指示的时延要求，否则第二AI模型不满足第三阈值所指示的时延要求。
若第六信息包括第四阈值，则网络设备可以判断第二AI模型是否满足网络设备的硬件要求，该硬件要求即第四阈值所指示的功耗要求。当网络设备使用第二AI模型执行通 信业务时，如果第二AI模型计算所产生的功耗不超过第四阈值，则表示第二AI模型满足第四阈值所指示的功耗要求，否则第二AI模型不满足第四阈值所指示的功耗要求。如果网络设备支持异构计算的能力，则需要判断网络设备在异构计算的多个计算单元上分别执行通信业务时，产生的总功耗是否超过第四阈值，在多个计算单元产生的总功耗不超过第四阈值时，则表示第二AI模型满足第四阈值所指示的功耗要求，否则第二AI模型不满足第四阈值所指示的功耗要求。
若第六信息包括网络设备的存储能力，则可以判断第二AI模型是否满足网络设备的硬件要求，该硬件要求为存储能力所指示的存储空间要求。当第二AI模型占用的存储空间不大于存储能力所指示的存储空间时，表示第二AI模型满足存储能力所指示的存储空间要求，否则第二AI模型不满足存储能力所指示的存储空间要求。
当判断第二AI模型是否满足第三阈值指示的时延要求和/或第四阈值指示的功耗要求时，第一网元可以根据第六信息指示的硬件能力确定网络设备的计算能力(即算力)，进一步判断网络设备的算力是否能够支持第二AI模型的运行，即在网络设备的算力基础上，使用第二AI模型是否满足时延要求或功耗要求。
图12实施例可以与图3实施例结合形成本申请需要保护的方案，例如，在S301和S302的基础上，还可以执行S1201和S1202。S301、S302、S1201和S1202的执行顺序本申请不做限定。终端设备获取第一AI模型，网络设备获取第二AI模型，终端设备在第一通信场景中使用第一AI模型，对应地，网络设备在该第一通信场景中使用第二AI模型。
例如，第一通信场景为CSI编码和解码，终端设备使用第一AI模型进行CSI编码，将编码数据发送给网络设备，网络设备使用第二AI模型对来自终端设备的接收信号进行恢复。第一AI模型可以称为AI编码器，第二AI模型可以称为AI解码器。
可选的，S1201中的第六信息还可以包括第一标识，该第一标识用于指示一个或多个通信场景，该通信场景可以是使用AI进行通信的场景。例如，该第一标识可以指示上述第一通信场景。例如可以是CSI编解码，也可以是基于AI的信道估计。其中CSI编解码的通信场景中，网络设备和终端设备都需要使用AI模型，并且需要配对使用，配对使用的AI模型需要联合训练获得，例如第一AI模型和第二AI模型是通过联合训练获得的。
使用AI进行通信的场景可以是无线通信系统中的一些通信场景。无线通信系统是由一些功能模块构成的，在无线通信系统中引入AI技术，是使用基于AI技术的模块替换传统通信系统中的模块。例如CSI编解码的通信场景，是使用基于AI的CSI编码模块替换传统的CSI编码模块、使用基于AI的CSI解码模块替换传统的CSI解码模块。
对应于第六信息包括第一标识，S302中网络设备向终端设备发送的第二信息也可以包括第一标识。这样终端设备和网络设备可以都启用第一标识对应的通信场景中对应的AI技术。第一标识可能包括或指示一组AI模型，该组AI模型中包括一个或多个AI模型，该组AI模型都用于该第一标识所指示的通信场景(例如：第一通信场景)。网络设备根据第一标识和自身的AI能力，从第一网元获取最终使用的第二AI模型。终端设备根据第一标识和自身的AI能力，从网络设备或第一设备获取最终使用的第一AI模型。
在一个应用场景中，当通信系统中网络设备和终端设备启动时，例如终端设备刚接入网络时，终端设备和网络设备基于传统的方案进行通信。在启动之后，终端设备和网络设备协商某个功能模块可以启用AI技术。
如图15所示，下面结合具体的应用场景，基于图12实施例与图3实施例，举例说明 AI模型传输方法的具体过程。
S1501.网络设备和终端设备协商启动AI通信模式。
网络设备和终端设备可以基于某一个或多个通信场景协商是否启动AI通信模式，其中，该一个或多个通信场景可能既可以用原始通信模式，也可以使用AI通信模式，通过协商决定使用AI通信模式进行通信。在协商过程中，终端设备可以向网络设备指示自身支持的AI模型格式，也可以向网络设备指示自身支持哪些通信场景使用AI通信模式。网络设备根据终端设备的指示，向终端设备发送第一指示信息，第一指示信息用于指示该一个或多个通信场景启动AI通信模式，第一指示信息中可以包括第一标识，该第一标识用于指示该一个或多个通信场景。
S1502.网络设备向终端设备发送第二信息。
第二信息用于指示第一AI模型格式表示的第一AI模型的获取方法信息，第一AI模型格式是根据第一信息确定的，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式。在S1502之前，终端设备可以向网络设备发送第一信息，对应地，网络设备接收来自终端设备的第一信息。
第二信息还可以包括第一标识，该第一标识用于指示一个或多个通信场景，该通信场景可以是使用AI进行通信的场景。可以理解的是，当第二信息包括第一标识时，第二信息可以代替上述S1501中第一指示信息的功能，即用于指示该一个或多个通信场景启动AI通信模式。
S1503.网络设备向第一网元发送第六信息。
第六信息包括网络设备支持的AI模型格式，网络设备支持的AI模型格式包括第三AI模型格式。
第六信息还可以包括第一标识，该第一标识用于指示一个或多个通信场景，该通信场景可以是使用AI进行通信的场景。
S1504.第一网元向第二网元发送模型请求消息，第二网元接收来自第一网元的该模型请求消息。
该模型请求消息用于请求第二AI模型。
S1505.第二网元向第一网元发送第二AI模型，对应地，第一网元接收来自第二网元的该第二AI模型。
S1506.第一网元向网络设备发送第七信息，对应地，网络设备接收来自第一网元的第七信息。
其中，第七信息包括第二AI模型，或者第七信息用于指示第二AI模型的获取方法信息，第二AI模型是第三AI模型格式表示的，第三AI模型格式是根据第六信息确定的。
S1507.终端设备向第一网元发送用于指示终端设备支持的AI模型格式的信息。
终端设备还可以向第一网元发送第一标识。
S1508.第一网元向第二网元发送模型请求消息，第二网元接收来自第一网元的该模型请求消息。该模型请求消息用于请求第一AI模型。
S1509.第二网元向第一网元发送第一AI模型，对应地，第一网元接收来自第二网元的该第一AI模型。
S1510.第一网元向终端设备发送第一AI模型，对应地，终端设备接收来自第一网元的该第一AI模型。
S1503～S1506是网络设备获取第二AI模型的过程，S1507～S1510是终端设备获取第二AI模型的过程，这两组步骤没有先后执行顺序，可以并行执行。终端设备在收到第二信息后就可以执行S1507。S1503-S1506与S1501-S1502这两组步骤也没有先后顺序，可以并行执行。
终端设备收到第一AI模型，网络设备收到第二AI模型后，终端设备和网络设备使用AI通信模式进行通信。
如图16所示，下面结合具体的应用场景，基于图12实施例与图3实施例，举例说明AI模型传输方法的具体过程。
S1601.网络设备和终端设备协商启动AI通信模式。
本步骤与S1501相同，重复之处不予赘述。
S1602.网络设备向终端设备发送第二信息。
第二信息用于指示第一AI模型格式表示的第一AI模型的获取方法信息，第一AI模型格式是根据第一信息确定的，第一信息包括终端设备支持的AI模型格式，终端设备支持的AI模型格式包括第一AI模型格式。在S1602之前，终端设备可以向网络设备发送第一信息，对应地，网络设备接收来自终端设备的第一信息。
第二信息还可以包括第一标识，该第一标识用于指示一个或多个通信场景，该通信场景可以是使用AI进行通信的场景。
本步骤与S1502相同，重复之处不予赘述。
S1603.网络设备向第一网元发送第六信息。
第六信息包括网络设备支持的AI模型格式，网络设备支持的AI模型格式包括第三AI模型格式。
第六信息还可以包括第一标识，该第一标识用于指示一个或多个通信场景，该通信场景可以是使用AI进行通信的场景。
S1604.第一网元向网络设备发送第二AI模型的获取方法信息。
第二AI模型是第三AI模型格式表示的，第三AI模型格式是根据第六信息确定的。
S1605.网络设备根据第二AI模型的获取方法信息，向第二网元发送模型请求信息，该模型请求信息用于请求第二AI模型。
S1606.第二网元向网络设备发送第二AI模型，对应地，网络设备接收来自第二网元的该第二AI模型。
S1607.终端设备根据第二信息向第一网元发送用于指示终端设备支持的AI模型格式的信息。
第二信息可以包括第一网元的地址。终端设备还可以向第一网元发送第一标识。
S1608.第一网元向终端设备发送第一AI模型的获取方法信息，该第一AI模型的获取方法信息可以包括第二网元的地址。
S1609.终端设备根据S1608中该第一AI模型的获取方法信息，向第二网元发送模型请求消息，第二网元接收来自终端设备的该模型请求消息。该模型请求消息用于请求第一AI模型。
S1610.第二网元向终端设备发送第一AI模型，对应地，终端设备接收来自第二网元的该第一AI模型。
S1603～S1606是网络设备获取第二AI模型的过程，S1607～S1610是终端设备获取第 二AI模型的过程，这两组步骤没有先后执行顺序，可以并行执行。终端设备在收到第二信息后就可以执行S1607。S1603-S1606与S1601-S1602这两组步骤也没有先后顺序，可以并行执行。
终端设备收到第一AI模型，网络设备收到第二AI模型后，终端设备和网络设备使用AI通信模式进行通信。
S1503～S1506、S1603～S1606是网络设备获取第二AI模型的两种方式，可以选择任意一种实现。S1507～S1510、S1607～S1610是终端设备获取第二AI模型的两种方式，可以选择任意一种实现。
在上述实施例中，可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机程序或指令。在计算机上加载和执行所述计算机程序或指令时，全部或部分地执行本申请实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、网络设备、用户设备或者其它可编程装置。所述计算机程序或指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，所述计算机程序或指令可以从一个网站站点、计算机、第一设备或数据中心通过有线或无线方式向另一个网站站点、计算机、第一设备或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是集成一个或多个可用介质的第一设备、数据中心等数据存储设备。所述可用介质可以是磁性介质，例如，软盘、硬盘、磁带；也可以是光介质，例如，数字视频光盘(digital video disc，DVD)；还可以是半导体介质，例如，固态硬盘(solid state drive，SSD)。
在本申请的各个实施例中，如果没有特殊说明以及逻辑冲突，不同的实施例之间的术语和/或描述具有一致性、且可以相互引用，不同的实施例中的技术特征根据其内在的逻辑关系可以组合形成新的实施例。
本申请中，“至少一个”是指一个或者多个，“多个”是指两个或两个以上。“和/或”，描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B的情况，其中A,B可以是单数或者复数。在本申请的文字描述中，字符“/”，一般表示前后关联对象是一种“或”的关系；在本申请的公式中，字符“/”，表示前后关联对象是一种“相除”的关系。本申请提供的各个实施例中，虚线表示的步骤为可选步骤。
可以理解的是，在本申请的实施例中涉及的各种数字编号仅为描述方便进行的区分，并不用来限制本申请的实施例的范围。上述各过程的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定。
结合上述实施例的描述，本申请还提供以下实施例。
第一方面，包括下述实施例1～实施例7。
实施例1.一种人工智能AI模型传输方法，其特征在于，包括：
终端设备向网络设备发送第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；
所述终端设备接收来自所述网络设备的所述第一AI模型格式表示的第一AI模型，所述第一AI模型格式是根据所述第一信息确定的。通过终端设备向网络设备上报支持的AI模型格式，能够使得网络设备获知终端设备的软件能力，即获知终端设备支持的AI模型 格式，这样，网络设备会根据终端设备支持的AI模型格式确定出第一AI模型格式，第一AI模型格式一定是终端设备支持的模型格式。网络设备向终端设备指示第一AI模型格式表示的第一AI模型，这样，第一AI模型的模型格式肯定是自身支持的，保证了终端设备和网络设备进行无线通信业务的AI模型的模型格式达成一致，且终端设备能够理解或者能够识别该AI模型格式，保证使用AI模型进行通信业务的可行性。
实施例2.如实施例1所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
深度学习框架输出的模型格式、中间表示层模型格式、或可在硬件上运行的可执行文件表示的模型格式。
实施例3.如实施例1或2所述的方法，其特征在于，所述第一信息还包括第一阈值和/或第二阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。通过终端设备上报第一阈值，这样网络设备能够根据第一阈值判断第一AI模型是否满足时延要求，使得第一AI模型更好的满足通信时延要求，更好在通信业务中应用AI技术。通过终端设备上报第二阈值，这样网络设备能够根据第二阈值判断第一AI模型是否满足功耗要求，使得第一AI模型更好的适配终端设备的功耗要求。
实施例4.如实施例1～3任一项所述的方法，其特征在于，所述第一信息还包括AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。这样根据AI模型编译能力确定出来的第一AI模型格式可以是AI模型的高层表示，也可以是AI模型的低层表示，可以根据网络设备的需求来选择更合适的AI模型格式，可选择的范围更大。
实施例5.如实施例1～4任一项所述的方法，其特征在于，所述第一信息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力、或支持异构计算的能力。这样能够根据第一信息的各种参数能够获得更匹配终端设备硬件能力的AI模型，使得在通信业务中更好的应用AI技术。
实施例6.如实施例1～5任一项所述的方法，其特征在于，所述方法还包括：
所述终端设备接收来自所述网络设备的工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式为AI模式，所述AI模式用于指示所述终端设备使用所述第一AI模型执行通信业务。通过工作模式配置信息，能够指示终端设备使用AI技术执行通信业务，能够达成终端设备与网络设备之间工作模式的一致。
实施例7.如实施例6所述的方法，其特征在于，所述工作模式配置信息还包括所述第一AI模型的生效时间。通过配置第一AI模型的生效时间，能够切换其它工作模式执行通信业务，更加灵活的配置执行通信业务的工作模式。
第二方面，包括下述实施例8～实施例20。第二方面的有益效果可以第一方面相应地方的描述。
实施例8.一种人工智能AI模型传输方法，其特征在于，包括：
网络设备接收来自终端设备的第一信息，所述第一信息包括所述终端设备支持的AI模型格式；
所述网络设备向所述终端设备发送所述第一AI模型格式表示的第一AI模型，所述第一AI模型格式是根据所述第一信息确定的。
实施例9.如实施例8所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
深度学习框架输出的模型格式、中间表示IR层模型格式、或可在硬件上运行的可执行文件表示的模型格式。
实施例10.如实施例8或9所述的方法，其特征在于，所述第一信息还包括第一阈值和/或第二阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
实施例11.如实施例10所述的方法，其特征在于，所述第一AI模型满足所述终端设备的硬件要求，所述硬件要求包括：所述第一阈值所指示的时延要求，和/或所述第一AI模型满足所述第二阈值所指示的功耗要求。
实施例12.如实施例8～11任一项所述的方法，其特征在于，所述第一信息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力、或支持异构计算的能力。
实施例13.如实施例12所述的方法，其特征在于，所述第一AI模型满足所述终端设备的硬件要求，所述硬件要求包括：所述存储能力所指示的存储空间要求。
实施例14.如实施例11所述的方法，其特征在于，所述方法还包括：所述网络设备向第一设备发送请求消息；其中，所述请求消息用于请求所述第一设备判断所述第一AI模型是否满足所述终端设备的所述硬件要求，所述请求消息包括所述终端设备支持的AI模型格式、所述第一阈值、和/或所述第二阈值；或者，所述请求消息用于请求所述第一AI模型格式。通过向第一设备发送请求消息，使第一设备判断第一AI模型是否满足所述终端设备的所述硬件要求，判断结果更加准确。通过请求第一AI模型格式，能够使得第一设备编译转换，得到低层表示的AI模型，使得第一AI模型格式更精准的匹配终端的能力，并且不会造成AI模型泄露。
实施例15.如实施例13所述的方法，其特征在于，所述方法还包括：所述网络设备向第一设备发送请求消息；其中，所述请求消息用于请求所述第一设备判断所述第一AI模型是否满足所述终端设备的所述硬件要求，所述请求消息包括以下一种或多种：所述终端设备支持的AI模型格式、所述硬件型号、所述硬件版本、所述存储能力、所述计算能力、或所述支持异构计算的能力；或者，所述请求消息用于请求所述第一AI模型格式。通过向第一设备发送请求消息，使第一设备判断第一AI模型是否满足所述终端设备的所述硬件要求，判断结果更加准确。通过请求第一AI模型格式，能够使得第一设备编译转换，得到低层表示的AI模型，使得第一AI模型格式更精准的匹配终端的能力，并且不会造成AI模型泄露。
实施例16.如实施例8～15任一项所述的方法，其特征在于，所述第一信息还包括AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
实施例17.如实施例16所述的方法，其特征在于，所述方法还包括：所述网络设备根据所述第一信息确定所述第一AI模式格式为深度学习框架输出的模型格式或中间表示IR层模型格式。
在终端有编译能力的情况下，指示高层表示的AI模型，由于高层表示的AI模型与终端设备的硬件无关，该高层表示的AI模型更容易理解，这种情况下，终端设备只需要上报AI模型编译能力，不需要上报其它硬件信息(例如硬件型号或硬件版本)，一方面保护 终端设备的隐私，另一方面使得该方案更适配于各种厂商的设备。
实施例18.如实施例8～15任一项所述的方法，其特征在于，所述方法还包括：
所述网络设备在所述终端设备不支持AI模型编译能力时，确定所述第一AI模式格式为可执行文件表示的模型格式。使得第一AI模型格式更精准的匹配终端的能力。
实施例19.如实施例8～18任一项所述的方法，其特征在于，所述方法还包括：
所述网络设备向所述终端设备发送工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式为AI模式，所述AI模式用于指示所述终端设备使用所述第一AI模型执行通信业务。
实施例20.如实施例19所述的方法，其特征在于，所述工作模式配置信息还包括所述第一AI模型的生效时间。
第三方面，包括下述实施例21～实施例27。第三方面的有益效果也可以参考第一方面或第二方面相应实施例的描述。
实施例21.一种人工智能AI模型传输方法，其特征在于，包括：
第一设备接收来自网络设备的请求消息；
所述第一设备根据所述请求消息，判断第一AI模型是否满足终端设备的硬件要求，所述第一AI模型的模型格式为第一AI模型格式；或者，所述第一设备根据所述请求消息，将第二AI模型格式编译转换为所述第一AI模型格式。可以通过第一设备存储和维护AI模型，第一设备的数量可以比网络设备的数量少，减少维护成本，降低网络设备的开销和功耗，提高通信系统的性能。
实施例22.如实施例21所述的方法，其特征在于，所述第二AI模型格式为深度学习框架输出的模型格式、或中间表示IR层模型格式，所述第一AI模型格式为可执行文件表示的模型格式。
实施例23.如实施例21或22所述的方法，其特征在于，所述请求消息携带第一阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延。
实施例24.如实施例23所述的方法，其特征在于，所述判断第一AI模型是否满足终端设备的硬件要求，包括：
在使用所述第一AI模型执行通信业务时该AI模型执行部分的AI模型执行通信业务时该AI模型执行部分的时延不超过第一阈值时，确定所述第一AI模型满足所述终端设备的时延要求；和/或，在使用所述第一AI模型执行通信业务时该AI模型执行部分的AI模型执行通信业务时该AI模型执行部分的时延超过所述第一阈值时，确定所述第一AI模型不满足所述终端设备的时延要求。
实施例25.如实施例21～24任一项所述的方法，其特征在于，所述请求消息携带第二阈值，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的AI模型执行通信业务时该AI模型执行部分的最大功耗。
实施例26.如实施例25所述的方法，其特征在于，所述判断第一AI模型是否满足终端设备的硬件要求，包括：
在使用所述第一AI模型执行通信业务时该AI模型执行部分的功耗不超过第二阈值时，确定所述第一AI模型满足所述终端设备的功耗要求；和/或，在使用所述第一AI模型执行通信业务时该AI模型执行部分的功耗超过所述第二阈值时，确定所述第一AI模型不满足所述终端设备的功耗要求。
实施例27.如实施例21～26任一项所述的方法，其特征在于，所述判断第一AI模型是否满足终端设备的硬件要求，包括：
在所述第一AI模型占用的存储空间不超过所述终端设备的存储能力所指示的存储空间时，确定所述第一AI模型满足所述终端设备的存储空间要求；和/或，在所述第一AI模型占用的存储空间超过所述终端设备的存储能力所指示的存储空间时，确定所述第一AI模型不满足所述终端设备的存储空间要求。
第四方面，包括下述实施例28～实施例36。第四方面的有益效果也可以参考第一方面相应实施例的描述。
实施例28.一种人工智能AI模型传输方法，其特征在于，包括：
终端设备向网络设备发送第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；
所述终端设备接收来自所述网络设备的所述第一AI模型格式表示的第一AI模型，所述第一AI模型格式是根据所述第一信息确定的。
实施例29.如实施例28所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
深度学习框架输出的模型格式、中间表示层模型格式、或可在硬件上运行的可执行文件表示的模型格式。
实施例30.如实施例28或29所述的方法，其特征在于，所述方法还包括：
所述终端设备判断所述第一AI模型是否满足所述终端设备的硬件要求。终端设备判断所述第一AI模型是否满足所述终端设备的硬件要求，更贴近于实际的运行硬件环境，使得判断的结果更加准确。
实施例31.如实施例30所述的方法，其特征在于，所述终端设备判断所述第一AI模型是否满足所述终端设备的硬件要求，包括：
在使用所述第一AI模型执行通信业务时该AI模型执行部分的时延不超过第一阈值时，确定所述第一AI模型满足所述终端设备的时延要求；和/或，在使用所述第一AI模型执行通信业务时该AI模型执行部分的时延超过所述第一阈值时，确定所述第一AI模型不满足所述终端设备的时延要求；
其中，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延。
实施例32.如实施例30或31所述的方法，其特征在于，所述终端设备判断所述第一AI模型是否满足所述终端设备的硬件要求，包括：
在使用所述第一AI模型执行通信业务时该AI模型执行部分的功耗不超过第二阈值时，确定所述第一AI模型满足所述终端设备的功耗要求；和/或，在使用所述第一AI模型执行通信业务时该AI模型执行部分的功耗超过所述第二阈值时，确定所述第一AI模型不满足所述终端设备的功耗要求；
其中，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
实施例33.如实施例30～32任一项所述的方法，其特征在于，所述终端设备判断所述第一AI模型是否满足所述终端设备的硬件要求，包括：
在所述第一AI模型占用的存储空间不超过所述终端设备的存储能力所指示的存储空 间时，确定所述第一AI模型满足所述终端设备的存储空间要求；和/或，在所述第一AI模型占用的存储空间超过所述终端设备的存储能力所指示的存储空间时，确定所述第一AI模型不满足所述终端设备的存储空间要求。
实施例34.如实施例28～33任一项所述的方法，其特征在于，所述方法还包括：
所述终端设备将所述第一AI模型格式编译转换为第二AI模型格式。
实施例35.如实施例34所述的方法，其特征在于，所述第一AI模型格式为深度学习框架输出的模型格式、或中间表示IR层模型格式，所述第二AI模型格式为可执行文件表示的模型格式。通过将高层表示编译转换成低层表示的AI模型，低层表示的AI模型是与运行硬件相关的AI模型，使用低层表示AI模型进行评估，能够使得评估结果更加准确。
实施例36.如实施例28～35任一项所述的方法，其特征在于，所述方法还包括：
所述终端设备在所述第一AI模型不满足所述终端设备的硬件要求时，向所述网络设备发送模式请求消息，所述模式请求消息用于请求使用传统通信模式执行通信业务。
第五方面，包括下述实施例37～实施例48。第五方面的有益效果也可以参考第二方面相应实施例的描述。
实施例37.一种人工智能AI模型传输方法，其特征在于，包括：
网络设备接收来自终端设备的第一信息，所述第一信息包括所述终端设备支持的AI模型格式；
所述网络设备向所述终端设备发送所述第一AI模型格式表示的第一AI模型，所述第一AI模型格式是根据所述第一信息确定的。
实施例38.如实施例37所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
深度学习框架输出的模型格式、中间表示层模型格式、或可在硬件上运行的可执行文件表示的模型格式。
实施例39.如实施例37或38所述的方法，其特征在于，所述第一信息还包括第一阈值和/或第二阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
实施例40.如实施例39所述的方法，其特征在于，所述第一AI模型满足所述终端设备的硬件要求，所述硬件要求包括：所述第一阈值所指示的时延要求，和/或所述第一AI模型满足所述第二阈值所指示的功耗要求。
实施例41.如实施例37～40任一项所述的方法，其特征在于，所述第一信息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力、或支持异构计算的能力。
实施例42.如实施例41所述的方法，其特征在于，所述第一AI模型满足所述终端设备的硬件要求，所述硬件要求包括：所述存储能力所指示的存储空间要求。
实施例43.如实施例37～42任一项所述的方法，其特征在于，所述第一信息还包括AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
实施例44.如实施例43所述的方法，其特征在于，所述方法还包括：所述网络设备根据所述第一信息确定所述第一AI模式格式为深度学习框架输出的模型格式或中间表示IR层模型格式。
实施例45.如实施例37～42任一项所述的方法，其特征在于，所述方法还包括：
所述网络设备在所述终端设备不支持AI模型编译能力时，确定所述第一AI模式格式为可执行文件表示的模型格式。
实施例46.如实施例37～45任一项所述的方法，其特征在于，所述方法还包括：
所述网络设备向所述终端设备发送工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式为AI模式，所述AI模式用于指示所述终端设备使用所述第一AI模型执行通信业务。
实施例47.如实施例46所述的方法，其特征在于，所述工作模式配置信息还包括所述第一AI模型的生效时间。
实施例48.如实施例37～47任一项所述的方法，其特征在于，所述方法还包括：
所述网络设备接收来自所述终端设备的模式请求消息，所述模式请求消息用于请求使用传统通信模式执行通信业务；
所述网络设备向所述终端设备发送模式确认消息，所述模式确认消息用于指示确认使用所述传统通信模式执行通信业务。
第六方面，包括下述实施例49～实施例54。
实施例49.提供一种人工智能AI模型传输方法，该方法可以由终端设备执行，也可以由终端设备的部件执行。该方法可以通过以下步骤实现：
终端设备向网络设备发送第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；所述终端设备接收来自所述网络设备的工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式包括AI模式或非AI模式，所述AI模式用于指示使用所述第一AI模型执行通信业务，所述非AI模式用于指示使用传统通信模式执行通信业务。通过工作模式配置信息，能够指示终端设备使用AI技术执行通信业务，能够达成终端设备与网络设备之间工作模式的一致。
实施例50.如实施例49所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：深度学习框架输出的模型格式、中间表示层模型格式、或可在硬件上运行的可执行文件表示的模型格式。
实施例51.如实施例49或50所述的方法，其特征在于，所述第一信息还包括第一阈值和/或第二阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
实施例52.如实施例49～51任一项所述的方法，其特征在于，所述第一信息还包括AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
实施例53.如实施例49～52任一项所述的方法，其特征在于，所述第一信息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力、或支持异构计算的能力。
实施例54.如实施例49～53任一项所述的方法，其特征在于，工作模式配置信息用于配置所述终端设备执行通信业务的工作模式为AI模式时，工作模式配置信息还包括所述第一AI模型的生效时间。
第七方面，实施例55.提供一种人工智能AI模型传输方法，该方法可以由网络设备执行，也可以由网络设备的部件执行。该方法可以通过以下步骤实现：
网络设备接收来自终端设备的第一信息，所述第一信息包括所述终端设备支持的AI 模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；所述网络设备向网络设备发送工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式包括AI模式或非AI模式，所述AI模式用于指示使用所述第一AI模型执行通信业务，所述非AI模式用于指示使用传统通信模式执行通信业务。
第七方面还可以包括如上述实施例50～54的方案。
第八方面，实施例56.提供一种人工智能AI模型传输方法，该方法可以由网络设备执行，也可以由网络设备的部件执行。该方法可以通过以下步骤实现：
网络设备向第一网元发送第六信息，所述第六信息包括所述网络设备支持的AI模型格式，所述网络设备支持的AI模型格式包括第三AI模型格式，和/或，所述第六消息用于指示第一通信场景；
所述网络设备接收来自所述第一网元的第七信息，所述第七信息用于指示所述第三AI模型格式表示的第二AI模型的获取方法信息，或者所述第七信息包括所述第三AI模型格式表示的第二AI模型，所述第三AI模型格式是根据所述第六信息确定的。
实施例57.如实施例56所述的方法，其特征在于，所述方法还包括：
所述网络设备根据所述获取方法信息，向第二网元发送第一请求消息，所述第一请求消息用于请求所述第二AI模型；
所述网络设备接收来自所述第二网元的所述第二AI模型。
实施例58.如实施例56或57所述的方法，其特征在于，所述第六信息还包括第三阈值和/或第四阈值，所述第三阈值为所述网络设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第四阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
实施例59.如实施例55～实施例58任一项所述的方法，其特征在于，所述第六信息还包括所述网络设备的以下一种或多种的信息：硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力、AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
实施例60.如实施例55～实施例59任一项所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
实施例61.如实施例55～实施例60任一项所述的方法，其特征在于，所述方法还包括：
所述网络设备向终端设备发送第一指示信息，所述第一指示信息用于指示一个或多个通信场景启用AI通信模式，所述一个或多个通信场景包括所述第一通信场景。
第八方面，实施例62.提供一种人工智能AI模型传输方法，该方法可以由网元执行，也可以由网元的部件执行，该网元可以记为第一网元。该方法可以通过以下步骤实现：第一网元接收来自网络设备的第六信息，所述第六信息包括所述网络设备支持的AI模型格式，所述网络设备支持的AI模型格式包括第三AI模型格式，和/或，所述第六消息用于指示第一通信场景；
所述第一网元向所述网络设备发送第七信息，所述第七信息用于指示所述第三AI模型格式表示的第二AI模型的获取方法信息，或者所述第七信息包括所述第三AI模型格式表示的第二AI模型，所述第三AI模型格式是根据所述第六信息确定的。
实施例63.如实施例62所述的方法，其特征在于，所述方法还包括：
所述第一网元向第二网元发送请求消息，所述请求消息用于请求所述第二AI模型；
所述第一网元接收来自所述第二网元的所述第二AI模型。
实施例64.如实施例62或63所述的方法，其特征在于，所述第六信息还包括第三阈值和/或第四阈值，所述第三阈值为所述网络设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第四阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
实施例65.如实施例62～实施例64任一项所述的方法，其特征在于，所述第六信息还包括所述网络设备的以下一种或多种的信息：硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力、AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
实施例66.如实施例62～实施例65任一项所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。

Claims (57)

  1. 一种人工智能AI模型传输方法，其特征在于，包括：
    终端设备向网络设备发送第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；
    所述终端设备接收来自所述网络设备的第二信息，所述第二信息用于指示第一AI模型的获取方法信息；其中，所述第一AI模型是第一AI模型格式表示的，所述第一AI模型格式是根据所述第一信息确定的，和/或，所述第一AI模型应用于第一通信场景。
  2. 如权利要求1所述的方法，其特征在于，所述方法还包括：
    所述终端设备根据所述获取方法信息，获取所述第一AI模型。
  3. 如权利要求2所述的方法，其特征在于，所述终端设备根据所述获取方法信息，获取所述第一AI模型，包括：
    所述终端设备向第一设备发送请求消息；其中所述请求消息用于请求所述第一AI模型格式表示的所述第一AI模型，和/或，所述请求消息用于指示所述第一通信场景；
    所述终端设备接收来自第一设备的第四信息，所述第四信息包括所述第一AI模型格式表示的所述第一AI模型。
  4. 如权利要求3所述的方法，其特征在于，所述请求消息还用于请求所述第一设备判断所述第一AI模型是否满足终端设备的硬件要求，所述第四信息还包括所述第一AI模型是否满足所述终端设备的硬件要求的判断结果。
  5. 如权利要求4所述的方法，其特征在于，所述请求消息包括以下一种或多种：第一阈值、第二阈值、硬件型号、硬件版本、存储能力、计算能力或支持异构计算的能力；其中，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
  6. 如权利要求1～5任一项所述的方法，其特征在于，所述方法还包括：
    所述终端设备向所述网络设备发送第三信息，所述第三信息用于指示所述终端设备是否使用所述第一AI模型执行通信业务。
  7. 如权利要求1～6任一项所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
    深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
  8. 如权利要求1～7任一项所述的方法，其特征在于，所述第一信息还包括第一阈值和/或第二阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
  9. 如权利要求1～8任一项所述的方法，其特征在于，所述第一信息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力、AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
  10. 如权利要求1～9任一项所述的方法，其特征在于，所述方法还包括：
    所述终端设备接收来自所述网络设备的工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式为AI模式，所述AI模式用 于指示所述终端设备使用所述第一AI模型执行通信业务。
  11. 如权利要求10所述的方法，其特征在于，所述工作模式配置信息还包括所述第一AI模型的生效时间。
  12. 一种人工智能AI模型传输方法，其特征在于，包括：
    网络设备接收来自终端设备的第一信息，所述第一信息包括所述终端设备支持的AI模型格式，所述终端设备支持的AI模型格式包括第一AI模型格式；
    所述网络设备向所述终端设备发送第二信息，所述第二信息用于指示第一AI模型的获取方法信息；其中，所述第一AI模型是第一AI模型格式表示的，所述第一AI模型格式是根据所述第一信息确定的，和/或，所述第一AI模型应用于第一通信场景。
  13. 如权利要求12所述的方法，其特征在于，所述获取方法信息包括所述第一AI模型的下载地址。
  14. 如权利要求12或13所述的方法，其特征在于，所述方法还包括：
    所述网络设备接收来自所述终端设备的第三信息，所述第三信息用于指示所述终端设备是否使用所述第一AI模型执行通信业务。
  15. 如权利要求12～14任一项所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
    深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
  16. 如权利要求12～15任一项所述的方法，其特征在于，所述第一信息还包括第一阈值和/或第二阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
  17. 如权利要求16所述的方法，其特征在于，所述第一AI模型满足所述终端设备的第一硬件要求，所述第一硬件要求包括：所述第一阈值所指示的时延要求，和/或所述第一AI模型满足所述第二阈值所指示的功耗要求。
  18. 如权利要求12～17任一项所述的方法，其特征在于，所述第一信息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力。
  19. 如权利要求18所述的方法，其特征在于，所述第一AI模型满足所述终端设备的第二硬件要求，所述第二硬件要求包括：所述存储能力所指示的存储空间要求。
  20. 如权利要求12～19任一项所述的方法，其特征在于，所述第一信息还包括AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
  21. 如权利要求20所述的方法，其特征在于，所述方法还包括：所述网络设备根据所述第一信息确定所述第一AI模式格式为深度学习框架输出的模型格式或中间表示层模型格式。
  22. 如权利要求12～19任一项所述的方法，其特征在于，所述方法还包括：
    所述网络设备在所述终端设备不支持AI模型编译能力时，确定所述第一AI模式格式为可执行文件表示的模型格式。
  23. 如权利要求12～22任一项所述的方法，其特征在于，所述方法还包括：
    所述网络设备向所述终端设备发送工作模式配置信息，所述工作模式配置信息用于配置所述终端设备执行通信业务的工作模式，所述工作模式为AI模式，所述AI模式用于指示所述终端设备使用所述第一AI模型执行通信业务。
  24. 如权利要求23所述的方法，其特征在于，所述工作模式配置信息还包括所述第一AI模型的生效时间。
  25. 一种人工智能AI模型传输方法，其特征在于，包括：
    第一设备接收来自终端设备的请求消息；
    所述第一设备根据所述请求消息，判断第一AI模型是否满足所述终端设备的硬件要求；和/或，所述第一设备根据所述请求消息，确定第一AI模型格式表示的第一AI模型；和/或，所述请求消息用于指示第一通信场景，所述第一设备根据所述请求消息，确定所述终端设备用于所述第一通信场景的第一AI模型；
    所述第一设备向所述终端设备发送第四信息，所述第四信息包括以下一项或多项：第一AI模型是否满足所述终端设备的硬件要求的判断结果，所述第一AI模型格式表示的第一AI模型，所述终端设备用于所述第一通信场景的第一AI模型。
  26. 如权利要求25所述的方法，其特征在于，所述第一AI模型格式包括以下一项或多项：深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
  27. 如权利要求25或26所述的方法，其特征在于，所述请求消息携带第一阈值，所述第一阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大时延。
  28. 如权利要求27所述的方法，其特征在于，所述判断第一AI模型是否满足终端设备的硬件要求，包括：
    在使用所述第一AI模型执行通信业务时该AI模型执行部分的时延不超过第一阈值时，确定所述第一AI模型满足所述终端设备的时延要求；和/或，在使用所述第一AI模型执行通信业务时该AI模型执行部分的时延超过所述第一阈值时，确定所述第一AI模型不满足所述终端设备的时延要求。
  29. 如权利要求25～28任一项所述的方法，其特征在于，所述请求消息携带第二阈值，所述第二阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
  30. 如权利要求29所述的方法，其特征在于，所述判断第一AI模型是否满足终端设备的硬件要求，包括：
    在使用所述第一AI模型执行通信业务时该AI模型执行部分的功耗不超过第二阈值时，确定所述第一AI模型满足所述终端设备的功耗要求；和/或，在使用所述第一AI模型执行通信业务时该AI模型执行部分的功耗超过所述第二阈值时，确定所述第一AI模型不满足所述终端设备的功耗要求。
  31. 如权利要求25～30任一项所述的方法，其特征在于，所述请求消息还包括以下一种或多种：硬件型号、硬件版本、存储能力、计算能力或支持异构计算的能力。
  32. 如权利要求31所述的方法，其特征在于，所述判断第一AI模型是否满足终端设备的硬件要求，包括：
    在所述第一AI模型占用的存储空间不超过所述终端设备的存储能力所指示的存储空间时，确定所述第一AI模型满足所述终端设备的存储空间要求；和/或，在所述第一AI模型占用的存储空间超过所述终端设备的存储能力所指示的存储空间时，确定所述第一AI模型不满足所述终端设备的存储空间要求。
  33. 一种通信装置，其特征在于，包括用于执行如权利要求1～11任一项所述方法的模块，或者包括用于执行如权利要求12～24任一项所述方法的模块，或者包括用于执行如权利要求25～32任一项所述方法的模块。
  34. 一种通信装置，其特征在于，包括处理器和接口电路，所述接口电路用于接收来自所述通信装置之外的其它通信装置的信号并传输至所述处理器或将来自所述处理器的信号发送给所述通信装置之外的其它通信装置，所述处理器通过逻辑电路或执行代码指令用于实现如权利要求1～11任一项所述的方法或如权利要求12～24任一项所述的方法或如权利要求25～32任一项所述的方法。
  35. 一种计算机可读存储介质，其特征在于，所述存储介质中存储有计算机程序或指令，当所述计算机程序或指令被通信装置执行时，实现如权利要求1～11任一项所述的方法或如权利要求12～24任一项所述的方法或如权利要求25～32任一项所述的方法。
  36. 一种包含指令的计算机程序产品，其特征在于，当所述计算机程序产品在计算机上运行时，使得如权利要求1～11任一项所述的方法或如权利要求12～24任一项所述的方法或如权利要求25～32任一项所述的方法被执行。
  37. 一种人工智能AI模型传输方法，其特征在于，包括：
    网络设备向第一网元发送第六信息，所述第六信息包括所述网络设备支持的AI模型格式，所述网络设备支持的AI模型格式包括第三AI模型格式，和/或，所述第六消息用于指示第一通信场景；
    所述网络设备接收来自所述第一网元的第七信息，所述第七信息用于指示所述第三AI模型格式表示的第二AI模型的获取方法信息，或者所述第七信息包括所述第三AI模型格式表示的第二AI模型，所述第三AI模型格式是根据所述第六信息确定的。
  38. 如权利要求37所述的方法，其特征在于，所述方法还包括：
    所述网络设备根据所述获取方法信息，向第二网元发送第一请求消息，所述第一请求消息用于请求所述第二AI模型；
    所述网络设备接收来自所述第二网元的所述第二AI模型。
  39. 如权利要求37或38所述的方法，其特征在于，所述第六信息还包括第三阈值和/或第四阈值，所述第三阈值为所述网络设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第四阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
  40. 如权利要求37～39任一项所述的方法，其特征在于，所述第六信息还包括所述网络设备的以下一种或多种的信息：硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力、AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
  41. 如权利要求37～40任一项所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
    深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
  42. 如权利要求37～41任一项所述的方法，其特征在于，所述方法还包括：
    所述网络设备向终端设备发送第一指示信息，所述第一指示信息用于指示一个或多个通信场景启用AI通信模式，所述一个或多个通信场景包括所述第一通信场景。
  43. 一种人工智能AI模型传输方法，其特征在于，包括：
    第一网元接收来自网络设备的第六信息，所述第六信息包括所述网络设备支持的AI模型格式，所述网络设备支持的AI模型格式包括第三AI模型格式，和/或，所述第六消息用于指示第一通信场景；
    所述第一网元向所述网络设备发送第七信息，所述第七信息用于指示所述第三AI模 型格式表示的第二AI模型的获取方法信息，或者所述第七信息包括所述第三AI模型格式表示的第二AI模型，所述第三AI模型格式是根据所述第六信息确定的。
  44. 如权利要求43所述的方法，其特征在于，所述方法还包括：
    所述第一网元向第二网元发送请求消息，所述请求消息用于请求所述第二AI模型；
    所述第一网元接收来自所述第二网元的所述第二AI模型。
  45. 如权利要求43或44所述的方法，其特征在于，所述第六信息还包括第三阈值和/或第四阈值，所述第三阈值为所述网络设备使用AI模型执行通信业务时该AI模型执行部分的最大时延，所述第四阈值为所述终端设备使用AI模型执行通信业务时该AI模型执行部分的最大功耗。
  46. 如权利要求43～45任一项所述的方法，其特征在于，所述第六信息还包括所述网络设备的以下一种或多种的信息：硬件型号、硬件版本、存储能力、计算能力、支持异构计算的能力、AI模型编译能力，所述AI模型编译能力包括编译转换AI模型格式的能力。
  47. 如权利要求43～46任一项所述的方法，其特征在于，所述AI模型格式包括以下一项或多项：
    深度学习框架输出的模型格式、中间表示层模型格式、或可执行文件表示的模型格式。
  48. 一种通信装置，其特征在于，包括用于执行如权利要求37～42任一项所述方法的模块，或者包括用于执行如权利要求43～47任一项所述方法的模块。
  49. 一种通信装置，其特征在于，包括处理器和接口电路，所述接口电路用于接收来自所述通信装置之外的其它通信装置的信号并传输至所述处理器或将来自所述处理器的信号发送给所述通信装置之外的其它通信装置，所述处理器通过逻辑电路或执行代码指令用于实现如权利要求37～42任一项所述的方法或如权利要求43～47任一项所述的方法。
  50. 一种计算机可读存储介质，其特征在于，所述存储介质中存储有计算机程序或指令，当所述计算机程序或指令被通信装置执行时，实现如权利要求37～42任一项所述的方法或如权利要求43～47任一项所述的方法。
  51. 一种包含指令的计算机程序产品，其特征在于，当所述计算机程序产品在计算机上运行时，使得如权利要求37～42任一项所述的方法或如权利要求43～47任一项所述的方法。
  52. 一种装置，其特征在于，包括电路，所述电路用于执行如权利要求1～32或37～47中任一项所述的方法。
  53. 一种装置，其特征在于，包括处理器，所述处理器与存储器耦合，所述存储器存储有指令，所述指令被执行时，使得所述装置执行如权利要求1～32或37～47中任一项所述的方法。
  54. 一种通信装置，其特征在于，包括处理器和存储器，所述处理器与所述存储器耦合；
    存储器存储有计算机程序；
    处理器，用于执行所述存储器中存储的计算机程序，以使得所述装置执行如权利要求1～32或37～47中任一所述的方法。
  55. 一种芯片，其特征在于，所述芯片与存储器耦合，用于读取并执行所述存储器中存储的程序指令，以执行如权利要求1～32或37～47中任一所述的方法。
  56. 一种通信系统，其特征在于，包括终端设备、网络设备和第一设备中的至少两项，所述终端设备用于执行如权利要求1～11中任一所述的方法，所述网络设备用于执行如权利要求12～24中任一所述的方法，所述第一设备用于执行如权利要求25～32中任一所述的 方法。
  57. 一种通信系统，其特征在于，包括网络设备和第一网元，所述网络设备用于执行如权利要求37～42中任一所述的方法，所述第一网元用于执行如权利要求43～47中任一所述的方法。
PCT/CN2022/112905 2021-08-17 2022-08-17 一种人工智能ai模型传输方法及装置 WO2023020516A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
EP22857827.4A EP4383774A1 (en) 2021-08-17 2022-08-17 Artificial intelligence ai model transmission method and apparatus

Applications Claiming Priority (6)

Application Number Priority Date Filing Date Title
CN202110945160 2021-08-17
CN202110945160.3 2021-08-17
CN202111128825 2021-09-26
CN202111128825.8 2021-09-26
CN202111576356.6 2021-12-22
CN202111576356.6A CN115942298A (zh) 2021-08-17 2021-12-22 一种人工智能ai模型传输方法及装置

Publications (1)

Publication Number Publication Date
WO2023020516A1 true WO2023020516A1 (zh) 2023-02-23

Family

ID=85240100

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/112905 WO2023020516A1 (zh) 2021-08-17 2022-08-17 一种人工智能ai模型传输方法及装置

Country Status (2)

Country Link
EP (1) EP4383774A1 (zh)
WO (1) WO2023020516A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024183608A1 (zh) * 2023-03-07 2024-09-12 华为技术有限公司 通信方法及相关装置
WO2024208308A1 (zh) * 2023-04-06 2024-10-10 维沃软件技术有限公司 人工智能ai单元的标识方法及通信设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180248831A1 (en) * 2017-02-24 2018-08-30 David Fletcher Methods and systems for electronic messaging management
CN111316227A (zh) * 2018-08-20 2020-06-19 华为技术有限公司 一种调试应用程序的方法及设备
CN111753948A (zh) * 2020-06-23 2020-10-09 展讯通信（上海）有限公司 模型处理方法及相关设备
CN111837425A (zh) * 2020-06-10 2020-10-27 北京小米移动软件有限公司 一种接入方法、接入装置及存储介质
CN112199385A (zh) * 2020-09-30 2021-01-08 北京百度网讯科技有限公司 用于人工智能ai的处理方法、装置、电子设备和存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180248831A1 (en) * 2017-02-24 2018-08-30 David Fletcher Methods and systems for electronic messaging management
CN111316227A (zh) * 2018-08-20 2020-06-19 华为技术有限公司 一种调试应用程序的方法及设备
CN111837425A (zh) * 2020-06-10 2020-10-27 北京小米移动软件有限公司 一种接入方法、接入装置及存储介质
CN111753948A (zh) * 2020-06-23 2020-10-09 展讯通信（上海）有限公司 模型处理方法及相关设备
CN112199385A (zh) * 2020-09-30 2021-01-08 北京百度网讯科技有限公司 用于人工智能ai的处理方法、装置、电子设备和存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024183608A1 (zh) * 2023-03-07 2024-09-12 华为技术有限公司 通信方法及相关装置
WO2024208308A1 (zh) * 2023-04-06 2024-10-10 维沃软件技术有限公司 人工智能ai单元的标识方法及通信设备

Also Published As

Publication number Publication date
EP4383774A1 (en) 2024-06-12

Similar Documents

Publication Publication Date Title
WO2023020516A1 (zh) 一种人工智能ai模型传输方法及装置
Ateya et al. Study of 5G services standardization: Specifications and requirements
US11432204B2 (en) Method and apparatus for enhancing handover procedure for supporting conditional handover in wireless communication system
TW202037208A (zh) 憑藉5g及其上之外之工業自動化
CN113329431B (zh) 一种无线承载的配置方法、终端、存储介质及芯片
CN114762388B (zh) 清除单个pdu传输和侧链路资源分配的侧链路许可的一部分
EP4391612A1 (en) Method, apparatus and system for downloading artificial intelligence model
US20230262728A1 (en) Communication Method and Communication Apparatus
CN114258151A (zh) 一种计算数据传输方法及装置
CN115942298A (zh) 一种人工智能ai模型传输方法及装置
Gebremariam et al. SoftPSN: software‐defined resource slicing for low‐latency reliable public safety networks
WO2023044904A1 (zh) 通信方法、终端设备及网络设备
WO2024060139A1 (zh) 通信方法、装置、存储介质及程序产品
WO2024174070A1 (en) Methods, devices, and computer readable medium for communication
WO2024197929A1 (en) Methods, devices, and computer readable medium for communication
WO2023206163A1 (zh) 无线通信的方法、网络设备和终端设备
EP4369253A1 (en) Dataset sharing transmission instructions for separated two-sided ai/ml based model training
WO2024193453A1 (zh) 定位方法、通信装置及存储介质
WO2024131885A1 (zh) 通信方法与装置、终端设备、网络设备和芯片
WO2023015499A1 (zh) 无线通信的方法和设备
WO2024046419A1 (zh) 一种通信方法及装置
WO2023066051A1 (zh) 一种区块链信息的传输方法, 装置及系统
WO2023125339A1 (zh) 参考信号的传输方法及装置
CN118474791A (zh) 信息传输方法和装置
KR20240113453A (ko) 무선 통신 시스템에서 통신을 수행하는 방법 및 장치

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: 22857827

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2022857827

Country of ref document: EP

Effective date: 20240307

NENP Non-entry into the national phase

Ref country code: DE