Communication method, communication apparatus, and communication system

Abstract

A communication method, a communication apparatus, and a communication system. On the basis of the method, the first model comprises a plurality of terminal-side models and a plurality of network-side models, a terminal uses a first terminal-side model among the plurality of terminal-side models to perform inference to obtain a first inference result, and on the basis of the first inference result and a first network-side model corresponding to the first terminal-side model, a network side continues to perform inference to obtain a second inference result. The method can enable selection of a suitable terminal-side model and network-side model from among a plurality of terminal-side models and a plurality of network-side models to perform inference, can ensure that an air interface transmission delay of a terminal when transmitting an inference result over an air interface does not change excessively, and thus can ensure an end-to-end service delay in a terminal-network co-inference scenario.

Description

Communication method, communication apparatus, and communication system

[0001] Cross-reference to Related Applications

[0002] This application claims priority to the Chinese Patent Application No. 202410985092.7, filed on July 22, 2024, and entitled “A Communication Method, Communication Apparatus, and Communication System”, the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0003] The present application relates to the field of communication technology, and in particular to a communication method, a communication apparatus, and a communication system. BACKGROUND

[0004] With the development of artificial intelligence (AI) technology, AI models gradually evolve from traditional small-scale models (referred to as small models), such as convolution neutral network (CNN) and regression neutral network (RNN), to large-scale models (referred to as large models). At the same time, terminal applications based on large models are emerging, such as question and answer systems, text-to-image, text-to-video, and various natural language processing and multi-modal applications. In such applications, in order to improve user experience, it is usually necessary to reduce the end-to-end service latency, for example, to control it within 300-500 milliseconds.

[0005] When deploying a model, a network management system can determine whether to split the model and how to split and distribute the model based on the computing power of a terminal. For example, if the computing power of the terminal meets the requirements of model inference, the network management system determines to split the model and determines the splitting rate of the model based on the computing power of the terminal. Hereinafter, the complete model before splitting is referred to as model #1, and it is assumed that the splitting rate is 1:9, which means that 10% of model #1 (hereinafter referred to as model #2) is deployed to the terminal, and the remaining 90% of the model (hereinafter referred to as model #3) is deployed to the network side. Taking model #1 including 100 layers as an example, the first 10 layers of model #1 (i.e., model #2) can be deployed to the terminal, and the last 90 layers of the model (i.e., model #3) can be deployed to the network side. Model #2 and model #3 jointly constitute model #1. When subsequent inference using the model is needed, the terminal first performs inference based on the deployed model #2 to obtain an intermediate inference result, and transmits the intermediate inference result to the network side through the air interface, and the network side continues to perform inference based on the intermediate inference result using model #3 to obtain a final inference result.

[0006] The inference manner jointly performed by the terminal and the network side can be referred to as end-network collaborative inference. The end-network collaborative inference has the following advantages: first, since the computing resources of the terminal are also used for model inference, system resource utilization can be maximized, which is beneficial to improving model inference speed and accuracy; second, since the terminal participates in model inference, it is possible to protect user privacy.

[0007] In the end-network collaborative inference scenario, the end-to-end service latency of user experience mainly depends on the air interface transmission latency of the inference result in the air interface transmission. However, the randomness of the air interface channel state is strong, which can cause the end-to-end service latency of user experience to be unable to be guaranteed. SUMMARY

[0008] Embodiments of the present application provide a communication method, a communication device and a communication system to guarantee the end-to-end service latency in the end-network collaborative inference scenario.

[0009] In a first aspect, embodiments of the present application provide a communication method, which can be applied to the terminal side, such as a terminal or a communication module in the terminal, or a circuit or chip responsible for communication functions in the terminal (such as a modem chip, also known as a baseband chip, or a system on chip (SoC) chip or a system in package (SIP) chip containing a modem core). The method comprises: sending a first request to a first device, the first request comprising an identifier of a first model and model selection assistance information, the first request being used to request a terminal-side model of the first model corresponding to the model selection assistance information, the model selection assistance information comprising a data volume of input data of the first model; receiving indication information from the first device, the indication information being used to indicate a first terminal-side model of the first model, the first terminal-side model being one of a plurality of terminal-side models of the first model, each terminal-side model in the plurality of terminal-side models being a sub-model of the first model processed for terminal inference; determining a first inference result according to the input data and the first terminal-side model; sending an inference request, the inference request comprising the first inference result and first information, the inference request being used to request continued inference based on the first inference result, the first information being used to determine a first network-side model of the first model, the first network-side model being one of a plurality of network-side models of the first model, each network-side model in the plurality of network-side models being a sub-model of the first model processed for network inference, the first network-side model corresponding to the first terminal-side model; and receiving a second inference result, the second inference result being obtained according to the first inference result and the first network-side model.

[0010] Based on the above scheme, when the terminal has reasoning requirements, the terminal requests a terminal-side model for reasoning from the first device, the first terminal-side model selected by the first device for reasoning is one of the plurality of terminal-side models of the first model, and the first terminal-side model is related to the data amount of the input data when the terminal performs model reasoning, that is, the method selects a suitable terminal-side model for model reasoning based on the data amount of the input data of the model reasoning. This method can ensure that the air interface transmission delay of the terminal when transmitting the reasoning result over the air interface will not change greatly, thus ensuring the end-to-end service delay in the end-network collaborative reasoning scenario.

[0011] In a possible implementation method, the model selection assistance information further includes a terminal-side estimated reasoning duration corresponding to the data amount of the input data.

[0012] Based on the above scheme, the terminal-side model for terminal model reasoning is selected based on the terminal-side estimated reasoning duration corresponding to the data amount of the input data, which helps to select a suitable terminal-side model and further helps to ensure the end-to-end service delay in the end-network collaborative reasoning scenario.

[0013] In a possible implementation method, the indication information includes an identifier of the first model and an identifier of the first terminal-side model.

[0014] Based on the above scheme, the selected terminal-side model for terminal model reasoning is indicated to the terminal by the identifier of the first model and the identifier of the first terminal-side model, which can accurately indicate the selected terminal-side model.

[0015] In a possible implementation method, the indication information includes an identifier of the first model and a first uplink transmission rate of the terminal, and the first uplink transmission rate corresponds to the first terminal-side model.

[0016] Based on the above scheme, the selected terminal-side model for terminal model reasoning is indicated to the terminal by the identifier of the first model and the first uplink transmission rate corresponding to the first terminal-side model, which can accurately indicate the selected terminal-side model.

[0017] In a possible implementation method, first association information is received, the first association information is used to indicate a one-to-one correspondence relationship between a plurality of uplink transmission rate ranges and the plurality of terminal-side models, the plurality of uplink transmission rate ranges include a first uplink transmission rate range, and the first uplink transmission rate is included in the first uplink transmission rate range; and the first terminal-side model is determined according to the first association information and the indication information.

[0018] Based on the above scheme, by indicating a one-to-one correspondence between the multiple uplink transmission rate ranges and the multiple terminal side models through the first association information, when the indication information received by the subsequent terminal includes a certain uplink transmission rate, the terminal can determine the terminal side model selected by the network side for the terminal to perform model inference based on the uplink transmission rate and the first association information, and accurate indication of the selected terminal side model can be realized.

[0019] In a possible implementation method, the indication information includes an identifier of the first model and a first collaboration level, and the first collaboration level corresponds to the first terminal side model.

[0020] Based on the above scheme, by indicating the selected terminal side model for the terminal to perform model inference to the terminal through the identifier of the first model and the first collaboration level corresponding to the first terminal side model, accurate indication of the selected terminal side model can be realized.

[0021] In a possible implementation method, second association information is received, the second association information is used to indicate a one-to-one correspondence between multiple collaboration levels and the multiple terminal side models, and the multiple collaboration levels include the first collaboration level; and the first terminal side model is determined according to the second association information and the indication information.

[0022] Based on the above scheme, by indicating a one-to-one correspondence between the multiple collaboration levels and the multiple terminal side models through the second association information, when the indication information received by the subsequent terminal includes a certain collaboration level, the terminal can determine the terminal side model selected by the network side for the terminal to perform model inference based on the collaboration level and the second association information, and accurate indication of the selected terminal side model can be realized.

[0023] In a possible implementation method, the first information includes an identifier of the first model and an identifier of the first terminal side model; or the first information includes an identifier of the first model and a first uplink transmission rate of the terminal, and the first uplink transmission rate corresponds to the first network side model; or the first information includes an identifier of the first model and a first collaboration level, and the first collaboration level corresponds to the first terminal side model and the first network side model.

[0024] Based on the above scheme, by indicating the first terminal side model of the terminal to perform model inference to the network side through the first information, the network side can select a corresponding first network side model to continue inference based on the first information, which helps to guarantee the accuracy of inference.

[0025] In a possible implementation method, a first session establishment request is sent, the first session establishment request includes an identifier of the first model, and the first session establishment request is used to request establishment of a session for inference of the first model.

[0026] In a second aspect, an embodiment of the present application provides a communication method, which can be applied to a terminal side, such as a terminal, or a communication module in the terminal, or a chip responsible for communication functions in the terminal, such as a modem chip (also known as a baseband chip) or a SoC or SIP chip containing a modem module. The method comprises: determining a first uplink transmission rate of the terminal; determining a first terminal-side model of a first model according to the first uplink transmission rate, the first terminal-side model being one of a plurality of terminal-side models of the first model, each terminal-side model in the plurality of terminal-side models being a sub-model of the first model processed for terminal inference; determining a first inference result according to input data of the first model and the first terminal-side model; sending an inference request, the inference request comprising an identifier of the first model, the first inference result, and first information, the inference request being used to request continued inference based on the first inference result, the first information being used to determine a first network-side model of the first model, the first network-side model being one of a plurality of network-side models of the first model, each network-side model in the plurality of network-side models being a sub-model of the first model processed for network inference, the first network-side model corresponding to the first terminal-side model; and receiving a second inference result, the second inference result being obtained according to the first inference result and the first network-side model.

[0027] Based on the above scheme, when the terminal has an inference requirement, a first terminal-side model for inference is selected from a plurality of terminal-side models of a first model based on a first uplink transmission rate of the terminal, that is, the method is to select a suitable terminal-side model for model inference based on the first uplink transmission rate of the terminal. This method can ensure that the air interface transmission delay of the terminal when transmitting the inference result over the air interface does not change too much, thus ensuring the end-to-end service delay in the terminal-network collaborative inference scenario.

[0028] In a possible implementation method, the first association information is received, the first association information being used to indicate a one-to-one correspondence relationship between a plurality of uplink transmission rate ranges and the plurality of terminal-side models, the plurality of uplink transmission rate ranges comprising a first uplink transmission rate range, the first uplink transmission rate being included in the first uplink transmission rate range; and the determining the first terminal-side model of the first model according to the first uplink transmission rate comprises: determining the first terminal-side model according to the first association information and the first uplink transmission rate.

[0029] Based on the above scheme, the first association information indicates a one-to-one correspondence relationship between the plurality of uplink transmission rate ranges and the plurality of terminal side models, and the terminal can accurately determine the terminal side model used for model inference of the terminal based on the determined first uplink transmission rate and in combination with the first association information.

[0030] In a possible implementation, the first information includes an identifier of the first model and an identifier of the first terminal side model; or the first information includes an identifier of the first model and the first uplink transmission rate.

[0031] Based on the above scheme, the first information indicates the first terminal side model used for model inference of the terminal to the network side, so that the network side can select a corresponding first network side model for further inference based on the first information, which helps to ensure the accuracy of inference.

[0032] In a possible implementation, the first information includes an identifier of the first model and a first collaboration level, and the first collaboration level corresponds to the first terminal side model and the first network side model.

[0033] Based on the above scheme, the first information indicates the first terminal side model used for model inference of the terminal to the network side, so that the network side can select a corresponding first network side model for further inference based on the first information, which helps to ensure the accuracy of inference.

[0034] In a possible implementation, second association information is received, the second association information is used to indicate a one-to-one correspondence relationship between a plurality of collaboration levels and the plurality of terminal side models, and the plurality of collaboration levels includes the first collaboration level; and the first information is determined according to the second association information and the first terminal side model.

[0035] In a possible implementation, a first session establishment request is sent, the first session establishment request includes an identifier of the first model, and the first session establishment request is used to request to establish a session for inference of the first model.

[0036] In a third aspect, the method can be applied to a network side, for example, a first device at the network side, a module (for example, a circuit, a chip, or a chip system, etc.) in the first device, or a logic node, a logic module, or software capable of implementing all or part of the functions of the first device. The first device can be an access network device or a core network device. The method includes: receiving a first request from a terminal, the first request including an identification of a first model and model selection assistance information, the first request being used to request a terminal-side model of the first model corresponding to the model selection assistance information, the model selection assistance information including a data volume of input data of the first model; determining a first terminal-side model of the first model according to the model selection assistance information, the first terminal-side model being one of a plurality of terminal-side models of the first model, each terminal-side model in the plurality of terminal-side models being a sub-model of the first model processed for terminal inference; and sending indication information to the terminal, the indication information being used to indicate the first terminal-side model.

[0037] Based on the above scheme, when the terminal has an inference requirement, the terminal requests a terminal-side model for inference from the first device. The first terminal-side model selected by the first device for inference is one of a plurality of terminal-side models of the first model, and the first terminal-side model is related to the data volume of the input data when the terminal performs model inference. That is, the method selects a suitable terminal-side model for model inference based on the data volume of the input data for model inference. This method can ensure that the air interface transmission delay of the terminal when transmitting the inference result over the air interface does not change greatly, and thus can ensure the end-to-end service delay in the end-network collaborative inference scenario.

[0038] In a possible implementation method, the determining the first terminal-side model of the first model according to the model selection assistance information includes: determining an estimated data volume of output data corresponding to the plurality of terminal-side models respectively according to the data volume of the input data; determining an estimated uplink transmission rate corresponding to the plurality of terminal-side models respectively according to the estimated data volume of the output data corresponding to the plurality of terminal-side models respectively; determining a first uplink transmission rate according to the estimated uplink transmission rates corresponding to the plurality of terminal-side models respectively, the first uplink transmission rate being a maximum estimated uplink transmission rate that can be satisfied among the estimated uplink transmission rates corresponding to the plurality of terminal-side models respectively; and determining the first terminal-side model according to the first uplink transmission rate.

[0039] Based on the above scheme, the first device can accurately determine the first terminal-side model for the terminal to perform model inference. The first terminal-side model is related to the data volume of the input data of the first model, which helps to ensure that the air interface transmission delay of the terminal when transmitting the inference result over the air interface does not change greatly, and thus can ensure the end-to-end service delay in the end-network collaborative inference scenario.

[0040] In a possible implementation, the determining the first uplink transmission rate according to the estimated uplink transmission rates corresponding to the plurality of terminal-side models comprises: determining the first uplink transmission rate according to the estimated uplink transmission rates corresponding to the plurality of terminal-side models, and one or more of the identifier of the first model, the channel quality between the terminal and the first device, or the load of the first device.

[0041] Based on the above scheme, when determining the terminal-side model for the terminal to perform model inference, one or more of the identifier of the first model, the channel quality between the terminal and the first device, or the load of the first device is further referred to, which helps to more accurately determine a suitable terminal-side model, thereby further guaranteeing the end-to-end service delay in the end-network collaborative inference scenario.

[0042] In a possible implementation, the receiving the first association information comprises: receiving the first association information used to indicate a one-to-one correspondence relationship between a plurality of uplink transmission rate ranges and the plurality of terminal-side models, wherein the plurality of uplink transmission rate ranges comprise a first uplink transmission rate range, and the first uplink transmission rate is included in the first uplink transmission rate range; and the determining the first terminal-side model according to the first uplink transmission rate comprises: determining the first terminal-side model according to the first uplink transmission rate and the first association information.

[0043] Based on the above scheme, the first association information is used to indicate the one-to-one correspondence relationship between the plurality of uplink transmission rate ranges and the plurality of terminal-side models, and when the first device subsequently receives the first information from the terminal and the first information comprises a certain uplink transmission rate, the first device can determine the first terminal-side model used by the terminal based on the uplink transmission rate and the first association information, and then the first device can accurately determine the first network-side model corresponding to the first terminal-side model and use the first network-side model to perform network-side inference, which helps to improve the accuracy of inference.

[0044] In a possible implementation, the model selection assistance information further comprises a terminal-side estimated inference time length corresponding to the data amount of the input data.

[0045] Based on the above scheme, the terminal-side model used for the terminal to perform model inference is selected based on the terminal-side estimated inference time length corresponding to the data amount of the input data, which helps to select a suitable terminal-side model, thereby helping to guarantee the end-to-end service delay in the end-network collaborative inference scenario.

[0046] In a possible implementation, the indication information includes an identifier of the first model and an identifier of the first terminal-side model; or the indication information includes an identifier of the first model and a first uplink transmission rate of the terminal; or the indication information includes an identifier of the first model and a first cooperation level corresponding to the first terminal-side model.

[0047] Based on the above scheme, the terminal-side model selected for the terminal to perform model inference can be indicated to the terminal through the indication information, and accurate indication of the selected terminal-side model can be implemented.

[0048] In a fourth aspect, the method can be applied to a network side, for example, a first device of the network side, a module (for example, a circuit, a chip, or a chip system, etc.) in the first device, or a logical node, a logical module, or software capable of implementing all or part of the functions of the first device. The first device can be an access network device or a core network device. Alternatively, the method can be applied to a task execution function (TEF) of the network side, a module (for example, a circuit, a chip, or a chip system, etc.) in the TEF network element, or a logical node, a logical module, or software capable of implementing all or part of the functions of the TEF network element. The method includes: receiving an inference request, the inference request including a first inference result and first information, the inference request being used to request to continue inference based on the first inference result, and the first information being used to determine a first network-side model of a first model, the first network-side model being one of a plurality of network-side models of the first model, and each network-side model in the plurality of network-side models being a sub-model of the first model processed for a network to perform inference; and transmitting a second inference result according to the first network-side model and the first inference result.

[0049] Based on the above scheme, the first model includes a plurality of terminal-side models and a plurality of network-side models, the terminal uses a first terminal-side model in the plurality of terminal-side models to obtain a first inference result, and the network side continues inference based on the first inference result and a first network-side model corresponding to the first terminal-side model to obtain a second inference result. This method can select appropriate terminal-side models and network-side models from the plurality of terminal-side models and the plurality of network-side models for inference, and can ensure that the air interface transmission delay of the terminal when transmitting the inference result will not change too much, thus ensuring the end-to-end service delay in the end-network collaborative inference scenario.

[0050] In a possible implementation, the first information includes an identifier of the first model and an identifier of the first terminal-side model.

[0051] Based on the above scheme, the first terminal side model for model inference of the terminal is indicated to the network side through the identifier of the first model and the identifier of the first terminal side model, so that the network side can select the corresponding first network side model for continuous inference based on the identifier of the first model and the identifier of the first terminal side model, which helps to ensure the accuracy of inference.

[0052] In a possible implementation method, the first information includes the identifier of the first model and a first uplink transmission rate of the terminal, and the first uplink transmission rate corresponds to the first network side model.

[0053] Based on the above scheme, the first terminal side model for model inference of the terminal is indicated to the network side through the identifier of the first model and the first uplink transmission rate, so that the network side can select the corresponding first network side model for continuous inference based on the identifier of the first model and the first uplink transmission rate, which helps to ensure the accuracy of inference.

[0054] In a possible implementation method, the first network side model is determined according to third association information and the first uplink transmission rate, the third association information is used to indicate a one-to-one correspondence relationship between a plurality of uplink transmission rate ranges and a plurality of network side models, the plurality of uplink transmission rate ranges include a first uplink transmission rate range, and the first uplink transmission rate is included in the first uplink transmission rate range.

[0055] Based on the above scheme, the one-to-one correspondence relationship between the plurality of uplink transmission rate ranges and the plurality of network side models is indicated through the third association information, and when the first information including a certain uplink transmission rate is received subsequently, the network side model used for model inference of the network side can be accurately determined based on the uplink transmission rate and the third association information.

[0056] In a possible implementation method, the third association information is received from a second device.

[0057] In a possible implementation method, the first information includes the identifier of the first model and a first collaboration level, and the first collaboration level corresponds to the first network side model.

[0058] Based on the above scheme, the first terminal side model for model inference of the terminal is indicated to the network side through the identifier of the first model and the first collaboration level, so that the network side can select the corresponding first network side model for continuous inference based on the identifier of the first model and the first collaboration level, which helps to ensure the accuracy of inference.

[0059] In a possible implementation, fourth association information is received, where the fourth association information is used to indicate a one-to-one correspondence relationship between a plurality of collaboration levels and a plurality of network side models, and the plurality of collaboration levels includes the first collaboration level; and the first network side model is determined according to the fourth association information and the first collaboration level.

[0060] Based on the foregoing scheme, the one-to-one correspondence relationship between the plurality of collaboration levels and the plurality of network side models is indicated by the fourth association information, and when the first information received subsequently includes a certain collaboration level, the network side model used for network side model inference can be accurately determined based on the collaboration level and the fourth association information.

[0061] In a possible implementation, the fourth association information is received from the second device.

[0062] In a possible implementation, information of the plurality of network side models is received from the second device, and the information of each network side model in the plurality of network side models includes an identifier of the network side model and parameters of the network side model.

[0063] In a possible implementation, first association information is sent to a terminal, where the first association information is used to indicate a one-to-one correspondence relationship between a plurality of uplink transmission rate ranges and a plurality of terminal side models of the first model.

[0064] In a possible implementation, second association information is sent to a terminal, where the second association information is used to indicate a one-to-one correspondence relationship between a plurality of collaboration levels and a plurality of terminal side models of the first model, and the plurality of collaboration levels includes a first collaboration level corresponding to the first network side model.

[0065] In a possible implementation, information of the plurality of terminal side models is sent to the terminal, and the information of each terminal side model in the plurality of terminal side models includes an identifier of the terminal side model and parameters of the terminal side model.

[0066] In a possible implementation, a first session establishment request is received from a terminal, where the first session establishment request includes an identifier of the first model, and the first session establishment request is used to request establishment of a session for inference of the first model; information of a task execution function network element is acquired; and a second session establishment request is sent to the task execution function network element, where the second session establishment request includes the identifier of the first model, and the second session establishment request is used to request establishment of a session for inference of the first model.

[0067] Fifthly, this method can be applied to the network side, such as a second device on the network side, a module (e.g., a circuit, chip, or chip system) within the second device, or a logical node, logical module, or software capable of implementing all or part of the functions of the second device. The second device can be an element management system (EMS), a network management system (NMS), or a service management and orchestration (SMO) system, or a device within the EMS, NMS, or SMO system. Alternatively, this method can be applied to a TEF on the network side, a module (e.g., a circuit, chip, or chip system) within the TEF network element, or a logical node, logical module, or software capable of implementing all or part of the functions of the TEF network element. The method includes: obtaining a model processing strategy for a first model; determining multiple terminal-side models and multiple network-side models of the first model according to the model processing strategy, wherein each terminal-side model is a sub-model of the first model after processing for inference on a terminal, and each network-side model is a sub-model of the first model after processing for inference on a network, and the multiple terminal-side models and the multiple network-side models correspond one-to-one.

[0068] Based on the above scheme, the first model includes multiple terminal-side models and multiple network-side models, which can select appropriate terminal-side models and network-side models for inference. This can ensure that the air interface transmission latency of the terminal when transmitting the inference results over the air interface does not change too much, thus ensuring the end-to-end service latency in the terminal-network collaborative inference scenario.

[0069] In one possible implementation, the model processing strategy is used to indicate the model processing methods corresponding to multiple uplink transmission rate ranges, wherein the multiple uplink transmission rate ranges, the multiple terminal-side models, and the multiple network-side models correspond one-to-one; and first association information is sent, wherein the first association information is used to indicate the one-to-one correspondence between the multiple uplink transmission rate ranges and the multiple terminal-side models.

[0070] Based on the above scheme and the model processing strategy, the terminal-side model and network-side model corresponding to multiple uplink transmission rate ranges can be determined respectively. The first association information indicates the one-to-one correspondence between multiple uplink transmission rate ranges and multiple terminal-side models. Subsequently, the terminal-side model used by the terminal for model inference can be accurately determined based on the first association information.

[0071] In one possible implementation, a third association information is sent, which is used to indicate the one-to-one correspondence between the plurality of uplink transmission rate ranges and the plurality of network-side models.

[0072] Based on the above scheme, the third association information indicates the one-to-one correspondence between multiple uplink transmission rate ranges and multiple network-side models. Subsequently, the network-side model used for network model inference can be accurately determined based on the third association information.

[0073] In one possible implementation, the model processing strategy is used to indicate the model processing methods corresponding to multiple coordination levels, wherein the multiple coordination levels, the multiple terminal-side models, and the multiple network-side models are in one-to-one correspondence; and second association information is sent, wherein the second association information is used to indicate the one-to-one correspondence between the multiple coordination levels and the multiple terminal-side models.

[0074] Based on the above scheme and the model processing strategy, the terminal-side model and network-side model corresponding to multiple collaboration levels can be determined respectively. The one-to-one correspondence between multiple collaboration levels and multiple terminal-side models can be indicated by the second association information. Subsequently, the terminal-side model used by the terminal when performing model inference can be accurately determined based on the second association information.

[0075] In one possible implementation, a fourth association information is sent, which is used to indicate the one-to-one correspondence between the plurality of collaboration levels and the plurality of network-side models.

[0076] Based on the above scheme, the fourth association information indicates the one-to-one correspondence between multiple collaboration levels and multiple network-side models. Subsequently, the network-side model used when performing model inference can be accurately determined based on the fourth association information.

[0077] In a sixth aspect, this application provides a communication device that has the functions of the first aspect described above. For example, the communication device includes modules, units, or means corresponding to the operations involved in the first aspect described above. These modules, units, or means can be implemented by software, hardware, or a combination of software and hardware.

[0078] In a seventh aspect, this application provides a communication device that has the functions of the second aspect above. For example, the communication device includes modules, units or means that perform the operations involved in the second aspect above. These modules, units or means can be implemented by software, hardware or a combination of software and hardware.

[0079] Eighthly, this application provides a communication device that has the functions of the third aspect above. For example, the communication device includes modules, units or means corresponding to the operations involved in the third aspect above. The modules, units or means can be implemented by software, or by hardware, or by a combination of software and hardware.

[0080] Ninthly, this application provides a communication device that has the functions of the fourth aspect above. For example, the communication device includes modules, units or means corresponding to the operations involved in the fourth aspect above. The modules, units or means can be implemented by software, or by hardware, or by a combination of software and hardware.

[0081] In a tenth aspect, this application provides a communication device that has the functions of the fifth aspect described above. For example, the communication device includes modules, units, or means corresponding to the operations involved in the fifth aspect described above. These modules, units, or means can be implemented by software, hardware, or a combination of software and hardware.

[0082] Eleventhly, this application provides a communication device including an interface circuit and one or more processors. The one or more processors are coupled to a memory. The memory stores part or all of the necessary computer programs or instructions for implementing the functions described in the first and second aspects above. The one or more processors can execute the computer programs or instructions, and when the computer programs or instructions are executed, cause the communication device to implement the methods in any possible design or implementation of the first and second aspects above. The interface circuit is used to implement the communication functions within the communication device and / or the communication functions between the communication device and other devices or components.

[0083] In one possible design, the processor is used to communicate with other devices or components through the interface circuit.

[0084] In one possible design, the communication device may also include the memory.

[0085] The aforementioned communication device may be a terminal, a communication module in a terminal, or a chip in a terminal that is responsible for communication functions, such as a modem chip (also known as a baseband chip) or a SoC or SIP chip containing a modem module.

[0086] In a twelfth aspect, this application provides a communication device including an interface circuit and one or more processors. The one or more processors are coupled to a memory. The memory stores part or all of a computer program or instructions necessary for implementing the functions described in the third aspect above. The one or more processors are executable to carry out the computer program or instructions, causing the communication device to implement the methods in any possible design or implementation of the third aspect above. The interface circuit is used to implement communication functions within the communication device and / or communication functions between the communication device and other devices or components.

[0087] The aforementioned communication device may be a first device, a module within the first device (e.g., a circuit, chip, or chip system), or a logical node, logical module, or software capable of implementing all or part of the functions of the first device. The first device may be an access network device or a core network device.

[0088] In a thirteenth aspect, this application provides a communication device including an interface circuit and one or more processors. The one or more processors are coupled to a memory. The memory stores part or all of the necessary computer program or instructions for implementing the functions described in the fourth aspect above. The one or more processors are executable to carry out the computer program or instructions, causing the communication device to implement the methods in any possible design or implementation of the fourth aspect above. The interface circuit is used to implement the communication functions within the communication device and / or the communication functions between the communication device and other devices or components.

[0089] The aforementioned communication device may be a first device, a module within the first device (e.g., a circuit, chip, or chip system), or a logical node, logical module, or software capable of implementing all or part of the functions of the first device. The first device may be an access network device or a core network device.

[0090] Alternatively, the aforementioned communication device may also be a TEF, a module (e.g., a circuit, chip, or chip system) in a TEF network element, or a logical node, logical module, or software that can implement all or part of the functions of a TEF network element.

[0091] In a fourteenth aspect, this application provides a communication device including an interface circuit and one or more processors. The one or more processors are coupled to a memory. The memory stores part or all of a computer program or instructions necessary for implementing the functions described in the fifth aspect above. The one or more processors are executable to carry out the computer program or instructions, causing the communication device to implement the methods in any possible design or implementation of the fifth aspect above when executed. The interface circuit is used to implement communication functions within the communication device and / or communication functions between the communication device and other devices or components.

[0092] The aforementioned communication device may be a second device, a module within the second device (e.g., a circuit, chip, or chip system), or a logic node, logic module, or software capable of implementing all or part of the functions of the second device. The second device may be an EMS, NMS, or SMO system, or a device within an EMS, NMS, or SMO system.

[0093] Alternatively, the aforementioned communication device may also be a TEF, a module (e.g., a circuit, chip, or chip system) in a TEF network element, or a logical node, logical module, or software that can implement all or part of the functions of a TEF network element.

[0094] In a fifteenth aspect, this application provides a computer-readable storage medium storing a computer program or instructions that, when executed, implement the method in any of the possible designs of the first to fifth aspects described above.

[0095] In a sixteenth aspect, this application provides a computer program product including instructions that, when executed, implement the method in any of the possible designs of the first to fifth aspects described above.

[0096] In a seventeenth aspect, this application provides a communication system including a terminal for performing the method in any possible design of the first aspect described above, and a first device for performing the method in any possible design of the third aspect described above.

[0097] In an eighteenth aspect, this application provides a communication system, including a communication device and a terminal for performing the method in any possible design of the second aspect described above. The communication device is configured to receive an inference request from the terminal, the inference request including an identifier of a first model, a first inference result, and first information, the inference request being used to request continued inference based on the first inference result, the first information being used to determine a first network-side model of the first model, the first network-side model being one of a plurality of network-side models of the first model, each of the plurality of network-side models being a sub-model of the first model processed for network inference, the first network-side model corresponding to the first terminal-side model; and to send a second inference result to the terminal, the second inference result being obtained based on the first inference result and the first network-side model.

[0098] The communication device can be a first device or a TEF network element.

[0099] In a nineteenth aspect, this application provides a communication system, including a terminal and a communication device for performing the method in any possible design of the fourth aspect described above. The terminal is configured to send an inference request to the communication device, the inference request including a first inference result and first information, the inference request being used to request continued inference based on the first inference result, the first information being used to determine a first network-side model of a first model, the first network-side model being one of a plurality of network-side models of the first model, each of the plurality of network-side models being a sub-model of the first model processed for network inference; and to receive a second inference result from the communication device.

[0100] The communication device can be a first device or a TEF network element. Attached Figure Description

[0101] Figures 1 and 2 are schematic diagrams of possible, non-limiting systems applicable to this application;

[0102] Figures 3 to 7 are schematic flowcharts of the communication method provided in the embodiments of this application;

[0103] Figure 8 is a possible exemplary block diagram of a communication device provided in an embodiment of this application;

[0104] Figure 9 is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0105] Figure 1 is a possible, non-limiting system schematic diagram. As shown in Figure 1, the communication system 10 includes a radio access network (RAN) 100 and a core network (CN) 200. Optionally, the communication system also includes an Internet 300. RAN 100 includes at least one RAN node (110a and 110b in Figure 1, collectively referred to as 110) and at least one terminal (120a-120j in Figure 1, collectively referred to as 120). RAN 100 may also include other RAN nodes, such as wireless relay devices and / or wireless backhaul devices (not shown in Figure 1). Terminal 120 is wirelessly connected to RAN node 110. RAN node 110 is wirelessly or wired connected to core network 200. The core network equipment in core network 200 and RAN node 110 in RAN 100 can be different physical devices, or they can be the same physical device integrating core network logical functions and radio access network logical functions.

[0106] RAN100 can be a cellular system related to the 3rd Generation Partnership Project (3GPP), such as 4th generation (4G), 5th generation (5G) mobile communication systems, or future-oriented evolution systems (such as 6th generation (6G) mobile communication systems). RAN100 can also be an open RAN (O-RAN or ORAN), a cloud radio access network (CRAN), or a wireless fidelity (WiFi) system. RAN100 can also be a communication system that integrates two or more of the above systems.

[0107] RAN node 110, sometimes also referred to as access network equipment, RAN entity, or access node, constitutes part of the communication system and is used to help terminals achieve wireless access. Multiple RAN nodes 110 in communication system 10 can be of the same type or different types. In some scenarios, the roles of RAN node 110 and terminal 120 are relative. For example, network element 120i in Figure 1 can be a helicopter or drone, which can be configured as a mobile base station. For terminals 120j accessing RAN 100 through network element 120i, network element 120i is a base station; but for base station 110a, network element 120i is a terminal. RAN node 110 and terminal 120 are sometimes both referred to as communication devices. For example, network elements 110a and 110b in Figure 1 can be understood as communication devices with base station functions, and network elements 120a-120j can be understood as communication devices with terminal functions.

[0108] In one possible scenario, the RAN node can be a base station, an evolved NodeB (eNodeB), an access point (AP), a transmission reception point (TRP), a next-generation NodeB (gNB), a next-generation base station in a 6G mobile communication system, a base station in a future mobile communication system, or an access node in a WiFi system. The RAN node can be a macro base station (as shown in Figure 1, 110a), a micro base station or indoor station (as shown in Figure 1, 110b), a relay node or donor node, or a radio controller in a CRAN scenario. Optionally, the RAN node can also be a server, wearable device, vehicle, or in-vehicle equipment. For example, the access network equipment in vehicle-to-everything (V2X) technology can be a roadside unit (RSU). All or part of the functions of the RAN node in this application can also be implemented through software functions running on hardware, or through virtualization functions instantiated on a platform (e.g., a cloud platform). The RAN node can also be equipped with communication modules, circuits, or chips that perform corresponding communication functions. The RAN node can also be configured with program instructions for performing corresponding communication functions, as well as corresponding program instructions. The RAN node in this application can also be a logical node, logical module, or software capable of implementing all or part of the RAN node's functions.

[0109] In another possible scenario, multiple RAN nodes collaborate to assist the terminal in achieving wireless access, with each RAN node performing a portion of the base station's functions. For example, RAN nodes can be central units (CUs), distributed units (DUs), CU-control plane (CPs), CU-user plane (UPs), or radio units (RUs), etc. CUs and DUs can be separate entities or included in the same network element, such as a baseband unit (BBU). RUs can be included in radio frequency equipment or radio frequency units, such as remote radio units (RRUs), active antenna units (AAUs), or remote radio heads (RRHs).

[0110] In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an ORAN system, CU can also be called O-CU (open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. For ease of description, this application uses CU, CU-CP, CU-UP, DU, and RU as examples. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software and hardware modules.

[0111] A terminal can be a device or module that accesses the aforementioned communication system and has corresponding communication functions. A terminal can also be called a terminal device, user equipment (UE), mobile station, mobile terminal, etc. Terminals can be widely used in various scenarios, such as device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IoT), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart transportation, smart cities, etc. Terminals can be mobile phones, tablets, computers with wireless transceiver capabilities, wearable devices, vehicles, drones, helicopters, airplanes, ships, robots, robotic arms, smart home devices, transportation vehicles with wireless communication capabilities, communication modules, etc. The embodiments of this application do not limit the device form of the terminal. A terminal typically contains a communication module, circuit, or chip that performs the corresponding communication function. The terminal can also be configured with program instructions for performing the corresponding communication function.

[0112] Figure 2 shows another possible, non-limiting system diagram. This system includes NMS, EMS, RAN domain network elements, and core network (CN) domain network elements.

[0113] The Network Management System (NMS) is responsible for the operation, management, and maintenance of the network. It can also be called a cross-domain management system. The NMS can manage the EMS of various domains, such as the EMS of the RAN domain and the EMS of the CN domain.

[0114] Each domain's EMS can manage network elements within that domain. For example, the RAN domain's EMS can manage access network equipment, while the CN domain's EMS can manage network data analytics function (NWDAF) network elements, user plane function (UPF) network elements, session management function (SMF) network elements, and access and mobility management function (AMF) network elements. EMS is also known as a domain management system or a single-domain management system.

[0115] TEF network elements have computational and execution functions, as well as being responsible for local model inference.

[0116] The task management function (TMF) network element has task management related functions, and is responsible for task authorization, task management (such as task creation, deletion, and modification).

[0117] The task control function (TCF) network element has computing control-related functions, and is responsible for the selection of computing nodes, the division of computing tasks, computing power registration and discovery, and general computing scheduling.

[0118] Among them, TEF network elements, TMF network elements, and TCF network elements can all be located in the RAN domain, all in the CN domain, or partly in the RAN domain and partly in the CN domain. Furthermore, TEF network elements, TMF network elements, and TCF network elements are managed by the EMS of their respective domains.

[0119] TEF, TMF, and TCF network elements can be deployed in any two pairs, all three together, or independently. Furthermore, one or more of the TEF, TMF, or TCF elements can be deployed on access network equipment or on core network equipment (e.g., NWDAF network elements). This application does not impose any restrictions on the deployment method of TEF, TMF, and TCF network elements.

[0120] In addition, ORAN also includes the SMO system. The role of the SMO system in the network architecture is similar to that of the NMS. It is responsible for the operation, management and maintenance of network services and orchestration functions in each domain. For example, it can manage network elements in each domain (such as access network equipment, NWDAF network elements, TEF network elements, TMF network elements, TCF network elements, etc.).

[0121] Data or model information in the network can be made available to the terminal manufacturer's over-the-top (OTT) server through the NMS or SMO system, and the OTT server will then send the data or model information to the terminal.

[0122] With the development of AI technology, AI models are gradually evolving from traditional small-scale models (referred to as small models), such as CNNs and RNNs, to large-scale models (referred to as large models). At the same time, terminal applications based on large models are constantly emerging, such as question-answering systems, text-to-image processing, text-to-video processing, and various other natural language processing and multimodal applications. In these applications, to improve user experience, it is usually necessary to reduce end-to-end service latency, for example, to within 300–500 milliseconds.

[0123] During model deployment, the network management system can determine whether and how to segment the model and perform distributed deployment based on the terminal's computing power. For example, if the terminal's computing power meets the requirements for model inference, the network management system determines to segment the model and sets the segmentation ratio based on the terminal's computing power. The complete model before segmentation is referred to as Model #1. Assuming a segmentation ratio of 1:9, this means deploying 10% of Model #1 (hereinafter referred to as Model #2) to the terminal and the remaining 90% (hereinafter referred to as Model #3) to the network side. Taking Model #1 as having 100 layers as an example, the first 10 layers (i.e., Model #2) of Model #1 can be deployed to the terminal, and the last 90 layers (i.e., Model #3) can be deployed to the network side. Model #2 and Model #3 together constitute Model #1. When subsequent inference is needed, the terminal first performs inference based on the deployed Model #2 to obtain intermediate inference results, which are then sent to the network side via the air interface. The network side then uses Model #3 to continue inference based on the intermediate inference results to obtain the final inference result.

[0124] Joint inference by the terminal and network can be called end-to-end collaborative inference. End-to-end collaborative inference has the following advantages: First, since the terminal's computing resources are also used for model inference, the system resource utilization can be maximized, which is conducive to improving the speed and accuracy of model inference; Second, since the terminal participates in model inference, it provides the possibility of protecting user privacy.

[0125] In the scenario of end-to-end collaborative inference, the end-to-end service latency of the user experience mainly depends on the air interface transmission latency of the intermediate inference results. However, the randomness of the air interface channel state may lead to the inability to guarantee the end-to-end service latency of the user experience.

[0126] To address the aforementioned issues, this application provides corresponding solutions.

[0127] The communication method and apparatus are described below with reference to the accompanying drawings. It is understood that this application uses a first device, a second device, a TEF network element, and a terminal as examples to illustrate the execution of the interaction, but this application does not limit the execution subject of the interaction. For example, the method executed by the first device in this application can also be implemented by a module (e.g., a circuit, chip, or chip system) in the first device, or a logic node, logic module, or software capable of implementing all or part of the functions of the first device. Similarly, the method executed by the second device in this application can also be implemented by a module (e.g., a circuit, chip, or chip system) in the second device, or a logic node, logic module, or software capable of implementing all or part of the functions of the second device. Likewise, the method executed by a TEF network element in this application can also be implemented by a module (e.g., a circuit, chip, or chip system) in the TEF network element, or a logic node, logic module, or software capable of implementing all or part of the functions of the TEF network element. Finally, the method executed by the terminal in this application can also be implemented by a communication module in the terminal, or by a circuit or chip in the terminal responsible for communication functions (such as a modem chip (also known as a baseband chip), or a SoC chip containing a modem core, or a SIP chip).

[0128] In this application, the first device may be an access network device or a core network device (e.g., an NWDAF network element, a UPF network element, an SMF network element, or an AMF network element, etc.). The second device may be an EMS, an NMS, or an SMO system, or a device within an EMS, an NMS, or an SMO system.

[0129] Figure 3 is a flowchart illustrating a communication method provided in an embodiment of this application. The method includes the following steps:

[0130] Step 301: The terminal sends a first request to the first device. Accordingly, the first device receives the first request.

[0131] The first request includes the identifier of the first model and model selection assistance information. The first request is used to request the terminal-side model of the first model corresponding to the model selection assistance information.

[0132] For example, the identifier of the first model can be an application identifier (APP ID) or a service identifier (service ID). That is, one application or one service corresponds to one model.

[0133] The model selection auxiliary information may include the amount of data in the input data of the first model, which is used to indicate the size of the input data used for model inference.

[0134] Optionally, the model selection auxiliary information also includes the terminal-side estimated inference time corresponding to the amount of input data. That is, the terminal-side estimated inference time represents the estimated time for the terminal to use the input data to perform model inference and obtain the inference result.

[0135] For example, the first request may also be referred to as a pre-scheduled request, a model request, etc.

[0136] Step 302: The first device selects auxiliary information based on the model to determine the first terminal-side model of the first model.

[0137] The first model corresponds to multiple terminal-side models and multiple network-side models, with a one-to-one correspondence between the multiple terminal-side models and the multiple network-side models. Each of the multiple terminal-side models is a sub-model of the first model processed for inference on the terminal, and each of the multiple network-side models is a sub-model of the first model processed for inference on the network.

[0138] The plurality of terminal-side models includes at least a first terminal-side model, meaning the first terminal-side model is one of the plurality of terminal-side models of the first model. The plurality of network-side models includes at least a first network-side model, meaning the first network-side model is one of the plurality of network-side models of the first model. The first terminal-side model corresponds to the first network-side model, meaning that the first model, under certain processing, yields both the first terminal-side model and the first network-side model.

[0139] In one implementation, the first model is divided into multiple sub-models according to the same model partitioning ratio. Each sub-model group includes a terminal-side model and a network-side model. The model partitioning ratio indicates the proportion in which the first model is divided into terminal-side and network-side models. For example, a model partitioning ratio of 1:9 means that the first 10% of the layers of the first model are partitioned into terminal-side models, which are used by the terminal for model inference, and the remaining 90% of the layers are partitioned into network-side models, which are used by the network (e.g., the first device or TEF network elements) for model inference. The model partitioning ratio is related to the terminal's capabilities (e.g., computing power, memory size); for example, the more powerful the terminal, the more layers of the terminal-side model are allocated to it. These multiple sub-model groups correspond to the same model partitioning ratio, but the compression ratio of the output layer matrix of the terminal-side model differs in different sub-model groups. With the same amount of input data, the higher the compression ratio of the output layer matrix of the terminal-side model, the less output data the terminal-side model has; conversely, the lower the compression ratio of the output layer matrix of the terminal-side model, the more output data the terminal-side model has. For example, taking model #1 as the first model, model #1 is divided into three sub-models according to the same model segmentation ratio and different model processing methods. The first sub-model includes terminal-side model #1_1 and network-side model #1_1, the second sub-model includes terminal-side model #1_2 and network-side model #1_2, and the third sub-model includes terminal-side model #1_3 and network-side model #1_3. Assuming that the compression ratio of the output layer matrix of terminal-side model #1 is the highest, the compression ratio of the output layer matrix of terminal-side model #2 is the second highest, and the compression ratio of the output layer matrix of terminal-side model #3 is the lowest, then with the same amount of input data, terminal-side model #1 has the least amount of output data, terminal-side model #2 has the second largest amount of output data, and terminal-side model #3 has the largest amount of output data. Furthermore, network-side models #1_1, #1_2, and #1_3 are also different from each other. Network-side model #1_1 corresponds to terminal-side model #1_1, and the input data of network-side model #1_1 is the output data of terminal-side model #1_1. Network-side model #1_2 corresponds to terminal-side model #1_2, and the input data of network-side model #1_2 is the output data of terminal-side model #1_2. Network-side model #1_3 corresponds to terminal-side model #1_3, and the input data of network-side model #1_3 is the output data of terminal-side model #1_3.

[0140] For example, step 302 specifically involves: the first device determining the estimated data volume of output data corresponding to each of the multiple terminal-side models of the first model based on the data volume of input data of the first model; then determining the estimated uplink transmission rate corresponding to each of the multiple terminal-side models based on the estimated data volume of output data of the multiple terminal-side models; next, determining a first uplink transmission rate based on the estimated uplink transmission rates corresponding to the multiple terminal-side models, wherein the first uplink transmission rate is the maximum estimated uplink transmission rate that can be satisfied among the estimated uplink transmission rates corresponding to the multiple terminal-side models, or the first uplink transmission rate is any one of the estimated uplink transmission rates that can be satisfied among the estimated uplink transmission rates corresponding to the multiple terminal-side models; finally, the first device determining a first terminal-side model based on the first uplink transmission rate, that is, determining the first terminal-side model corresponding to the first uplink transmission rate among the multiple terminal-side models. The first uplink transmission rate may be the uplink transmission rate that the first device will configure for the terminal.

[0141] Continuing with the previous example, the first model is model #1, and the multiple terminal-side models of model #1 include terminal-side model #1_1, terminal-side model #1_2, and terminal-side model #1_3. Based on the amount of input data for model #1, the first device determines the estimated data volume #1 of the output data corresponding to terminal-side model #1_1, the estimated data volume #2 of the output data corresponding to terminal-side model #1_2, and the estimated data volume #3 of the output data corresponding to terminal-side model #1_3. Then, it determines the estimated uplink transmission rate #1 corresponding to the estimated data volume #1, the estimated uplink transmission rate #2 corresponding to the estimated data volume #2, and the estimated uplink transmission rate #3 corresponding to the estimated data volume #3, respectively. Assuming the estimated uplink transmission rate #1 is the highest, the estimated uplink transmission rate #2 is the second highest, and the estimated uplink transmission rate #3 is the lowest, if the terminal can satisfy both estimated uplink transmission rates #2 and #3 during uplink transmission, then the first device can choose the highest achievable estimated uplink transmission rate (i.e., estimated uplink transmission rate #2), or any one of the achievable estimated uplink transmission rates (i.e., estimated uplink transmission rate #2 or estimated uplink transmission rate #3). The estimated uplink transmission rate chosen by the first device is called the first uplink transmission rate. After determining the first uplink transmission rate, the first device determines the first terminal-side model corresponding to the first uplink transmission rate among multiple terminal-side models.

[0142] In one implementation method, when determining the first uplink transmission rate based on the estimated uplink transmission rates corresponding to multiple terminal-side models, the first device may refer to one or more of the following: the terminal-side estimated inference time corresponding to the amount of input data, the identifier of the first model, the channel quality between the terminal and the first device, or the load of the first device. In other words, the first device determines the first uplink transmission rate based on the estimated uplink transmission rates corresponding to the multiple terminal-side models, as well as the identifier of the first model, the channel quality between the terminal and the first device, or the load of the first device. Based on this method, the first uplink transmission rate can be accurately determined.

[0143] In one implementation method, a first device is locally configured with first association information or receives first association information from another device (e.g., a second device), and after determining a first uplink transmission rate, determines a first terminal-side model based on the first uplink transmission rate and the first association information. The first association information indicates a one-to-one correspondence between multiple uplink transmission rate ranges and multiple terminal-side models of the first model. These multiple uplink transmission rate ranges include the first uplink transmission rate range, and the first uplink transmission rate is contained within the first uplink transmission rate range, which corresponds to the first terminal-side model. For example, continuing with the previous example, the terminal-side model of model #1 includes terminal-side model #1, terminal-side model #2, and terminal-side model #3. The first association information is used to indicate the following correspondence: uplink transmission rate range #1 corresponds to terminal-side model #1, uplink transmission rate range #2 corresponds to terminal-side model #2, and uplink transmission rate range #3 corresponds to terminal-side model #3. Assuming that the first uplink transmission rate is included in the uplink transmission rate range #3 (i.e., the first uplink transmission rate range), the first device determines that the uplink transmission rate range #3 corresponds to terminal-side model #3 (i.e., the first terminal-side model).

[0144] Step 303: The first device sends an instruction message to the terminal. Accordingly, the terminal receives the instruction message.

[0145] This indication information is used to indicate the first terminal-side model of the first model. That is, the first device indicates to the terminal through the indication information that the terminal-side model selected for the first model in model inference is the first terminal-side model.

[0146] In one implementation method, the indication information includes the identifier of the first model and the identifier of the first terminal-side model.

[0147] In another implementation, the indication information includes the identifier of the first model and the first uplink transmission rate of the terminal, which corresponds to the first terminal-side model. For example, the first uplink transmission rate may be the first uplink transmission rate determined by the first device based on the amount of input data for the first model in step 302. After receiving the indication information, the terminal can determine the first terminal-side model based on the first uplink transmission rate and the first association information corresponding to the first model. The meaning of the first association information can be found in the description in step 302. The terminal may have the first association information configured locally or may have received it in advance from the first device or the second device. Specifically, the terminal determines the first terminal-side model based on the first uplink transmission rate and the first association information by: determining a first uplink transmission rate range that includes the first uplink transmission rate based on the first association information, and determining the first terminal-side model corresponding to the first uplink transmission rate range.

[0148] In another implementation method, the indication information includes the identifier of the first model and the first coordination level. The first coordination level corresponds one-to-one with the first terminal-side model and the first network-side model. It can be understood that the first coordination level can be used to uniquely index the first terminal-side model and the first network-side model.

[0149] For example, the first device is locally configured with second association information or receives second association information in advance from other devices (e.g., the second device). Based on the second association information corresponding to the first model, the first device determines the first coordination level corresponding to the first terminal-side model and sends indication information (i.e., the identifier of the first model and the first coordination level) to the terminal. The second association information is used to indicate a one-to-one correspondence between multiple coordination levels and multiple terminal-side models of the first model, and these multiple coordination levels include the first coordination level.

[0150] Continuing with the previous example, suppose model #1 is divided into three sub-models according to different processing methods. The first sub-model group includes terminal-side model #1_1 and network-side model #1_1; the second sub-model group includes terminal-side model #1_2 and network-side model #1_2; and the third sub-model group includes terminal-side model #1_3 and network-side model #1_3. For example, terminal-side model #1_1 and network-side model #1_1 correspond to collaboration level #1, terminal-side model #1_2 and network-side model #1_2 correspond to collaboration level #2, and terminal-side model #1_3 and network-side model #1_3 correspond to collaboration level #3. Furthermore, the second association information corresponding to model #1 indicates that collaboration level #1 corresponds to terminal-side model #1_1, collaboration level #2 corresponds to terminal-side model #1_2, and collaboration level #3 corresponds to terminal-side model #1_3. Assuming the first terminal-side model determined by the first device is terminal-side model #1_3, then the indication information sent by the first device to the terminal according to the second association information includes the identifier of model #1 and the collaboration level #3. After receiving the instruction, the terminal can determine the terminal-side model #1_3 (i.e., the first terminal-side model) based on the collaboration level #3 and the second association information. The terminal may have the second association information configured locally or may have received the second association information in advance from the first or second device.

[0151] Based on the above scheme, when a terminal has inference needs, it requests a terminal-side model for inference from a first device. The first terminal-side model selected by the first device for inference is one of multiple terminal-side models of the first model, and the first terminal-side model is related to the amount of input data when the terminal performs model inference. In other words, this method selects a suitable terminal-side model for model inference based on the amount of input data for model inference. This method ensures that the air interface transmission latency when the terminal transmits the inference results over the air interface does not vary excessively, thus guaranteeing end-to-end service latency in end-to-end collaborative inference scenarios. For example, when the air interface quality is relatively stable or the load of the first device is relatively stable, when the amount of input data for model inference is large, a terminal-side model with a higher compression ratio of the output layer matrix can be selected for model inference; when the amount of input data for model inference is small, a terminal-side model with a lower compression ratio of the output layer matrix can be selected. This ensures that the amount of data in the inference results generated by the terminal during different inference processes does not vary too much, thereby guaranteeing that the air interface transmission latency for transmitting the inference results over the air interface does not vary excessively. For example, when the amount of input data is relatively stable, if the air interface quality is poor or the load on the first device is heavy, the terminal-side model with a higher compression ratio of the output layer matrix can be selected for model inference. If the air interface quality is good or the load on the first device is light, the terminal-side model with a lower compression ratio of the output layer matrix can be selected for model inference. This allows the amount of data generated by the inference results when the terminal performs different inferences to change accordingly with the air interface quality or the load on the first device, thereby ensuring that the air interface transmission latency of the inference results does not change too much.

[0152] Figure 4 is a flowchart illustrating a communication method provided in an embodiment of this application. The method includes the following steps:

[0153] Step 401: The terminal determines its first uplink transmission rate.

[0154] That is, the terminal can measure its own uplink transmission rate, and the measured uplink transmission rate is the first uplink transmission rate. The first uplink transmission rate can be the terminal's current uplink transmission rate or the maximum uplink transmission rate supported by the terminal.

[0155] Step 402: The terminal determines the first terminal-side model of the first model based on the first uplink transmission rate.

[0156] Regarding the relationship between the first model and the first terminal-side model, please refer to the relevant description in the embodiment of Figure 3 above.

[0157] The first terminal-side model is the terminal-side model selected by the terminal from multiple terminal-side models of the first model for model inference.

[0158] In one implementation method, the terminal may be locally configured with first association information or may receive first association information in advance from a first device or a second device. The meaning of this first association information can be found in the relevant description in the embodiment of Figure 3 above. The terminal can determine a first terminal-side model of a first model based on the first association information and a first uplink transmission rate. That is, the terminal determines a first uplink transmission rate range containing the first uplink transmission rate based on the first association information, and then determines a first terminal-side model of the first model corresponding to the first uplink transmission rate range. For example, the first model is model #1, and the terminal-side model of model #1 includes terminal-side model #1, terminal-side model #2, and terminal-side model #3. The first association information is used to indicate the following correspondence: uplink transmission rate range #1 corresponds to terminal-side model #1, uplink transmission rate range #2 corresponds to terminal-side model #2, and uplink transmission rate range #3 corresponds to terminal-side model #3. Assuming the first uplink transmission rate is included in uplink transmission rate range #3, the terminal determines that the terminal-side model #3 (i.e., the first terminal-side model) corresponds to uplink transmission rate range #3.

[0159] Based on the above scheme, when a terminal has inference requirements, a first terminal-side model for inference is selected from multiple terminal-side models of the first model based on the terminal's first uplink transmission rate. In other words, this method selects a suitable terminal-side model for model inference based on the terminal's first uplink transmission rate. This method ensures that the air interface transmission latency of the terminal when transmitting inference results over the air interface does not vary excessively, thus guaranteeing end-to-end service latency in terminal-network collaborative inference scenarios. For example, when the terminal's first uplink transmission rate is low, a terminal-side model with a higher compression ratio of the output layer matrix can be selected for model inference; when the terminal's first uplink transmission rate is high, a terminal-side model with a lower compression ratio of the output layer matrix can be selected for model inference. This ensures that the amount of data generated by the inference results varies with the uplink transmission rate, thereby guaranteeing that the air interface transmission latency of the inference results does not vary excessively.

[0160] The following section, with reference to Figure 5, describes the implementation method of model inference through the cooperation between the terminal and the network (i.e., terminal-network collaboration).

[0161] Figure 5 is a flowchart illustrating a communication method provided in an embodiment of this application. The method includes the following steps:

[0162] Step 501: The terminal determines the first inference result based on the input data of the first model and the first terminal-side model of the first model.

[0163] For example, the terminal can determine to perform model inference using the first terminal-side model of the first model based on the method of the embodiment of FIG3 above, that is, execute the embodiment of FIG5 after the embodiment of FIG3 above. Alternatively, the terminal can also determine to perform model inference using the first terminal-side model of the first model based on the method of the embodiment of FIG4 above, that is, execute the embodiment of FIG5 after the embodiment of FIG4 above.

[0164] Specifically, the terminal inputs the input data of the first model into the first terminal-side model and obtains the output result, that is, determines the first inference result, which can also be called the intermediate result or intermediate inference result.

[0165] Step 502: The terminal sends an inference request. Correspondingly, the first device or TEF network element receives the inference request.

[0166] In one implementation method, the terminal sends an inference request to a first device, and the first device receives the inference request. Optionally, after receiving the inference request, the first device obtains information about the TEF network element (e.g., the address information of the TEF network element) and then forwards the inference request to the TEF network element.

[0167] The inference request includes a first inference result and first information, and is used to request continued inference based on the first inference result. Specifically, the inference can be continued by a first device based on the first inference result, or by a TEF network element based on the first inference result.

[0168] The first information is used to determine the first network-side model of the first model. This first network-side model is one of multiple network-side models of the first model. Each of these multiple network-side models is a sub-model of the first model used for network inference after processing. This first network-side model corresponds to the first terminal-side model. That is, the first model obtains the first terminal-side model and the first network-side model after a certain model processing.

[0169] In one implementation method, the first information includes an identifier of a first model (e.g., an application identifier or a service identifier) ​​and an identifier of a first terminal-side model. A first device or TEF network element determines a first network-side model based on the identifier of the first model and the identifier of the first terminal-side model.

[0170] In another implementation, the first information includes an identifier of the first model and a first uplink transmission rate, where the first uplink transmission rate corresponds to the first network-side model. For example, upon receiving this first information, a first device or TEF network element can determine the first network-side model of the first model based on third association information and the first uplink transmission rate. Specifically, the first device or TEF network element may have locally configured third association information, or a second device may receive third association information. This third association information indicates a one-to-one correspondence between multiple uplink transmission rate ranges and multiple network-side models of the first model. These multiple uplink transmission rate ranges include the first uplink transmission rate range, and the first uplink transmission rate is contained within the first uplink transmission rate range. The first uplink transmission rate range corresponds to the first network-side model. In detail, the first device or TEF network element determines the first uplink transmission rate range containing the first uplink transmission rate based on the third association information, and then determines the first network-side model corresponding to the first uplink transmission rate range.

[0171] In another implementation, the first information includes an identifier of a first model and a first coordination level, where the first coordination level corresponds to a first terminal-side model and a first network-side model. Exemplarily, the terminal is locally configured with second association information or receives second association information from other devices (e.g., the first device or the second device). The terminal determines the first information, which includes the identifier of the first model and the first coordination level, based on the second association information and the first terminal-side model. The terminal then carries this first information in the sent inference request. The second association information indicates a one-to-one correspondence between multiple coordination levels and multiple terminal-side models, where the multiple coordination levels include the first coordination level, which corresponds to the first terminal-side model. Exemplarily, the first device or TEF network element receives the first information and determines the first network-side model of the first model based on fourth association information and the first coordination level. The first device or TEF network element is locally configured with fourth association information or receives fourth association information from other devices (e.g., the second device). This fourth association information indicates a one-to-one correspondence between multiple coordination levels and multiple network-side models, where the multiple coordination levels include the first coordination level, which corresponds to the first network-side model.

[0172] In one implementation method, if a second device performs model processing (e.g., model segmentation) on a first model to obtain multiple terminal-side models and multiple network-side models of the first model, the second device sends information about the multiple terminal-side models of the first model to the terminal. The information of each terminal-side model includes the identifier of the terminal-side model and the parameters of the terminal-side model (e.g., at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix). The second device also sends information about the multiple network-side models of the first model to the first device. The information of each network-side model includes the identifier of the network-side model and the parameters of the network-side model (e.g., at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix).

[0173] In another implementation, if the first device processes the first model (e.g., model segmentation) to obtain multiple terminal-side models and multiple network-side models of the first model, then the first device sends information about the multiple terminal-side models of the first model to the terminal. The information for each terminal-side model includes its identifier and parameters (e.g., at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix). The first device also locally stores information about the multiple network-side models of the first model. The information for each network-side model includes its identifier and parameters (e.g., at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix). Subsequent network-side model inference is performed by the first device.

[0174] In another implementation, if the TEF network element performs model processing (e.g., model segmentation) on the first model to obtain multiple terminal-side models and multiple network-side models of the first model, then the TEF network element sends information about the multiple terminal-side models of the first model to the terminal. The information for each terminal-side model includes its identifier and parameters (e.g., at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix). The TEF network element also locally stores information about the multiple network-side models of the first model. The information for each network-side model includes its identifier and parameters (e.g., at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix). Subsequent network-side model inference is performed by this TEF network element.

[0175] Step 503: The first device or TEF network element sends the second inference result to the terminal. Accordingly, the terminal receives the second inference result.

[0176] The second inference result is obtained based on the first inference result and the first network-side model. Specifically, the first device or TEF network element inputs the first inference result into the first network-side model to obtain the second inference result.

[0177] Based on the above scheme, the first model includes multiple terminal-side models and multiple network-side models. The terminal uses the first terminal-side model among the multiple terminal-side models to obtain a first inference result. The first device or TEF network element continues inference based on the first inference result and the first network-side model corresponding to the first terminal-side model to obtain a second inference result. This method can select appropriate terminal-side models and network-side models from multiple terminal-side models and multiple network-side models for inference, and can ensure that the air interface transmission latency when the terminal transmits the inference result over the air interface does not change too much. Therefore, it can guarantee the end-to-end service latency in the terminal-network collaborative inference scenario.

[0178] Regarding the embodiment in Figure 5 above, in one possible implementation method, if the terminal performs model inference on the terminal side and the TEF network element performs model inference on the network side, and the terminal and the network perform inference only once, then the terminal and the network may not establish a user plane session, and the first device selects the TEF network element for performing model inference on the network side.

[0179] Regarding the embodiment in Figure 5 above, in another possible implementation, if the terminal performs model inference on the terminal side and the TEF network element performs model inference on the network side, and the terminal and network need to perform multiple inference operations, then the terminal and network can establish a user plane session. During the establishment of the user plane session, a TEF network element for network-side model inference is selected. Subsequently, each time the first device receives an inference request from the terminal, it forwards the inference request to that TEF network element, which then continues the inference process. That is, when the terminal and network need to perform multiple inference operations, by establishing a user plane session, only one TEF network element selection operation needs to be performed, and it is not necessary to reselect the TEF network element each time inference occurs, thus saving signaling overhead. For example, the process of establishing a user plane session is as follows: The first device receives a first session establishment request from the terminal. The first session establishment request includes the identifier of the first model. The first session establishment request is used to request the establishment of a session for inference of the first model. The first device obtains information of the TEF network element (such as address information or identifier). Then, the first device sends a second session establishment request to the TEF network element. The second session establishment request includes the identifier of the first model. The second session establishment request is used to request the establishment of a session for inference of the first model.

[0180] Figure 6 is a flowchart illustrating a communication method provided in an embodiment of this application. The method includes the following steps:

[0181] Step 601: The second device or TEF network element obtains the model processing strategy of the first model.

[0182] In one implementation method, the second device or TEF network element is locally configured with the model processing strategy of the first model.

[0183] In another implementation method, the second device or TEF network element receives the model processing strategy of the first model from an external device (such as a third-party server).

[0184] The model processing strategy of the first model is used to indicate the model processing method of the first model. This model processing method indicates the model segmentation method (e.g., segmentation based on a segmentation ratio) and the compression ratio of the output layer matrix of the segmented terminal-side model. For example, the model processing strategy includes multiple model processing methods, each of which includes a model segmentation ratio and the compression ratio of the output layer matrix of the segmented terminal-side model.

[0185] As a specific example, the model processing strategy includes model processing method #1, model processing method #2, and model processing method #3. Model processing method #1 includes a model segmentation ratio of 1:9 and a compression rate of 90% for the output layer matrix of the terminal-side model; model processing method #2 includes a model segmentation ratio of 1:9 and a compression rate of 80% for the output layer matrix of the terminal-side model; and model processing method #3 includes a model segmentation ratio of 1:9 and a compression rate of 65% for the output layer matrix of the terminal-side model. This example uses the same model segmentation ratio across all processing methods; however, these ratios can differ in actual applications.

[0186] Step 602: The second device or TEF network element determines multiple terminal-side models and multiple network-side models of the first model according to the model processing strategy.

[0187] Regarding the relationship between the first model and multiple terminal-side models and multiple network-side models, please refer to the relevant description in the embodiment of Figure 3 above.

[0188] In one implementation method, a model processing strategy is used to indicate the model processing methods corresponding to multiple uplink transmission rate ranges. A second device or TEF network element, based on the model processing strategy and the model processing method corresponding to each uplink transmission rate range, obtains a set of sub-models for the first model. This set of sub-models includes a terminal-side model and a network-side model, and the terminal-side model corresponds to the compression ratio of an output layer matrix. For example, the model processing strategy includes model processing method #1 corresponding to uplink transmission rate range #1, model processing method #2 corresponding to uplink transmission rate range #2, and model processing method #3 corresponding to uplink transmission rate range #3. The second device or TEF network element determines the terminal-side model 1_1 and network-side model 1_1 of model #1 based on model processing method #1, determines the terminal-side model 1_2 and network-side model 1_2 of model #1 based on model processing method #2, and determines the terminal-side model 1_3 and network-side model 1_3 of model #1 based on model processing method #3. Further, the second device or TEF network element determines the first association information and the third association information. The first association information is used to indicate the one-to-one correspondence between multiple uplink transmission rate ranges and multiple terminal-side models. For example, the first association information includes the correspondence between uplink transmission rate range #1 and terminal-side model 1_1, the correspondence between uplink transmission rate range #2 and terminal-side model 1_2, and the correspondence between uplink transmission rate range #3 and terminal-side model 1_3. The third association information is used to indicate the one-to-one correspondence between multiple uplink transmission rate ranges and multiple network-side models. For example, the first association information includes the correspondence between uplink transmission rate range #1 and network-side model 1_1, the correspondence between uplink transmission rate range #2 and network-side model 1_2, and the correspondence between uplink transmission rate range #3 and network-side model 1_3. For specific usage of the first and third association information, please refer to the embodiments shown in Figures 3 to 5 above.

[0189] For example, if the first association information and the third association information are determined by the second device according to the model processing strategy, and the model inference operation on the network side is performed by the TEF network element, then after determining the first association information and the third association information, the second device sends the first association information to the terminal and the third association information to the TEF network element. If the first association information and the third association information are determined by the second device according to the model processing strategy, and the model inference operation on the network side is performed by the first device, then after determining the first association information and the third association information, the second device sends the first association information to the terminal and the third association information to the first device. If the first association information and the third association information are determined by the TEF network element according to the model processing strategy, and the model inference operation on the network side is performed by the first device, then after determining the first association information and the third association information, the TEF network element sends the first association information to the terminal and the third association information to the first device. If the first association information and the third association information are determined by the TEF network element according to the model processing strategy, and the model inference operation on the network side is executed by the TEF network element, then after determining the first association information and the third association information, the TEF network element sends the first association information to the terminal and saves the third association information locally.

[0190] In another implementation, the model processing strategy is used to indicate the model processing methods corresponding to multiple coordination levels. The second device or TEF network element, based on the model processing strategy and the model processing method corresponding to each coordination level, obtains a set of sub-models for the first model. This set of sub-models includes a terminal-side model and a network-side model, and the terminal-side model corresponds to the compression ratio of an output layer matrix. For example, the model processing strategy includes model processing method #1 corresponding to coordination level #1, model processing method #2 corresponding to coordination level #2, and model processing method #3 corresponding to coordination level #3. The second device or TEF network element determines the terminal-side model 1_1 and network-side model 1_1 of model #1 based on model processing method #1, determines the terminal-side model 1_2 and network-side model 1_2 of model #1 based on model processing method #2, and determines the terminal-side model 1_3 and network-side model 1_3 of model #1 based on model processing method #3. Further, the second device or TEF network element determines the second association information and the fourth association information. The second association information is used to indicate the one-to-one correspondence between multiple coordination levels and multiple terminal-side models. For example, the second association information includes the correspondence between coordination level #1 and terminal-side model 1_1, coordination level #2 and terminal-side model 1_2, and coordination level #3 and terminal-side model 1_3. The fourth association information is used to indicate the one-to-one correspondence between multiple coordination levels and multiple network-side models. For example, the second association information includes the correspondence between coordination level #1 and network-side model 1_1, coordination level #2 and network-side model 1_2, and coordination level #3 and network-side model 1_3. For specific usage of the second and fourth association information, please refer to the embodiments shown in Figures 3 to 5 above.

[0191] For example, if the second device determines the second and fourth association information according to the model processing strategy, and the network-side model inference operation is performed by the TEF network element, then after determining the second and fourth association information, the second device sends the second association information to the terminal and the fourth association information to the TEF network element. If the second device determines the second and fourth association information according to the model processing strategy, and the network-side model inference operation is performed by the first device, then after determining the second and fourth association information, the second device sends the second association information to the terminal and the fourth association information to the first device. If the TEF network element determines the second and fourth association information according to the model processing strategy, and the network-side model inference operation is performed by the first device, then after determining the second and fourth association information, the TEF network element sends the second association information to the terminal and the fourth association information to the first device. If the second and fourth association information are determined by the TEF network element according to the model processing strategy, and the model inference operation on the network side is executed by the TEF network element, then after determining the second and fourth association information, the TEF network element sends the second association information to the terminal and saves the fourth association information locally.

[0192] Based on the above scheme, the first model includes multiple terminal-side models and multiple network-side models, which can select appropriate terminal-side models and network-side models for inference. This can ensure that the air interface transmission latency of the terminal when transmitting the inference results over the air interface does not change too much, thus ensuring the end-to-end service latency in the terminal-network collaborative inference scenario.

[0193] The embodiments of Figures 3 to 6 described above will now be described in detail with reference to the accompanying drawings. The embodiment shown in Figure 7 is a specific example of the embodiments of Figures 3 to 6 described above. In the following embodiments, model inference is performed by the terminal on the terminal side and by the TEF network element on the network side.

[0194] Figure 7 is a flowchart illustrating a communication method provided in an embodiment of this application. The method includes the following steps:

[0195] Step 701: The second device sends a first message to the TEF network element. Correspondingly, the TEF network element receives the first message.

[0196] The first message includes information about the first model and the model processing strategy for the first model.

[0197] The information of the first model includes an identifier for the first model, which may be an application identifier or a business identifier, etc. Optionally, the information of the first model also includes model parameters of the first model, which include at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, or edge weights.

[0198] This model processing strategy is used to indicate the model processing method corresponding to multiple uplink transmission rate ranges. For example, uplink transmission rate range #1 corresponds to model processing method #1, uplink transmission rate range #2 corresponds to model processing method #2, and uplink transmission rate range #3 corresponds to model processing method #3.

[0199] Alternatively, the model processing strategy can be used to indicate the model processing method corresponding to multiple collaboration levels. For example, collaboration level #1 corresponds to model processing method #1, collaboration level #2 corresponds to model processing method #2, and collaboration level #3 corresponds to model processing method #3.

[0200] This application does not limit the specific implementation of the first message; the first message can be either a user plane message or a control plane message. For example, the first message can be an interface message between the second device and the TEF network element, or it can be other types of messages.

[0201] For example, the second device can send the first message to the TEF network element through the interface message between the second device and the TEF network element, or it can send the first message to the TEF network element through the routing of one or more intermediate devices.

[0202] Step 702: The TEF network element processes the first model according to the model processing strategy of the first model to obtain multiple terminal-side models and multiple network-side models of the first model.

[0203] For details on the specific implementation of step 702, please refer to the description in the embodiment shown in Figure 6 above.

[0204] Step 703: The TEF network element sends information about multiple terminal-side models of the first model to the terminal. Correspondingly, the terminal receives the information about multiple terminal-side models of the first model.

[0205] For example, the TEF network element can send information about multiple terminal-side models of the first model to the OTT server of the terminal, and then the OTT server sends the information about multiple terminal-side models of the first model to the terminal.

[0206] The information of the multiple terminal-side models in the first model includes the identifiers of the multiple terminal-side models and the model parameters of the multiple terminal-side models. For example, the model parameters include at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix.

[0207] Based on the aforementioned example, the TEF network element sends information about the terminal-side model #1_1, the terminal-side model #1_2, and the terminal-side model #1_3 of the first model to the terminal.

[0208] It should be noted that when the TEF network element sends the terminal-side model information of the first model to the terminal, it also needs to send the identifier of the first model.

[0209] The TEF network element also deploys multiple network-side models of the processed first model locally. For example, combining the previous example, the TEF network element deploys network-side models #1_1, #1_2, and #1_3 of the first model locally.

[0210] In steps 701 to 703 above, the TEF network element performs model processing and deploys the processed terminal-side model to the terminal. In another implementation method, a second device can also perform model processing and deploy the processed terminal-side model to the terminal. When the second device performs model processing and deploys the processed terminal-side model to the terminal, steps 701 to 703 above can be replaced by steps A to C.

[0211] Step A: The second device processes the first model according to the model processing strategy of the first model to obtain multiple terminal-side models and multiple network-side models of the first model.

[0212] For specific methods and examples of model processing, please refer to the description in the embodiment shown in Figure 6 above.

[0213] In step B, the second device sends information about multiple terminal-side models to the terminal. Correspondingly, the terminal receives the information about the multiple terminal-side models.

[0214] For example, the second device can send information about multiple terminal-side models of the first model to the terminal's OTT server, and then the OTT server sends the information about the multiple terminal-side models of the first model to the terminal.

[0215] The information of the multiple terminal-side models in the first model includes the identifiers of the multiple terminal-side models and the model parameters of the multiple terminal-side models. For example, the model parameters include at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix.

[0216] Based on the aforementioned example, the second device sends information about the terminal-side model #1_1, the terminal-side model #1_2, and the terminal-side model #1_3 of the first model to the terminal.

[0217] It should be noted that when the second device sends the terminal-side model information of the first model to the terminal, it also needs to send the identifier of the first model.

[0218] In step C, the second device sends information about multiple network-side models to the TEF network element. Correspondingly, the TEF network element receives information about multiple network-side models.

[0219] The information of the multiple network-side models of the first model includes the identifiers of the multiple network-side models and the model parameters of the multiple network-side models. For example, the model parameters include at least one of the following: number of model layers, number of nodes, connections between nodes, node weights, edge weights, or compression ratio of the output layer matrix.

[0220] Based on the aforementioned example, the second device sends information about the network side model #1_1, network side model #1_2, and network side model #1_3 of the first model to the TEF network element.

[0221] It should be noted that when the second device sends the network-side model information of the first model to the TEF network element, it also needs to send the identifier of the first model.

[0222] The order of steps B and C above is not limited.

[0223] In step 704, the third device of the TEF network element sends a second message. Correspondingly, the third device receives the second message.

[0224] The third device is either a TMF network element or a TCF network element.

[0225] The second message includes the identifier of the TEF network element, the identifier of the first model, and configuration information.

[0226] Wherein, when the model processing strategy of the first model is used to indicate the model processing method corresponding to multiple uplink transmission rate ranges respectively, the configuration information includes the first association information and the third association information.

[0227] When the model processing strategy of the first model is used to indicate the model processing methods corresponding to multiple collaboration levels, the configuration information includes the second association information and the fourth association information.

[0228] This application does not limit the specific implementation of the second message; the second message can be either a user plane message or a control plane message. For example, the second message can be an interface message between a third device and a TEF network element, or it can be other types of messages.

[0229] For example, the TEF network element can send the second message to the third device through the interface message between the third device and the TEF network element, or it can send the second message to the third device through the routing of one or more intermediate devices.

[0230] In step 705, the third device sends a third message to the first device. Accordingly, the first device receives the third message.

[0231] The third message includes the identifier and configuration information of the first model.

[0232] This application does not limit the specific implementation of the third message; the third message can be either a user plane message or a control plane message. For example, the third message can be an interface message between a third device and a first device, or it can be other types of messages.

[0233] For example, a third device can send a third message to a first device through an interface message between the third device and the first device, or it can send the third message to the first device through routing of one or more intermediate devices.

[0234] In step 706, the first device sends a fourth message to the terminal. Correspondingly, the terminal receives the fourth message.

[0235] The fourth message includes the identifier and configuration information of the first model.

[0236] This application does not limit the specific implementation of the fourth message; the fourth message can be either a user plane message or a control plane message. For example, the fourth message can be an interface message between the first device and the terminal, or it can be other types of messages.

[0237] For example, the first device can send the fourth message to the terminal through the interface message between the first device and the terminal, or it can send the third message to the terminal through the routing of one or more intermediate devices.

[0238] For example, when the first device is an access network device, the fourth message can be an air interface message between the terminal and the access network device, such as a radio resource control (RRC) message, a medium access control element (MAC CE) message, etc.

[0239] For example, when the first device is an AMF network element, the fourth message can be a message between the terminal and the AMF network element, such as a non-access stratum (NAS) message.

[0240] In step 707, the terminal sends a first request to the first device. Accordingly, the first device receives the first request.

[0241] For example, the first request may also be referred to as a pre-scheduled request, a model request, etc.

[0242] In one implementation method, the first request includes an identifier of a first model and model selection assistance information. The first request is used to request a terminal-side model of the first model corresponding to the model selection assistance information. The model selection assistance information includes the data volume of the input data of the first model, which indicates the size of the input data used for model inference. Optionally, the model selection assistance information also includes the terminal-side estimated inference time corresponding to the data volume of the input data. That is, the terminal-side estimated inference time represents the estimated time for the terminal to use the input data to perform model inference and obtain the inference result.

[0243] In another implementation, the first request includes an identifier for the first model and the terminal's current location information. The terminal's current location information indicates the terminal's current location.

[0244] For example, the identifier of the first model can be an application identifier (APP ID) or a service identifier (service ID). That is, one application or one service corresponds to one model.

[0245] For example, when the first device is an access network device, the first request can be an air interface message between the terminal and the access network device, such as an RRC message.

[0246] For example, when the first device is an AMF network element, the first request can be a message between the terminal and the AMF network element, such as a NAS message.

[0247] This application does not limit the specific implementation of the first request.

[0248] Step 708: The first device determines the instruction information.

[0249] In one scenario, when the first request includes the identifier of the first model and model selection assistance information, the first device determines the specific implementation method of the instruction information, which can be referred to the description of steps 302 and 303 in the embodiment of FIG3 above.

[0250] In another scenario, if the first request includes the identifier of the first model and the current location information of the terminal, the first device can determine the first coordination level corresponding to the current location information of the terminal based on the current location information of the terminal and the correspondence between the location information of at least one terminal within the most recent set time period and the coordination level. Then, it determines indication information based on the first coordination level, which includes the first coordination level and the identifier of the first model. For example, if the first request includes the identifier of the first model and the current location information of terminal #1, the first device determines the location information of the terminal closest to the current location information of terminal #1 (assuming it is terminal #2) from the correspondence between the location information of at least one terminal within the most recent set time period and the coordination level, and determines the coordination level corresponding to the location information of terminal #2 as the first coordination level of terminal #1. As another example, if the first request includes the identifier of the first model and the current location information of terminal #1, the first device defines a circular range with the location indicated by the location information of terminal #1 as the center and a preset length as the radius, and determines the average, maximum, or minimum value of the coordination level corresponding to the location information of each terminal within this circular range, and determines the first coordination level of terminal #1 based on this average, maximum, or minimum value.

[0251] In step 709, the first device sends a first response to the terminal. Accordingly, the terminal receives the first response.

[0252] The first response includes instructions.

[0253] For example, when the first device is an access network device, the first response can be an air interface message between the terminal and the access network device, such as an RRC message, MAC CE, etc.

[0254] For example, when the first device is an AMF network element, the first response can be a message between the terminal and the AMF network element, such as a NAS message. This application does not limit the specific implementation of the first response.

[0255] In step 710, the terminal sends an inference request to the first device. Accordingly, the first device receives the inference request.

[0256] The reasoning request includes first information and first reasoning result.

[0257] The first information includes the identifier of the first model and the identifier of the first terminal-side model, or the first information includes the identifier of the first model and the first uplink transmission rate, or the first information includes the identifier of the first model and the first coordination level.

[0258] For a detailed explanation of the implementation of step 710, please refer to step 502 of the embodiment in Figure 5.

[0259] Step 711: The first device and the third device send an inference request. Accordingly, the third device receives the inference request.

[0260] The reasoning request includes first information and first reasoning result.

[0261] In step 712, the third device sends an inference request to the TEF network element. Correspondingly, the TEF network element receives the inference request.

[0262] The reasoning request includes first information and first reasoning result.

[0263] Step 713: The TEF network element sends an inference response to the terminal. Correspondingly, the terminal receives the inference response.

[0264] The reasoning response includes the identifier of the first model and the result of the second reasoning.

[0265] TEF network elements can send inference responses directly to the terminal, or they can send inference responses to the terminal via a third device or a first device.

[0266] After receiving the inference request, the TEF network element determines the first network-side model of the first model based on the first information, and inputs the first inference result into the first network-side model to continue inference and obtain the second inference result.

[0267] As one implementation method, if multiple end-to-end collaborative inference operations are required between the terminal and the TEF network element, then after step 706 and before step 707, the terminal can initiate a first session establishment request to the first device to request the establishment of a session. This first session establishment request carries the identifier of the first model and is used to request the establishment of a session for inference of the first model. Then, the first device sends a second session establishment request to the TEF network element via a third device. This second session establishment request includes the identifier of the first model and is used to request the establishment of a session for inference of the first model, thereby completing the session establishment between the terminal and the TEF network element. When the terminal subsequently sends an inference request, since a session has already been established, the terminal can directly find the TEF network element and send the inference request to it. That is, steps 710 to 712 can be replaced by the following step: the terminal sends an inference request to the TEF network element. Furthermore, the subsequent steps starting from step 707 can be executed multiple times to achieve multiple end-to-end collaborative inference operations.

[0268] Based on the above scheme, the first model includes multiple terminal-side models and multiple network-side models. The terminal uses the first terminal-side model among the multiple terminal-side models to obtain a first inference result. The TEF network element continues inference based on the first inference result and the first network-side model corresponding to the first terminal-side model to obtain a second inference result. This method can select appropriate terminal-side models and network-side models from multiple terminal-side models and multiple network-side models for inference, and can ensure that the air interface transmission latency when the terminal transmits the inference result over the air interface does not change too much. Therefore, it can guarantee the end-to-end service latency in the terminal-network collaborative inference scenario.

[0269] Figure 8 illustrates a possible exemplary block diagram of the communication device involved in the embodiments of this application. As shown in Figure 8, the communication device 800 may include modules or units for implementing the methods described above. In one possible design, the communication device 800 includes a processing unit 802 and a communication unit 803. Optionally, the communication device 800 may further include a storage unit 801 for storing device program code and / or data.

[0270] The communication device 800 can be a terminal device-side device in the above embodiments, such as a terminal device or a communication module in a terminal device, or a circuit or chip in a terminal device that is responsible for communication functions.

[0271] For example, in one embodiment, the communication unit 803 is configured to send a first request to a first device, the first request including an identifier of a first model and model selection assistance information, the first request being used to request a terminal-side model of the first model corresponding to the model selection assistance information, the model selection assistance information including the amount of input data of the first model; and to receive indication information from the first device, the indication information being used to indicate a first terminal-side model of the first model, the first terminal-side model being one of a plurality of terminal-side models of the first model, each of the plurality of terminal-side models being a sub-model of the first model processed for inference by the terminal; the processing unit 802 is configured to, according to the The communication unit 803 is further configured to: receive the input data and the first terminal-side model to determine a first inference result; send an inference request, the inference request including the first inference result and first information, the inference request being used to request continued inference based on the first inference result, the first information being used to determine a first network-side model of the first model, the first network-side model being one of a plurality of network-side models of the first model, each of the plurality of network-side models being a sub-model of the first model processed for network inference, the first network-side model corresponding to the first terminal-side model; and receive a second inference result, the second inference result being obtained based on the first inference result and the first network-side model.

[0272] In one possible implementation, the model selection auxiliary information further includes the terminal-side estimated inference time corresponding to the amount of input data.

[0273] In one possible implementation, the indication information includes the identifier of the first model and the identifier of the first terminal-side model.

[0274] In one possible implementation, the indication information includes the identifier of the first model and the first uplink transmission rate of the terminal, wherein the first uplink transmission rate corresponds to the first terminal-side model.

[0275] In one possible implementation, the communication unit 803 is further configured to receive first association information, which indicates a one-to-one correspondence between multiple uplink transmission rate ranges and multiple terminal-side models, wherein the multiple uplink transmission rate ranges include a first uplink transmission rate range and the first uplink transmission rate is included in the first uplink transmission rate range; the processing unit 802 is further configured to determine the first terminal-side model based on the first association information and the indication information.

[0276] In one possible implementation, the indication information includes the identifier of the first model and a first collaboration level, wherein the first collaboration level corresponds to the first terminal-side model.

[0277] In one possible implementation, the communication unit 803 is further configured to receive second association information, the second association information being used to indicate a one-to-one correspondence between multiple collaboration levels and multiple terminal-side models, the multiple collaboration levels including the first collaboration level; the processing unit 802 is further configured to determine the first terminal-side model based on the second association information and the indication information.

[0278] In one possible implementation, the first information includes the identifier of the first model and the identifier of the first terminal-side model; or, the first information includes the identifier of the first model and the first uplink transmission rate of the terminal, the first uplink transmission rate corresponding to the first network-side model; or, the first information includes the identifier of the first model and the first coordination level, the first coordination level corresponding to the first terminal-side model and the first network-side model.

[0279] In one possible implementation, the communication unit 803 is further configured to send a first session establishment request, the first session establishment request including an identifier of the first model, the first session establishment request being used to request the establishment of a session for inference of the first model.

[0280] For example, in another embodiment, processing unit 802 is configured to determine a first uplink transmission rate of the terminal; determine a first terminal-side model of a first model based on the first uplink transmission rate, wherein the first terminal-side model is one of a plurality of terminal-side models of the first model, and each of the plurality of terminal-side models is a sub-model of the first model processed for inference by the terminal; determine a first inference result based on the input data of the first model and the first terminal-side model; communication unit 803 is configured to send an inference request, wherein the inference request includes an identifier of the first model, the first inference result, and first information, wherein the inference request is used to request continued inference based on the first inference result, and the first information is used to determine a first network-side model of the first model, wherein the first network-side model is one of a plurality of network-side models of the first model, and each of the plurality of network-side models is a sub-model of the first model processed for inference by the network, and the first network-side model corresponds to the first terminal-side model; and receive a second inference result, wherein the second inference result is obtained based on the first inference result and the first network-side model.

[0281] In one possible implementation, the communication unit 803 is further configured to receive first association information, the first association information being used to indicate a one-to-one correspondence between multiple uplink transmission rate ranges and the multiple terminal-side models, the multiple uplink transmission rate ranges including a first uplink transmission rate range, the first uplink transmission rate being included in the first uplink transmission rate range; the processing unit 802 is configured to determine a first terminal-side model of the first model based on the first uplink transmission rate, including: determining the first terminal-side model based on the first association information and the first uplink transmission rate.

[0282] In one possible implementation, the first information includes the identifier of the first model and the identifier of the first terminal-side model; or, the first information includes the identifier of the first model and the first uplink transmission rate.

[0283] In one possible implementation, the first information includes the identifier of the first model and a first coordination level, the first coordination level corresponding to the first terminal-side model and the first network-side model.

[0284] In one possible implementation, the communication unit 803 is further configured to receive second association information, the second association information being used to indicate a one-to-one correspondence between multiple collaboration levels and the multiple terminal-side models, the multiple collaboration levels including the first collaboration level; the processing unit 802 is further configured to determine the first information based on the second association information and the first terminal-side model.

[0285] In one possible implementation, the communication unit 803 is further configured to send a first session establishment request, the first session establishment request including an identifier of the first model, the first session establishment request being used to request the establishment of a session for inference of the first model.

[0286] In one possible design, when the communication device 800 is a terminal device or a communication module within a terminal device, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor may include a modem chip, or a system-on-a-chip (SoC) chip or a SIP chip containing a modem core. The function of the communication unit 803 can be implemented by transceiver circuitry.

[0287] In one possible design, when the communication device 800 is a circuit or chip responsible for communication functions in a terminal device, such as a modem chip or a system-on-a-chip (SoC) or SIP chip containing a modem core, the function of the processing unit 802 can be implemented by a circuit system in the aforementioned chip that includes one or more processors or processor cores. The function of the communication unit 803 can be implemented by an interface circuit or data transceiver circuit on the aforementioned chip.

[0288] The communication device 800 can also be a network-side device in the above embodiments, such as a first device on the network side, a module (e.g., a circuit, chip, or chip system) in the first device, or a logic node, logic module, or software that can implement all or part of the functions of the first device.

[0289] For example, in one embodiment, the communication unit 803 is configured to receive a first request from a terminal, the first request including an identifier of a first model and model selection assistance information, the first request being used to request a terminal-side model of the first model corresponding to the model selection assistance information, the model selection assistance information including the amount of input data of the first model; the processing unit 802 is configured to determine a first terminal-side model of the first model according to the model selection assistance information, the first terminal-side model being one of a plurality of terminal-side models of the first model, each of the plurality of terminal-side models being a sub-model of the first model processed for inference by the terminal; the communication unit 803 is further configured to send indication information to the terminal, the indication information being used to indicate the first terminal-side model.

[0290] In one possible implementation, the processing unit 802 is configured to determine a first terminal-side model of the first model based on the model selection auxiliary information, including: determining the estimated data volume of the output data corresponding to each of the plurality of terminal-side models based on the data volume of the input data; determining the estimated uplink transmission rate corresponding to each of the plurality of terminal-side models based on the estimated data volume of the output data corresponding to each of the plurality of terminal-side models; determining a first uplink transmission rate based on the estimated uplink transmission rate corresponding to each of the plurality of terminal-side models, wherein the first uplink transmission rate is the maximum estimated uplink transmission rate that can be satisfied among the estimated uplink transmission rates corresponding to each of the plurality of terminal-side models; and determining the first terminal-side model based on the first uplink transmission rate.

[0291] In one possible implementation, the processing unit 802 is configured to determine a first uplink transmission rate based on the estimated uplink transmission rates corresponding to the plurality of terminal-side models, including: determining the first uplink transmission rate based on the estimated uplink transmission rates corresponding to the plurality of terminal-side models, and one or more of the identifier of the first model, the channel quality between the terminal and the first device, or the load of the first device.

[0292] In one possible implementation, the communication unit 803 is further configured to receive first association information, the first association information being used to indicate a one-to-one correspondence between multiple uplink transmission rate ranges and the multiple terminal-side models, the multiple uplink transmission rate ranges including a first uplink transmission rate range, the first uplink transmission rate being included in the first uplink transmission rate range; the processing unit 802 is configured to determine the first terminal-side model based on the first uplink transmission rate, including: determining the first terminal-side model based on the first uplink transmission rate and the first association information.

[0293] In one possible implementation, the model selection auxiliary information further includes the terminal-side estimated inference time corresponding to the amount of input data.

[0294] In one possible implementation, the indication information includes the identifier of the first model and the identifier of the first terminal-side model; or, the indication information includes the identifier of the first model and the first uplink transmission rate of the terminal; or, the indication information includes the identifier of the first model and the first coordination level, wherein the first coordination level corresponds to the first terminal-side model.

[0295] The communication device 800 can also be a network-side device in the above embodiments, such as a first device on the network side, a module (e.g., a circuit, chip, or chip system) in the first device, or a logic node, logic module, or software that can implement all or part of the functions of the first device. Alternatively, it can be a TEF network element on the network side, a module (e.g., a circuit, chip, or chip system) in the TEF network element, or a logic node, logic module, or software that can implement all or part of the functions of the TEF network element.

[0296] For example, in one embodiment, the communication unit 803 is configured to receive an inference request, the inference request including a first inference result and first information, the inference request being configured to request continued inference based on the first inference result, the first information being configured to determine a first network-side model of the first model, the first network-side model being one of a plurality of network-side models of the first model, each of the plurality of network-side models being a sub-model of the first model processed for network inference; the processing unit 802 is configured to send a second inference result through the communication unit 803 based on the first network-side model and the first inference result.

[0297] In one possible implementation, the first information includes the identifier of the first model and the identifier of the first terminal-side model.

[0298] In one possible implementation, the first information includes the identifier of the first model and the first uplink transmission rate of the terminal, wherein the first uplink transmission rate corresponds to the first network-side model.

[0299] In one possible implementation, the processing unit 802 is further configured to determine the first network-side model based on the third association information and the first uplink transmission rate, wherein the third association information is used to indicate a one-to-one correspondence between multiple uplink transmission rate ranges and the multiple network-side models, wherein the multiple uplink transmission rate ranges include a first uplink transmission rate range, and the first uplink transmission rate is included in the first uplink transmission rate range.

[0300] In one possible implementation, the communication unit 803 is further configured to receive the third association information from the second device.

[0301] In one possible implementation, the first information includes the identifier of the first model and a first coordination level, the first coordination level corresponding to the first network-side model.

[0302] In one possible implementation, the communication unit 803 is further configured to receive fourth association information, which indicates a one-to-one correspondence between multiple coordination levels and multiple network-side models, wherein the multiple coordination levels include the first coordination level; the processing unit 802 is further configured to determine the first network-side model based on the fourth association information and the first coordination level.

[0303] In one possible implementation, the communication unit 803 is further configured to receive the fourth association information from the second device.

[0304] In one possible implementation, the communication unit 803 is further configured to receive information from the plurality of network-side models from the second device, wherein the information of each network-side model includes the identifier of the network-side model and the parameters of the network-side model.

[0305] In one possible implementation, the communication unit 803 is further configured to send first association information to the terminal, the first association information being used to indicate a one-to-one correspondence between multiple uplink transmission rate ranges and multiple terminal-side models of the first model.

[0306] In one possible implementation, the communication unit 803 is further configured to send second association information to the terminal, the second association information being used to indicate a one-to-one correspondence between multiple coordination levels and multiple terminal-side models of the first model, the multiple coordination levels including a first coordination level corresponding to the first network-side model.

[0307] In one possible implementation, information about the plurality of terminal-side models is sent to the terminal, wherein the information about each terminal-side model includes the identifier of the terminal-side model and the parameters of the terminal-side model.

[0308] In one possible implementation, the communication unit 803 is further configured to receive a first session establishment request from the terminal, the first session establishment request including the identifier of the first model, the first session establishment request being used to request the establishment of a session for inference of the first model; the processing unit 802 is further configured to obtain information of the task execution function network element; the communication unit 803 is further configured to send a second session establishment request to the task execution function network element, the second session establishment request including the identifier of the first model, the second session establishment request being used to request the establishment of a session for inference of the first model.

[0309] The communication device 800 can also be a network-side device in the above embodiments, such as a second device on the network side, a module (e.g., a circuit, chip, or chip system) in the second device, or a logic node, logic module, or software that can implement all or part of the functions of the second device. Alternatively, it can be a TEF network element on the network side, a module (e.g., a circuit, chip, or chip system) in the TEF network element, or a logic node, logic module, or software that can implement all or part of the functions of the TEF network element.

[0310] For example, in one embodiment, the processing unit 802 is used to obtain a model processing strategy for a first model; and according to the model processing strategy, determine multiple terminal-side models and multiple network-side models of the first model, wherein each terminal-side model is a sub-model of the first model after processing for inference on the terminal, and each network-side model is a sub-model of the first model after processing for inference on the network, and the multiple terminal-side models and the multiple network-side models correspond one-to-one.

[0311] In one possible implementation, the model processing strategy is used to indicate the model processing methods corresponding to multiple uplink transmission rate ranges, wherein the multiple uplink transmission rate ranges, the multiple terminal-side models, and the multiple network-side models correspond one-to-one; the communication unit 803 is used to send first association information, wherein the first association information is used to indicate the one-to-one correspondence between the multiple uplink transmission rate ranges and the multiple terminal-side models.

[0312] In one possible implementation, the communication unit 803 is used to send third association information, which is used to indicate the one-to-one correspondence between the plurality of uplink transmission rate ranges and the plurality of network-side models.

[0313] In one possible implementation, the model processing strategy is used to indicate the model processing methods corresponding to multiple collaboration levels, wherein the multiple collaboration levels, the multiple terminal-side models, and the multiple network-side models correspond one-to-one; the communication unit 803 is used to send second association information, wherein the second association information is used to indicate the one-to-one correspondence between the multiple collaboration levels and the multiple terminal-side models.

[0314] In one possible implementation, the communication unit 803 is used to send fourth association information, which is used to indicate the one-to-one correspondence between the plurality of coordination levels and the plurality of network-side models.

[0315] It is understood that the division of units in the above-described device is merely a logical functional division. One function can correspond to one functional unit, or two or more functions can be integrated into one functional unit. In actual implementation, all or some units can be integrated onto a single physical entity, or distributed across different physical entities. Furthermore, the aforementioned functional units can be implemented in hardware, software, or a combination of both. Whether a function is executed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for specific applications, but such implementations should not be considered beyond the scope of this application.

[0316] In one example, the functional unit in any of the above devices may be one or more integrated circuits configured to implement the above methods, such as: one or more application-specific integrated circuits (ASICs), or one or more central processing units (CPUs), one or more microcontroller units (MCUs), one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs), or a combination of at least two of these integrated circuit forms.

[0317] In one example, storage unit 801 may include random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory and / or registers, etc.

[0318] Figure 9 is a schematic diagram of the structure of a terminal device 900 provided in an embodiment of this application. This terminal device 900 corresponds to the terminal devices shown in Figures 1 and 2 and is used to implement the operation of the terminal devices in the above embodiments. As shown in Figure 9, the terminal device includes: one or more antennas 910, a radio frequency processing system 920, and a processor system 930.

[0319] In the downlink or sidelink direction, the RF processing system 920 receives RF signals through the antenna 910 and sends the RF-processed signals to the processor system 930 for further processing. In the uplink or sidelink direction, the processor system 930 processes the information from the terminal device side and sends it to the RF processing system 920, which then processes the signal and transmits it through the antenna 910.

[0320] In one example, the radio frequency (RF) processing system 920 serves as the communication interface for external communication of the terminal device and may include a radio frequency frontend (RFFE) 921 and an RF transceiver 922. The RFFE 921 is primarily used for one or more processing operations, such as shaping, passband selection, or gain adjustment, on the RF signals received by the antenna or those to be transmitted through the antenna. It may include one or more components such as RF switches, duplexers, filters, power amplifiers, antenna tuners, and low-noise amplifiers. The RFFE 921 can be a circuit system composed of multiple discrete components or integrated into one or more chips. The RF transceiver 922 processes the RF signals received by the RFFE into baseband / IF signals for further processing by the processor system 930, and processes the baseband / IF signals provided by the processor system 930 into RF signals for transmission to the RFFE 921. The baseband / IF signals transmitted between the RF transceiver 922 and the processor system 930 can be digital or analog signals. The RF transceiver 922 can be implemented by one or more chips, which are commonly referred to as RF ICs.

[0321] In one example, the processor system 930 may include one or more processors for processing signals and executing one or more communication protocols. Optionally, the processor system 930 may also include a memory 936. In one example, the one or more processors include at least one baseband processor 931 (also known as a modem processor). The memory 936 is used to store data and / or computer program instructions. Optionally, the processor system 930 may also include one or more application processors 932 for implementing processing of the terminal device's operating system and application layer. Optionally, the processor system 930 may also include one or more of a voice subsystem 933, a multimedia subsystem 934, or an interface circuit 935. The voice subsystem 933 is used to process voice signals, the multimedia subsystem 934 is used to handle multimedia-related operations, such as video encoding / decoding, image processing, etc., and the interface circuit 935 is used to enable communication with other terminal device components, such as a display 940, an input device 950, a memory 960, etc. The above-mentioned components in the processor system 930 can communicate with each other via a bus or communication interface circuit.

[0322] In one example, the processor system 930 can be packaged as a single processor chip, such as a SoC chip or a SIP chip. In another example, the processor system 930 can be a system composed of multiple chips; for example, the baseband processor 931 can be packaged as a single chip, or packaged with part or all of the circuitry of the radio frequency processing system into a single chip.

[0323] In one example, memory 936 can be on-chip memory, i.e., located on the system-on-a-chip (SoC) 930. In another example, memory 960 can be off-chip memory, i.e., located outside the SoC 930.

[0324] In one example, the baseband processor 931 may include one or more processor cores 9311 and interface circuitry 9314. The one or more processor cores 9311 are used to process signals and execute one or more communication protocols. Optionally, the baseband processor 931 may also include a memory 9312 for storing at least a portion of the corresponding computer program instructions and / or data. In one example, the one or more processor cores 9311 execute the computer program instructions stored in the memory 9312 to implement the relevant operations in the above method embodiments. In this disclosure, the memory 9312 storing the corresponding computer program instructions and / or data may mean that the memory 9312 stores all the corresponding computer program instructions and / or data for the processor core 9311 to execute; or it may mean that the memory 9312 stores a portion of the corresponding computer program instructions and / or data, which includes the computer program instructions and / or data currently required to be executed by the processor core 9311. The memory 9312 can store different portions of computer program instructions and / or data multiple times for the processor core 9311 to execute in order to implement the relevant operations in the above method embodiments. Interface circuit 9314 serves as a communication interface for communication with other components, such as transmitting signals with RF processing system 920, communicating with other subsystems and related components of processor system 930 via bus, such as transmitting data control signals with application processor 932, and transmitting data or computer program instructions with memory 936 or memory 960. Optionally, to reduce the load on the processor core, baseband signal processing circuit 9313 can also be provided to perform at least some baseband signal processing, including one or more of signal demodulation, modulation, encoding, or decoding.

[0325] In one example, the communication device provided in this application may be a terminal device 900, including a communication module comprising a processor system 930 and a radio frequency system 920, or a baseband processor 931.

[0326] The processor, processor system, application processor, baseband processor, processor circuit, or processor core mentioned above can be collectively referred to as a processor. The processor may include one or more of the following: central processing unit (CPU), digital signal processor (DSP), microprocessor unit (MPU), microcontroller unit (MCU), graphics processing unit (GPU), field programmable gate array (FPGA), artificial intelligence processor (AI processor), or neural processing unit (NPU).

[0327] The aforementioned memory may include one or more of the following storage media: random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), phase-change memory (PCM), resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), cache, register, read-only memory (ROM), flash memory, erasable programmable read-only memory (EPROM), hard disk, etc. In one example, computer program instructions for executing the above embodiments may be stored on non-volatile memory, such as at least a portion of the aforementioned memory 960 (e.g., one or more of ROM, flash memory, EPROM, or hard disk). When the terminal device is running, the corresponding computer program instructions may be partially or wholly loaded onto a memory with a faster transfer speed than the processor, such as at least a portion of the memory 936 and / or memory 9312 (e.g., one or more of RAM, SRAM, DRAM, PCM, RERAM, MRAM, FRAM, cache, or register), for the processor to execute in order to implement the steps in the above method embodiments.

[0328] In one example, the RF transceiver 922 and the RF front-end 921 can also be packaged in a single chip. In another example, the RF transceiver 922, the RF front-end 921, and the baseband processor 931 can also be packaged in a single chip.

[0329] In one possible implementation, this application provides a computer-readable storage medium storing a computer program or instructions that, when executed, implement any of the above-described method embodiments.

[0330] In one possible implementation, this application provides a computer program product including instructions that, when executed, implement any of the above-described method embodiments.

[0331] In one possible implementation, this application provides a communication system including a terminal and a first device. The terminal is used to implement the operations performed by the terminal in the embodiment of FIG3. The first device is used to implement the operations performed by the first device in the embodiment of FIG3.

[0332] In one possible implementation, this application provides a communication system including a terminal and a first device. The terminal is used to implement the operations performed by the terminal in the embodiment of FIG. 5. The first device is used to implement the operations performed by the first device in the embodiment of FIG. 5.

[0333] In one possible implementation, this application provides a communication system including a terminal and a TEF network element. The terminal is used to implement the operations performed by the terminal in the embodiment of FIG. 5. The TEF network element is used to implement the operations performed by the TEF network element in the embodiment of FIG. 5.

[0334] The terms "system" and "network" in this application embodiment are used interchangeably. "At least one" refers to one or more, and "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, "at least one of A, B, or C" includes A, B, C, AB, AC, BC, or ABC; "at least one of A, B, and C" can also be understood as including A, B, C, AB, AC, BC, or ABC. Furthermore, unless otherwise specified, the ordinal numbers such as "first" and "second" mentioned in this application embodiment are used to distinguish multiple objects and are not used to limit the order, sequence, priority, or importance of multiple objects.

[0335] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0336] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0337] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0338] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0339] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A communication method characterized by comprising: The method comprises: sending a first request to a first device, the first request comprising an identification of a first model and model selection assistance information, the first request being used to request a terminal-side model of the first model corresponding to the model selection assistance information, the model selection assistance information comprising a data volume of input data of the first model; receiving indication information from the first device, the indication information being used to indicate a first terminal-side model of the first model, the first terminal-side model being one of a plurality of terminal-side models of the first model, each terminal-side model of the plurality of terminal-side models being a sub-model of the first model processed for terminal inference; determining a first inference result according to the input data and the first terminal-side model; sending an inference request, the inference request comprising the first inference result and first information, the inference request being used to request continued inference based on the first inference result, the first information being used to determine a first network-side model of the first model, the first network-side model being one of a plurality of network-side models of the first model, each network-side model of the plurality of network-side models being a sub-model of the first model processed for network inference, the first network-side model corresponding to the first terminal-side model; receiving a second inference result, the second inference result being obtained according to the first inference result and the first network-side model.

2. The method of claim 1, wherein, The model selection assistance information further comprises a terminal-side estimated inference time length corresponding to the data volume of the input data.

3. The method of claim 1 or 2, wherein, The indication information comprises an identification of the first model and an identification of the first terminal-side model.

4. The method of claim 1 or 2, wherein, The indication information comprises an identification of the first model and a first uplink transmission rate of the terminal, the first uplink transmission rate corresponding to the first terminal-side model.

5. The method of claim 4, wherein, The method further comprises: receiving first association information, the first association information being used to indicate a one-to-one correspondence relationship between a plurality of uplink transmission rate ranges and the plurality of terminal-side models, the plurality of uplink transmission rate ranges comprising a first uplink transmission rate range, the first uplink transmission rate being included in the first uplink transmission rate range; determining the first terminal-side model according to the first association information and the indication information.

6. The method of claim 1 or 2, wherein, The indication information comprises an identification of the first model and a first cooperation level, the first cooperation level corresponding to the first terminal-side model.

7. The method of claim 6, wherein, The method further comprises: receiving second association information, the second association information being used to indicate a one-to-one correspondence relationship between a plurality of cooperation levels and the plurality of terminal-side models, the plurality of cooperation levels comprising the first cooperation level; determining the first terminal-side model according to the second association information and the indication information.

8. The method of any one of claims 1 to 7, wherein: the first information comprises an identification of the first model and an identification of the first terminal-side model; or the first information comprises an identification of the first model and a first uplink transmission rate of the terminal, the first uplink transmission rate corresponding to the first network-side model; or The first information includes an identifier of the first model and a first collaboration level corresponding to the first terminal-side model and the first network-side model.

9. The method of any one of claims 1 to 8, wherein, Further comprising: sending a first session establishment request including an identifier of the first model, the first session establishment request being used to request establishment of a session for inference of the first model.

10. A communication method characterized by comprising: comprising: determining a first uplink transmission rate of a terminal; determining, according to the first uplink transmission rate, a first terminal-side model of the first model, the first terminal-side model being one of a plurality of terminal-side models of the first model, each terminal-side model of the plurality of terminal-side models being a sub-model of the first model processed for inference by a terminal; determining a first inference result according to input data of the first model and the first terminal-side model; sending an inference request including an identifier of the first model, the first inference result, and first information, the inference request being used to request continued inference based on the first inference result, the first information being used to determine a first network-side model of the first model, the first network-side model being one of a plurality of network-side models of the first model, each network-side model of the plurality of network-side models being a sub-model of the first model processed for inference by a network, the first network-side model corresponding to the first terminal-side model; receiving a second inference result obtained according to the first inference result and the first network-side model.

11. The method of claim 10, wherein, Further comprising: receiving first association information indicating a one-to-one correspondence between a plurality of uplink transmission rate ranges and the plurality of terminal-side models, the plurality of uplink transmission rate ranges including a first uplink transmission rate range, the first uplink transmission rate being included in the first uplink transmission rate range; The determining, according to the first uplink transmission rate, of the first terminal-side model of the first model comprises: determining, according to the first association information and the first uplink transmission rate, the first terminal-side model.

12. The method of claim 10 or 11, wherein the first information includes an identifier of the first model and an identifier of the first terminal-side model; or the first information includes an identifier of the first model and the first uplink transmission rate.

13. The method of claim 10 or 11, wherein the first information includes an identifier of the first model and a first collaboration level corresponding to the first terminal-side model and the first network-side model.

14. The method of claim 13, wherein, Further comprising: receiving second association information indicating a one-to-one correspondence between a plurality of collaboration levels and the plurality of terminal-side models, the plurality of collaboration levels including the first collaboration level; determining the first information according to the second association information and the first terminal-side model.

15. The method of any one of claims 10 to 14, wherein, Further comprising: sending a first session establishment request, the first session establishment request comprising an identification of the first model, the first session establishment request being used to request establishment of a session for inference of the first model.

16. A method of communication, comprising: comprising: receiving a first request from a terminal, the first request comprising an identification of a first model and model selection assistance information, the first request being used to request a terminal-side model of the first model corresponding to the model selection assistance information, the model selection assistance information comprising a data volume of input data of the first model; determining, according to the model selection assistance information, a first terminal-side model of the first model, the first terminal-side model being one of a plurality of terminal-side models of the first model, each terminal-side model of the plurality of terminal-side models being a sub-model of the first model processed for inference by a terminal; sending indication information to the terminal, the indication information being used to indicate the first terminal-side model.

17. The method of claim 16, wherein, The determining, according to the model selection assistance information, a first terminal-side model of the first model, comprises: determining, according to the data volume of the input data, an estimated data volume of output data corresponding to the plurality of terminal-side models respectively; determining, according to the estimated data volume of output data corresponding to the plurality of terminal-side models respectively, an estimated uplink transmission rate corresponding to the plurality of terminal-side models respectively; determining, according to the estimated uplink transmission rate corresponding to the plurality of terminal-side models respectively, a first uplink transmission rate, the first uplink transmission rate being a maximum estimated uplink transmission rate that can be satisfied among the estimated uplink transmission rates corresponding to the plurality of terminal-side models respectively; determining, according to the first uplink transmission rate, the first terminal-side model.

18. The method of claim 17, wherein, The determining, according to the estimated uplink transmission rate corresponding to the plurality of terminal-side models respectively, a first uplink transmission rate, comprises: determining, according to the estimated uplink transmission rate corresponding to the plurality of terminal-side models respectively, and one or more of the identification of the first model, a channel quality between the terminal and the first device, or a load of the first device, the first uplink transmission rate.

19. The method of claim 17 or 18, wherein, Further comprising: receiving first association information, the first association information being used to indicate a one-to-one correspondence relationship between a plurality of uplink transmission rate ranges and the plurality of terminal-side models, the plurality of uplink transmission rate ranges comprising a first uplink transmission rate range, the first uplink transmission rate being contained in the first uplink transmission rate range; The determining, according to the first uplink transmission rate, the first terminal-side model, comprises: determining, according to the first uplink transmission rate and the first association information, the first terminal-side model.

20. The method of any one of claims 16 to 19, wherein, The model selection assistance information further comprises a terminal-side estimated inference time length corresponding to the data volume of the input data.

21. The method of any one of claims 16 to 20, wherein: the indication information comprises the identification of the first model and the identification of the first terminal-side model; or the indication information comprises the identification of the first model and a first uplink transmission rate of the terminal; or The indication information comprises an identifier of the first model and a first collaboration level, and the first collaboration level corresponds to the first terminal-side model.

22. A communications device, characterized by A module for performing the method of any one of claims 1-9, or the method of any one of claims 10-15, or the method of any one of claims 16-21.

23. A communications device, characterized by A device comprising a processor and an interface circuit, the processor configured to communicate with other devices via the interface circuit to implement the method of any one of claims 1-9, or to implement the method of any one of claims 10-15, or to implement the method of any one of claims 16-21.

24. A computer program product, characterised in that, The computer program product comprises instructions which, when executed, implement the method of any one of claims 1-9, or implement the method of any one of claims 10-15, or implement the method of any one of claims 16-21.

25. A computer-readable storage medium, characterized in that, The storage medium stores a computer program or instructions which, when executed, implement the method of any one of claims 1-9, or implement the method of any one of claims 10-15, or implement the method of any one of claims 16-21.

26. A communication system, characterized by Comprising: a terminal for implementing the method of any one of claims 1-9; a first device for implementing the method of any one of claims 16-21.

27. A communication system, characterized by Comprising: a terminal for implementing the method of any one of claims 10-15; a communication device for receiving an inference request from the terminal, the inference request comprising an identifier of a first model, a first inference result, and first information, the inference request being used to request to continue inference based on the first inference result, the first information being used to determine a first network-side model of the first model, the first network-side model being one of a plurality of network-side models of the first model, each network-side model of the plurality of network-side models being a sub-model of the first model processed for network inference, the first network-side model corresponding to the first terminal-side model; and sending a second inference result to the terminal, the second inference result being obtained according to the first inference result and the first network-side model. ​

Images