Configuration method for distributed ai task, management node, and computing node

Abstract

The present application provides a configuration method for a distributed AI task, a management node, and a computing node. In the method, a management node determines at least two sub-models on the basis of task information of a first task and a plurality of stored functional modules, and sends attribute information and configuration information of the at least two sub-models to at least two computing nodes, so that the at least two sub-models deployed on the at least two computing nodes can jointly process the first task. The at least two sub-models are sub-models assembled by the management node using the functional modules on the basis of the task information, so that the management node can achieve generalization of AI models via the stored functional modules without needing to store specific AI models for tasks in advance, and the management node can determine, on the basis of the task information, sub-models suitable for processing the task, thereby improving the performance of each distributed computing node in providing AI services.

Description

A configuration method, management node and computing node for distributed AI tasks

[0001] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office on March 28, 2024, with application number 202410376209.1 and invention name “A configuration method, management node and computing node for distributed AI tasks”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The embodiments of the present application relate to the field of communications, and in particular to a configuration method, management node, and computing node for distributed AI tasks. Background Art

[0003] With the development of artificial intelligence (AI) technology, AI technology has spawned intelligent applications in various industries and has also spawned powerful functions in the communications field.

[0004] In the communications field, communication modules utilizing AI technology can replace traditional communication modules, enabling more intelligent network functions. For example, AI4Net (AI for Network) can improve network performance, efficiency, and user experience through AI. For example, AI technology applied to the physical layer can enable intelligent beam management, channel prediction, and resource allocation. Another example is AI technology applied to higher layers, enabling terminal trajectory prediction, load balancing, and energy saving. Furthermore, the network's computing, transmission capabilities, and perceived information can also support third-party AI applications on mobile terminals. For example, by offloading computations for third-party AI applications to computing nodes within the network, the network can leverage environmental information and user behavior to provide better personalized services. Given these two potential scenarios and driving forces, the deep integration of AI and communications has become one of the key visions of the sixth-generation mobile communication technology (6G) system: through the deep integration of AI capabilities, mobile networks will provide native AI services for both network-wide use and third-party applications.

[0005] In traditional technologies, multiple computing nodes are distributed within the access network. A management node, which manages these computing nodes, receives task requests from base stations or third-party AI applications. Based on these task requests, it selects an AI model with specific functionality and a participating computing node for the task. The management node then sends the selected AI model to each computing node, enabling it to deploy the model, process the task data using it, and output the results of the AI ​​task.

[0006] However, the number of AI models with specific functions that can be stored in a management node is limited, and the task processing strategy for AI models with specific functions is fixed. The management node may not be able to find the appropriate AI model for the current AI task and deploy it on each participating computing node, which in turn affects the performance of the AI ​​service provided by each distributed computing node. Summary of the Invention

[0007] The present application provides a configuration method, management node, and computing node for distributed AI tasks, which are used to achieve the generalization of the AI ​​model within the management node, so that the sub-model deployed by the management node instructs each computing node to form a target model suitable for processing the task, thereby improving the performance of each distributed computing node in providing AI services.

[0008] In the first aspect, the present application provides a method for configuring a distributed AI task, which can be executed by a management node that manages computing nodes, or by a component of the management node (for example, a processor, a chip, or a chip system). Taking the management node as an example, the management node receives a first task request, and the first task request includes task information of the first task; then, the management node determines at least two sub-models and attribute information of the sub-model based on the task information and the attribute information of multiple functional modules stored by the management node. Wherein, the sub-model includes at least one functional module and / or a part of a functional module. Then, the management node sends the attribute information of the sub-models to be deployed by each computing node and the sub-model to at least two computing nodes, so that at least two sub-models are distributedly deployed in at least two computing nodes. Wherein, the attribute information of the sub-model is determined based on the attribute information of the functional modules constituting the sub-model, and the attribute information of the sub-model is used for computing nodes to deploy sub-models. Then, the management node sends the configuration information of the sub-model deployed by the computing node to at least two computing nodes based on the task information. The configuration information of the sub-model is used to configure the model parameters used by the sub-model when participating in processing the first task and to establish a connection relationship between the sub-model and the sub-model deployed on different computing nodes.

[0009] In this aspect, the management node can determine at least two sub-models based on the task information of the first task and the stored multiple functional modules, and send the attribute information and configuration information of the at least two sub-models to at least two computing nodes, so that the at least two sub-models deployed on the at least two computing nodes can jointly process the first task. Since the at least two sub-models are sub-models spliced ​​together by the management node using functional modules based on the task information, the management node realizes the generalization of the AI ​​model through the stored functional modules, and there is no need for the management node to store specific AI models for the task in advance, so that the management node determines the sub-model suitable for processing the task based on the task information, thereby improving the performance of each distributed computing node in providing AI services.

[0010] In one possible implementation, the attribute information of a functional module includes a functional type and functional module requirements. The functional module requirements indicate the requirements for data processing and / or data transmission when implementing the functional module's functions. The functional type of a functional module indicates the AI ​​processing function implemented by the functional module and the adaptation conditions for the functional module to implement the AI ​​processing function.

[0011] In this embodiment, the attribute information of the functional module includes the requirements of the functional module, which is conducive to the management node determining a sub-model that can meet the task requirements of the first task based on the requirements of the functional module, thereby improving the performance of the computing node of the deployed sub-model determined by the management node in providing AI services.

[0012] In one possible implementation, the task information of the first task includes the task type of the first task and the task requirement of the first task, where the task requirement of the first task is used to indicate the requirements of the distributed AI processing service requested by the first task for data processing and / or data transmission.

[0013] In this embodiment, the task information of the first task includes the task requirements of the first task, which is conducive to the management node determining a sub-model that can meet the task requirements of the first task based on the requirements of the functional module, thereby improving the performance of the computing node of the deployed sub-model determined by the management node in providing AI services.

[0014] In one possible implementation, the management node determines at least two sub-models and attribute information of the sub-models based on task information and attribute information of multiple functional modules stored in the management node, including: the management node determines a target model that supports processing a first task based on the task information and attribute information of multiple functional modules, the target model is composed of at least one functional module among the multiple functional modules, the task type of the first task is used to determine the model type of the target model, the performance requirements of the target model meet the task requirements of the first task, and the performance requirements of the target model are determined based on the requirements of at least one functional module; then, the management node determines at least two sub-models and attribute information of each sub-model based on the target model.

[0015] In this embodiment, the management node can screen out functional modules that can meet the task requirements of the first task based on the task requirements of the first task and the performance requirements of the functional modules, and then ensure that the target model composed of multiple functional modules can also meet the task requirements of the first task, thereby improving the matching rate of the management node configuration tasks and improving the performance of the AI ​​processing service.

[0016] In one possible implementation, the attribute information of the sub-model includes the AI ​​resource requirements of the sub-model; before the management node sends the sub-models to be deployed and the attribute information of the sub-models to at least two computing nodes, the method further includes: the management node obtains resource status information of at least one computing node, where the resource status information is used to indicate the usage status of the AI ​​resources of the computing node, where the AI ​​resources include model resources, computing resources, and data resources; the management node determines at least two computing nodes based on the resource status information of the at least one computing node, and determines the sub-model to be deployed for each of the at least two computing nodes, where the AI ​​resources of the computing nodes meet the AI ​​resource requirements of the sub-models to be deployed by the computing nodes.

[0017] In this embodiment, the management node determines the sub-model to be deployed for each computing node based on the collected resource status of each computing node, which is conducive to ensuring that the computing node has sufficient AI resources to deploy the sub-model, reducing the probability of failure of the computing node to deploy the sub-model, and improving the efficiency of the management node in configuring the sub-model for each computing node.

[0018] In a possible embodiment, the method also includes: the management node receives network status information from the access network device connected to the computing node, the network status information is used to indicate the network status of the communication device applying for the first task; the management node determines the configuration information of the sub-model to be deployed by the computing node based on the task information, the resource status information of the computing node and the network status information; wherein the network status information includes first network status information and / or second network status information, the first network status information is used to indicate the air interface status of the access network device connected to the computing node, and the second network status information is used to indicate the air interface status of the terminal device sending the first task request.

[0019] In this embodiment, when determining the configuration information of the sub-model to be deployed by the computing node, the management node not only considers the task information of the first task, but also considers the resource status information and network status information of the computing node, which is conducive to determining the appropriate configuration for each sub-model, thereby improving the performance of the AI ​​model and improving the performance of the computing node in providing AI services.

[0020] In one possible implementation, the attribute information of the functional module also includes information of multiple network layers contained in the functional module, the input dimension of the functional module, and the output dimension of the functional module; the attribute information of the sub-model includes information of multiple network layers contained in the sub-model.

[0021] In a possible implementation, the configuration information of the sub-model includes parameters of an enabled network layer, where the parameters of the enabled network layer are used to indicate the network layer that the sub-model deployed on the computing node enables when participating in processing the first task.

[0022] In this implementation, the management node can flexibly indicate which network layers in the sub-model are enabled through parameters of the enabled network layers, thereby improving the efficiency of the management node in configuring the sub-model.

[0023] In a possible implementation, the configuration information of the sub-model further includes a connection relationship between the sub-model and sub-models deployed on other computing nodes.

[0024] In this embodiment, the management node can indicate to the computing node the connection relationship between the sub-model and the sub-model deployed on other computing nodes to ensure that data interaction can be carried out between the sub-models deployed on different computing nodes, which is conducive to improving the efficiency of the management node in configuring the sub-model.

[0025] In one possible implementation, the management node sends configuration information of a sub-model deployed by the computing nodes to at least two computing nodes based on task information, including: the management node sends first configuration information of a first sub-model to a first computing node, the first configuration information including a first model parameter, the first model parameter being used to indicate enabling of a first network layer in the first sub-model, the first network layer being at least one of a plurality of network layers indicated by attribute information of the first sub-model.

[0026] In this embodiment, the management node sends at least a first model parameter to the first computing node. The first model parameter indicates some of the multiple network layers indicated by the attribute information of the first sub-model, indicating that the attribute information of the first sub-model indicates some redundant network layers, that is, network layers that are not enabled when processing the first task. It can be understood that the management node configures the network layers that need to be enabled and the redundant network layers for the first sub-model through the attribute information of the first sub-model, and indicates the network layers that need to be enabled for the first task through the configuration information of the first sub-model. This is beneficial for the computing node to enhance the function or capability of the sub-model based on the redundant network layer, and improves the adaptability of the sub-model deployed in the computing node to the requirements of the first task.

[0027] In a possible embodiment, the method also includes: the management node sends second configuration information of the first sub-model to the first computing node, the second configuration information includes second model parameters, the second model parameters are used to indicate enabling of the second network layer in the first sub-model, the second network layer is at least one network layer in the first computing node, and the second network layer is different from the first network layer.

[0028] In this embodiment, the management node instructs the first computing node to enable the new network layer in the first sub-model by issuing new configuration information, thereby enhancing the function of the first sub-model, which is conducive to improving the adaptability of the sub-model deployed in the computing node to the requirements of the first task.

[0029] In a possible embodiment, the method further includes: the management node sends the target model and attribute information of the target model to the first computing node, the attribute information of the target model includes information of the network layer contained in each sub-model in at least two sub-models, and the connection relationship between the network layers of different sub-models in at least two sub-models.

[0030] In this embodiment, the management node sends the target model and the attribute information of the target model to the computing node, which is conducive to the computing node autonomously modifying the network layer enabled by the first sub-model based on the attribute information of the target model, thereby enhancing the sub-model, which is conducive to improving the sub-model in the computing node and improving the service quality of the AI ​​service.

[0031] In a possible implementation, the method further includes: the management node sending configuration information of the participating node to at least one participating node, where the configuration information of the participating node is used to indicate a data interaction strategy between the participating node and the computing node.

[0032] In a possible implementation, the method further includes: the management node receiving auxiliary information of the first task, the auxiliary information including prediction information of the access network device on the network environment and / or user behavior, and the auxiliary information is related to the task type of the first task.

[0033] In this embodiment, the access network device provides auxiliary information of the first task to the management node, so that the management node sends the pre-processed auxiliary information to the computing node, thereby prompting the computing node to use the auxiliary information as input data of the AI ​​model, which is conducive to enhancing the performance of the sub-model in processing the first task, improving the ability of the sub-model, and providing personalized AI services based on user preferences.

[0034] In a possible embodiment, the method also includes: the management node receives at least one backbone model and attribute information of the backbone model, the attribute information of the backbone model is used to indicate the model function and model parameters of the backbone model; the management node determines multiple functional modules and attribute information of each functional module based on at least one backbone model, a functional module is used to implement a sub-function in the model function of the backbone model, and different functional modules are used to implement different sub-functions in the model function of the backbone model or different precision / granularity of the same sub-function.

[0035] In this implementation, the management node can split the received backbone model into functional modules for storage, rather than directly storing the backbone model. This not only helps conserve management node storage resources but also facilitates the management node to generate sub-models from the functional modules, increasing the diversity of sub-models that the management node can generate. For deployment nodes, the interconnection between functional modules can flexibly support various types of distributed AI paradigms.

[0036] In a possible implementation, the method further includes: the management node receiving at least one functional module and attribute information of the functional module.

[0037] In the second aspect, the present application provides a method for configuring a distributed AI task, which can be executed by a computing node managed by a management node, or by a component of the computing node (for example, a processor, a chip, or a chip system). Taking the first computing node as an example, the first computing node is a node managed by the management node and participating in the processing of a distributed AI task. After the management node determines the first sub-model and the attribute information of the first sub-model sent to the first computing node, the first computing node can receive the first sub-model and the attribute information of the first sub-model of the first task sent by the management node, and the attribute information of the first sub-model is used to indicate the structure and function of the first sub-model; then, the first computing node deploys the first sub-model based on the attribute information of the first sub-model; then, the first computing node receives the configuration information of the first sub-model sent by the management node, and the configuration information of the first sub-model is used to configure the model parameters used by the first sub-model when participating in the processing of the first task and the connection relationship between the sub-models deployed on different computing nodes. The first computing node configures the first sub-model based on the configuration information of the first sub-model, and establishes a connection relationship between the first sub-model and the sub-models deployed on different computing nodes.

[0038] In a possible implementation, the attribute information of the first sub-model includes information of multiple network layers included in the first sub-model.

[0039] In a possible implementation, the configuration information of the first sub-model includes parameters of an enabled network layer, where the parameters of the enabled network layer are used to indicate the network layer that the first sub-model deployed on the first computing node enables when participating in processing the first task.

[0040] In a possible implementation, the configuration information of the first sub-model further includes a connection relationship between the first sub-model and sub-models deployed on other computing nodes.

[0041] In one possible implementation, the configuration information of the first sub-model includes first configuration information, the first configuration information includes a first model parameter, the first model parameter is used to indicate enabling of a first network layer in the first sub-model, the first network layer being at least one network layer among multiple network layers indicated by the attribute information of the first sub-model. In this case, the first computing node deploying the first sub-model based on the attribute information of the first sub-model includes: the first computing node enabling the first network layer in the first sub-model based on the first model parameter.

[0042] In a possible embodiment, the method also includes: the first computing node receives second configuration information of the first sub-model, the second configuration information includes a second model parameter, the second model parameter is used to indicate enabling of a second network layer in the first sub-model, the second network layer is at least one network layer in the first computing node, and the second network layer is different from the first network layer; the first computing node enables the second network layer in the first sub-model based on the second model parameter.

[0043] In a possible embodiment, the method further includes: the first computing node receives the target model and attribute information of the target model, the attribute information of the target model includes information of the network layers contained in each sub-model in at least two sub-models, and the connection relationship between the network layers of different sub-models in at least two sub-models.

[0044] In a possible implementation, the method further includes: the first computing node sending resource status information of the first computing node, where the resource status information is used to indicate the usage status of AI resources of the first computing node, where the AI ​​resources include model resources, computing resources, and data resources.

[0045] In a third aspect, an embodiment of the present application provides a device, which may be the management node in the aforementioned embodiment, or a chip within the management node. The device may include a processing module and a transceiver module. When the device is a management node, the processing module may be a processor, and the transceiver module may be a transceiver; the management node may further include a storage module, and the storage module may be a memory; the storage module is used to store instructions, and the processing module executes the instructions stored in the storage module so that the management node executes the method in any one of the embodiments of the first aspect. When the device is a chip within a management node, the processing module may be a processor, and the transceiver module may be an input / output interface, a pin, or a circuit, etc.; the processing module executes the instructions stored in the storage module so that the management node executes the method in any one of the embodiments of the first aspect. The storage module may be a storage module within the chip (for example, a register, a cache, etc.), or it may be a storage module within the management node that is located outside the chip (for example, a read-only memory, a random access memory, etc.).

[0046] In a fourth aspect, an embodiment of the present application provides a device, which may be a computing node in the aforementioned embodiment, or a chip within the computing node. The device may include a processing module and a transceiver module. When the device is a computing node, the processing module may be a processor, and the transceiver module may be a transceiver; the computing node may further include a storage module, which may be a memory; the storage module is used to store instructions, and the processing module executes the instructions stored in the storage module to enable the computing node to perform the method in the second aspect or any one of the embodiments of the second aspect. When the device is a chip within a computing node, the processing module may be a processor, and the transceiver module may be an input / output interface, a pin, or a circuit, etc.; the processing module executes the instructions stored in the storage module to enable the computing node to perform the method in the second aspect or any one of the embodiments of the second aspect. The storage module may be a storage module within the chip (e.g., a register, a cache, etc.), or a storage module within the computing node located outside the chip (e.g., a read-only memory, a random access memory, etc.).

[0047] In a fifth aspect, the present application provides a device, which may be an integrated circuit chip. The integrated circuit chip includes a processor. The processor is coupled to a memory, which stores programs or instructions. When the program or instructions are executed by the processor, the device performs the method described in any of the embodiments of the aforementioned aspects.

[0048] In a sixth aspect, an embodiment of the present application provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute a method as described in any one of the aforementioned aspects.

[0049] In a seventh aspect, an embodiment of the present application provides a computer-readable storage medium comprising instructions, which, when executed on a computer, enable the computer to execute a method as described in any one of the embodiments in the foregoing aspects.

[0050] In an eighth aspect, an embodiment of the present application provides a communication system, which includes a management node executing the aforementioned first aspect and any one of the implementations of the first aspect, a computing node executing the aforementioned second aspect and any one of the implementations of the second aspect, and an access network device.

[0051] In the ninth aspect, an embodiment of the present application provides a communication system, which includes a management node executing the aforementioned first aspect and any one of the implementations of the first aspect, a computing node executing the aforementioned second aspect and any one of the implementations of the second aspect, an access network device and a terminal device. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] FIG1 is an example diagram of a network system to which the configuration method for distributed AI tasks provided in this application is applicable;

[0053] FIG2 is a flow chart of a method for configuring distributed AI tasks provided by this application;

[0054] FIG3A is an example diagram of a target model provided in this application;

[0055] FIG3B is another example diagram of the target model provided by this application;

[0056] FIG4 is another flow chart of the configuration method of the distributed AI task provided by this application;

[0057] FIG5 is another flowchart of the configuration method of the distributed AI task provided by this application;

[0058] FIG6 is a schematic diagram of the device provided in this application;

[0059] FIG7 is another schematic diagram of the device provided in this application. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0061] The terms "first," "second," "third," "fourth," and the like (if any) in the specification and claims of this application and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0062] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be single or multiple. In addition, the character " / " in this article generally indicates that the previous and next associated objects are in an "or" relationship. In addition, "at least one of the following" or similar expressions in this article is used to represent any combination of the listed items; for example, at least one of A, B and (or) C can represent the following six situations: A exists alone, B exists alone, C exists alone, A and B exist at the same time, B and C exist at the same time, A and C exist at the same time, and A, B and C exist at the same time, where A, B, and C can be single or multiple.

[0063] To facilitate understanding, the following first introduces the system architecture applicable to the configuration method of distributed AI tasks proposed in this application:

[0064] The configuration method of distributed AI tasks proposed in this application can be applied to the fifth generation mobile communication technology (5G) system, the sixth generation mobile communication technology (6G) system and other subsequent evolutionary standards, and this application is not limited to this.

[0065] Figure 1 is an example diagram of a network system applicable to the configuration method of distributed AI tasks provided in this application. As shown in Figure 1, the network system includes at least terminal devices, access network devices, computing nodes, and management nodes.

[0066] Among them, the terminal device includes a device that provides voice and / or data connectivity to the user. For example, it may include a handheld device with wireless connection function or a processing device connected to a wireless modem. The terminal device can communicate with the core network (for example, 5G core network (5th generation core, 5GC)) via the radio access network (RAN) and can exchange voice and / or data with the RAN. The terminal device may also be referred to as a terminal (Terminal), user equipment (UE), wireless terminal device, mobile terminal (MT) device, subscriber unit, subscriber station, mobile station (MS), mobile station (mobile), remote station (remote station), access point (AP), remote terminal device (remote terminal), access terminal device (access terminal), user terminal device (user terminal), user agent (user agent), or user equipment (user device), etc. In addition, the terminal device can be a mobile phone, a tablet computer, a computer with wireless transceiver function, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal in industrial control, a wireless terminal in self-driving, a wireless terminal in remote medical surgery, a wireless terminal in a smart grid, a wireless terminal in transportation safety, a wireless terminal in a smart city, a wireless terminal in a smart home, etc. It should be understood that the embodiments of the present application do not limit the specific technology and specific device form adopted by the terminal device. In this embodiment and subsequent embodiments, the terminal device is used as an example for introduction.

[0067] Access network equipment can be any device with wireless transceiver capabilities and can be used to perform air interface-related functions, such as radio link maintenance, radio resource management, and some mobility management functions. The access network equipment can also be configured with a baseband unit (BBU) that has baseband signal processing capabilities. The access network equipment can be a radio access network (RAN) device (or RAN node) that currently provides services to terminal devices. Currently, some common examples of access network equipment include: Node B (NB), evolved Node B (eNB), next generation Node B (gNB) in 5G new radio (NR) systems, nodes in 6G systems (e.g., xNodeB), transmission reception point (TRP), radio network controller (RNC), base station controller (BSC), base transceiver station (BTS), home base station (e.g., home evolved NodeB or home Node B (HNB)), etc. In addition, in network structures such as cloud radio access network (CloudRAN) or open radio access network (ORAN), access network equipment can be a device including a centralized unit (CU) (also known as a control unit) and / or a distributed unit (DU). Among them, the RAN equipment including CU and DU splits the protocol layer of the gNB in ​​the NR system, placing the functions of some protocol layers in the CU for centralized control, and the functions of the remaining part or all of the protocol layers in the DU, which is centrally controlled by the CU. Multiple DUs can share one CU. The division of CU and DU can be divided according to the protocol stack.For example, one possible segmentation approach is to deploy the radio resource control (RRC), service data adaptation protocol (SDAP), and packet data convergence protocol (PDCP) layers in the CU, and the remaining radio link control (RLC), media access control (MAC), and physical layers (PHY) in the DU. The CU and DU are connected via the F1 interface. The CU represents the gNB to which it connects with the core network via the NG interface, and the CU represents the gNB to which it connects with other gNBs (or other CUs) via the Xn interface. In the actual deployment of traditional RAN equipment, in addition to the logical gNB composed of the CU and DU, the RAN equipment also includes the RU. The RU is a hardware unit that contains some PHY layer functions and / or antenna equipment. Optionally, the RU can be configured to be independent of the antenna device (for example, an antenna line device (ALD) (also known as an antenna linear device), or it can be integrated with the antenna device. For example, in a 5G NR system, the aforementioned RU may be an active antenna unit (AAU), that is, a processing unit that integrates a remote radio unit (RRU) (or a remote radio head (RRH)) and an antenna device. In some deployments, the RAN device mentioned in the embodiments of the present application may be a device including a CU or a DU; or, the RAN device is a device including a CU and a DU; or, the RAN device is a device including a control plane CU node (central unit control plane (central unit-control plane, CU-CP)) and a user plane CU node (central unit user plane (central unit-user RAN equipment includes equipment for the gNB-CU-CP, gNB-CU-UP, and gNB-DU. In other deployments, multiple RAN nodes collaborate to assist terminals in achieving wireless access, with different RAN nodes implementing portions of the base station's functionality. For example, a RAN node may be a CU, DU, CU-CP, CU-UP, or RU. The CU and DU may be separate or included in the same network element, such as a BBU. The RU may be included in a radio frequency device or radio frequency unit, such as an RRU, AAU, or RRH.In one possible design, the processing unit for implementing baseband functions in the BBU is called a baseband high layer (BBH) unit, and the processing unit for implementing baseband functions in the RRU / AAU / RRH is called a baseband low layer (BBL) unit. It should be understood that in different systems, CU (or CU-CP and CU-UP), DU or RU may have different names, but those skilled in the art can understand their meanings. For example, in the ORAN system, CU may also be called O-CU (Open CU), DU may also be called O-DU (Open DU), CU-CP may also be called O-CU-CP, CU-UP may also be called O-CU-UP, and RU may also be called O-RU. Any of the CU (or CU-CP, CU-UP), DU and RU in this application may be implemented by a software module, a hardware module, or a combination of a software module and a hardware module. It should be understood that the access network device in the embodiments of the present application may be any of the above-mentioned devices or a chip in the above-mentioned devices, and the specific details are not limited here. In this embodiment and subsequent embodiments, the access network device is mainly used as an example for description.

[0068] A compute node, also known as an in-network compute node, is a node in a wireless network responsible for processing AI tasks. The compute node is configured with various AI resources, such as AI model resources, AI computing resources, and AI data resources. Optionally, the compute node is deployed with an AI Server module to perform AI task processing. Optionally, the compute node is also deployed with a task management function (TMF) responsible for local scheduling of computing tasks. Exemplarily, the compute node may be a Node C or a Node I. Node C is a compute node that provides perception, computing, and other services in a mobile network. Node I is a dedicated node that provides data collection, model training, and management in a mobile network. Generally, the compute node is connected to an access network device, such as a gNB; or a CU; or a DU; or, the compute node is connected to both a CU and a DU, although this application is not restrictive. A compute node can be connected to multiple access network devices (e.g., a gNB, CU, or DU) and provide AI task processing capabilities for the multiple connected access network devices. An access network device can also be connected to multiple computing nodes so that multiple computing nodes can jointly provide AI task processing functions for an access network device. When there are multiple distributed computing nodes within the mobile network, each computing node communicates with each other through the access network device. The computing node in the embodiment of the present application can be independent of the access network device or integrated with the access network device, which is not limited by the present application. In this embodiment and subsequent embodiments, the computing node is mainly used as an example for introduction.

[0069] The management node is a new intelligent plane network element in the mobile network, which is used to respond to AI services and perform the orchestration and configuration of distributed AI tasks. For example, the management node can determine the appropriate computing node for the AI ​​task initiated by the access network device or the third-party AI application, and send the AI ​​task to each computing node so that each computing node can execute the AI ​​task in a distributed manner. Exemplarily, the management node can be an AI management function (AI Management Function, AIMF). The management node can be connected to the access network device, and the management node communicates with each computing node through the access network device; or, there is a communication interface between the management node and the computing node, and the management node communicates directly with each computing node. The management node in the embodiment of the present application can be deployed in the access network, for example, the management node can be independent of the access network device, or it can be integrated with the access network device; the management node can also be deployed in the core network, for example, the management node is the network element responsible for AI management in the core network. This application does not limit the specific implementation form of the management node.

[0070] In the system architecture shown in Figure 1, a management node receives task requests from access network devices or terminal devices (e.g., user terminals deployed with third-party AI applications) and, based on the task requests, selects an AI model with specific functionality and participating compute nodes for the task. The management node then sends the selected AI model to each compute node, enabling it to deploy the model, process the task data using it, and output the results of the AI ​​task.

[0071] However, the number of AI models with specific functions that can be stored in a management node is limited, and the task processing strategy for AI models with specific functions is fixed. The management node may not be able to find the appropriate AI model for the current AI task and deploy it on each participating computing node, which in turn affects the performance of the AI ​​service provided by each distributed computing node.

[0072] In this regard, the present application provides a configuration method, management node and computing node for distributed AI tasks, which are used to realize the generalization of the AI ​​model within the management node, so that the sub-model deployed by the management node to each computing node according to the task can constitute a target model suitable for processing the task, thereby improving the performance of each distributed computing node in providing AI services.

[0073] The following describes the main process of the distributed AI task configuration method provided by this application in conjunction with Figure 2.

[0074] As shown in Figure 2, a flowchart of a configuration method for a distributed AI task provided by the present application is provided. The configuration method of the distributed AI task is illustrated by taking the interaction between the management node and the computing node as an example. Of course, the subject that executes the management node action in the method can also be a device or module in the management node, such as a chip, processor or processing unit in the management node; the subject that executes the computing node action in the method can also be a device or module in the computing node, such as a chip, processor or processing unit in the computing node, etc., and the embodiment of the present application does not specifically limit this. The processing performed by a single execution subject (for example, a management node or a computing node) in the embodiment of the present application can also be divided into executions by multiple execution subjects, and these execution subjects can be logically and / or physically separated. Exemplarily, as shown in Figure 2, the configuration method of the distributed AI task includes the following steps:

[0075] Step 201: The management node receives a first task request.

[0076] Among them, the first task request is a request for applying for an AI service. Based on the different initiators of the application for the AI ​​service, the first task request may be from an access network device or from a third-party application of a terminal device. In one embodiment, the first task request comes from an access network device. For example, the access network device sends a first task request to the management node to request the AI ​​application service on the RAN side; accordingly, the management node receives the first task request from the access network device. In another embodiment, the first task request comes from the terminal device. For example, the terminal device is configured with a client of a third-party application, and the terminal device sends a first task request to the management node through the access network device to request the AI ​​service of the third-party application; accordingly, the management node receives the first task request from the terminal device through the access network device.

[0077] The first task request includes task information of the first task. Optionally, the task information of the first task includes a task type of the first task and a task requirement of the first task.

[0078] Among them, the task type of the first task refers to what type of AI task the first task is, that is, the task type of the AI ​​task. In one embodiment, if the first task request comes from an access network device, the task type of the AI ​​task is related to the physical layer application or high-level application of the access network device. For example, the task type of the first task can be any type of physical layer AI task such as beam management, channel prediction, or resource allocation; it can also indicate any type of high-level AI task such as terminal trajectory prediction, terminal load balancing, or terminal energy saving. In another embodiment, if the first task request comes from a third-party application of the terminal device, the task type of the AI ​​task is related to the application type of the third-party application. For example, the task type of the first task can indicate AI tasks such as navigation path planning and congestion prediction. It should be understood that in actual applications, the first task can also be other types of AI tasks, and this application does not limit the specific type of the first task.

[0079] In addition, the task requirements of the first task are used to indicate the demand (or requirement) for data processing and / or data transmission of the distributed AI processing service requested by the first task. The task requirements of the first task may be QoS requirements or performance requirements. Optionally, the task requirements of the first task include performance requirements of the AI ​​model (for example, the task processing accuracy that the model can provide, the scale of the model, the robustness, personalization or generalization of the model, etc.) and service efficiency requirements (for example, response delay, energy consumption, etc.). Optionally, the task requirements of the first task also include overhead requirements for providing AI services (for example, storage overhead, computing overhead, transmission overhead, etc.).

[0080] Step 202: The management node determines at least two sub-models and attribute information of the sub-models based on the task information of the first task and attribute information of multiple functional modules stored in the management node.

[0081] Among them, the management node stores multiple functional modules, a functional module can realize specific AI processing functions, and different functional modules realize different AI processing functions. A functional module includes at least one network layer, and the functions or parameters of different network layers are not exactly the same. Among them, the network layer refers to a layer in a deep learning model constructed layer by layer such as a neural network and a Transformer. For example, an image feature extraction functional module includes at least a convolution layer, and may include a pooling layer (for example, a max-pooling layer or an average-pooling layer), etc. For another example, a classification functional module generally includes a fully connected layer with an activation function (for example, a softmax layer or a sigmoid layer). It should be understood that the aforementioned neural network can be a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN) or other neural networks, etc., and this application is not limited.

[0082] Furthermore, a functional module is an AI model with a specific AI processing function; or, a functional module is a part of an AI model with a specific AI processing function. Multiple functional modules with different functions can constitute an AI model that processes a specific AI task; accordingly, an AI model that processes a specific AI task can be split into multiple different functional modules based on their functions.

[0083] For example, as shown in Figure 3A, an example diagram of an image classification AI model listed in this application is shown. In this example, the image classification AI model includes five functional modules: a low-level feature extraction module, a complex feature extraction module, an abstract feature extraction module, a feature fusion processing module, and a classification processing module. The low-level feature extraction module includes two larger convolutional layers (i.e., layers a1 and a2), one maximum pooling layer (i.e., layer b1), and two smaller convolutional layers (i.e., layers a3 and a4). The complex feature extraction module includes one mean pooling layer (i.e., layer c1) and three convolutional layers (i.e., layers a5, a6, and a7). The abstract feature extraction module includes one maximum pooling layer (i.e., layer b2), two convolutional layers (i.e., layers a8 and a9), one mean pooling layer (i.e., layer c2), and one smaller maximum pooling layer (i.e., layer b3). The feature fusion processing module includes two fully connected layers (i.e., layers d1 and d2). The classification processing module includes (i.e., layers e1 and e2). In this example, the low-level feature extraction module, the complex feature extraction module, the abstract feature extraction module, the feature fusion processing module, and the classification processing module constitute an image classification AI model.

[0084] In addition, the management node also stores the attribute information of each functional module. The attribute information of the functional module includes the function type of the functional module and the requirements of the functional module. Among them, the function type of the functional module is used to indicate what kind of AI processing function the functional module implements and the adaptation conditions for the functional module to implement the AI ​​processing function. In addition, the requirements of the functional module are used to indicate the requirements (or requirements) for data processing and / or data transmission when implementing the functions of the functional module. In some scenarios, the requirements of the functional module can be referred to as the performance requirements or QoS requirements of the functional module (for example, computing-transmission QoS requirements).

[0085] Exemplarily, if the management node stores a feature extraction function module, the attribute information of the feature extraction function module includes the function type of the feature extraction function module and the performance requirements of the feature extraction function module. The function type of the feature extraction function module indicates the feature extraction function, and the performance requirements of the feature extraction function module refer to the requirements (or demands) for performance indicators (or QoS indicators) such as training time, generalization performance, robustness, storage overhead, computing overhead, and energy consumption when implementing the feature extraction function.

[0086] Optionally, the attribute information of the functional module also includes information of each network layer contained in the functional module (for example, network layer parameters such as data dimension, quantization information, and pruning information of the network layer), the input dimension of the functional module (i.e., the dimension of the data input to the first network layer of the functional module) and the output dimension of the functional module (i.e., the dimension of the data output by the last network layer of the functional module), etc. Taking the complex feature extraction module in the example shown in Figure 3A as an example, the complex feature extraction module includes 1 mean pooling layer (i.e., layer c1) and 3 convolutional layers (i.e., layer a5, layer a6, and layer a7), then the attribute information of the complex feature extraction module includes network layer parameters such as data dimension, quantization information, and pruning information of layer c1, layer a5, layer a6, and layer a7. In addition, the attribute information of the complex feature extraction module also includes the input dimension of the complex feature extraction module (i.e., the dimension of the data input to layer c1) and the output dimension of the complex feature extraction module (i.e., the dimension of the data output by layer a7).

[0087] Optionally, the attribute information of the functional module also includes the compatibility of the computing node, which is used to indicate the requirements of the functional module on the hardware or software of the computing node of part or all of the network layer where the functional module is deployed.

[0088] In this step, after receiving the task information of the first task, the management node can determine what type of AI task the first task is and the performance requirements (or QoS requirements) of the AI ​​task based on the task information of the first task. Then, based on the task information of the first task and the attribute information of multiple functional modules stored by the management node, the management node filters out at least one functional module from the multiple functional modules to constitute at least two sub-models for distributed processing of the first task, and determines the attribute information of the sub-model based on the attribute information of the filtered functional modules.

[0089] A submodel is a model assigned to a distributed computing node for processing task data. A submodel can include at least one functional module or a portion of a functional module. The submodel's attribute information is used by the distributed computing node to deploy the submodel. The submodel's attribute information is related to the attribute information of the functional modules that comprise the submodel. These are described below:

[0090] In one possible embodiment, a sub-model includes at least one functional module. The attribute information of the sub-model is related to the attribute information of at least one functional module constituting the sub-model. For example, if sub-model 1 is composed of functional module A and functional module B, and the sub-model 1 can implement function A and function B, then the attribute information of sub-model 1 includes the attribute information of functional module A and the attribute information of functional module B. Taking Figure 3A as an example, sub-model 2 includes a complex feature extraction module and an abstract feature extraction module. The attribute information of sub-model 2 includes information of each network layer constituting the complex feature extraction module and information of each network layer constituting the abstract feature extraction module, namely, the relevant parameters of the mean pooling layer (i.e., layer c1) and the convolution layer (i.e., layer a5, layer a6, and layer a7) in the complex feature extraction module, as well as the relevant parameters of the maximum pooling layer (i.e., layer b2), the convolution layer (i.e., layer a8 and layer a9), the mean pooling layer (i.e., layer c2), and the smaller-sized maximum pooling layer (i.e., layer b3) in the abstract feature extraction module.

[0091] In another possible implementation, a submodel comprises a portion of a functional module. The attribute information of a submodel is related to all or part of the attribute information of the functional modules that comprise the submodel. For example, submodel 2 may comprise a portion of functional module C, and submodel 3 may comprise another portion of functional module C, with submodels 2 and 3 jointly implementing function C. In this example, the attribute information of submodel 2 includes all attribute information of functional module C or attribute information related to the portion of functional module C that comprises submodel 2, and the attribute information of submodel 3 includes all attribute information of functional module C or attribute information related to the other portion of functional module C that comprises submodel 3. Taking Figure 3B as an example, submodel 1 comprises a portion of a low-level feature extraction module. The attribute information of submodel 1 may include information about each network layer that comprises the low-level feature extraction module, such as layer a1, layer a2, layer b1, layer a3, and layer a4; alternatively, the attribute information of submodel 1 may include information about all network layers that comprise the low-level feature extraction module, such as layer a1, layer a2, layer b1, layer a3, and layer a4.

[0092] It should be understood that among the at least two sub-models determined by the management node based on at least one filtered functional module, one part of the sub-models may include at least one functional module, another part of the sub-models may only include a part of a functional module, and another part of the sub-models may include at least one functional module and a part of a functional module. This application does not limit this.

[0093] In addition, the attribute information of the sub-model is used to indicate the structure and function of the sub-model. Among them, the structure of the sub-model refers to the multiple network layers contained in the sub-model and the order between the multiple network layers. Specifically, the attribute information of the sub-model includes information about each network layer contained in the sub-model, and the information of the network layer includes the data input dimension and data output dimension of the network layer. Optionally, the attribute information of the sub-model also includes the function of the network layer contained in the sub-model, that is, which network layers in the sub-model are part of which functional module. Taking the sub-model 3 in the example shown in Figure 3A as an example, the sub-model 3 includes a mean pooling layer (i.e., layer c2) and a maximum pooling layer (i.e., b3) located in the abstract feature extraction module, 2 fully connected layers (i.e., layer d1 and layer d2) located in the feature fusion processing module, and 2 classification layers (i.e., layer e1 and layer e2) located in the classification processing module. The attribute information of sub-model 3 includes the information of the aforementioned network layers (i.e., the data input dimension and data output dimension of each network layer in layer c2, layer b3, layer d1, layer d2, layer e1, and layer e2), as well as the functional modules to which the network layers included in sub-model 3 belong (i.e., layers c2 and b3 belong to the abstract feature extraction module, layers d1 and d2 belong to the feature fusion processing module, and layers e1 and e2 belong to the classification processing module).

[0094] In addition, the at least two sub-models determined by the management node are used to be distributedly deployed in at least two computing nodes, with one sub-model deployed on each computing node, and different sub-models are deployed on different computing nodes. The at least two sub-models deployed on the at least two computing nodes constitute the target model for processing the first task. For example, if the management node determines sub-model 1, sub-model 2, and sub-model 3, and determines that sub-model 1, sub-model 2, and sub-model 3 are deployed on computing nodes 1, computing nodes 2, and computing nodes 3, respectively, then the aforementioned sub-model 1, sub-model 2, and sub-model 3 together constitute the target model for processing the first task.

[0095] Specifically, the management node first determines a target model for processing the first task based on the task information of the first task and the attribute information of multiple functional modules. Then, the management node determines at least two sub-models and the attribute information of each sub-model based on the target model, so that the at least two sub-models are distributedly deployed in at least one computing node. For example, based on the task information of the first task and the attribute information of multiple functional modules, the management node selects at least one functional module from multiple functional modules to form a target model, so that the target model can process the first task; then, the management node splits the target model into at least two sub-models, so that the at least two sub-models are distributedly deployed in at least two computing nodes, thereby enabling the sub-models located in different computing nodes to jointly process the first task.

[0096] It should be noted that when the management node splits the target model into at least two sub-models, the management node can split the model at the granularity of functional modules, for example, one sub-model includes at least one functional module; the management node can also split the model at the granularity of network layers, for example, one sub-model includes a portion of the network layer of at least one functional module. For details, please refer to the previous introduction to sub-models and will not be repeated here.

[0097] It should also be noted that when the management node splits the target model into at least two sub-models, the sub-model determined by the management node may include partially redundant network layers to improve the reliability and flexibility of the sub-model and facilitate subsequent functional enhancement or dynamic adjustment of the sub-model. Specifically, the management node may divide the same network layer into different sub-models, that is, different sub-models all contain at least one identical network layer. For example, the two sub-models determined by the management node to have a connection relationship each contain the same network layer of the same functional module. Exemplarily, as shown in Figure 3A, the management node determines that the target model for processing the first task includes a low-level feature extraction module, a complex feature extraction module, an abstract feature extraction module, a feature fusion processing module, and a classification processing module. Then, the management node splits the target model into three sub-models, namely sub-model 1 including the low-level feature extraction module and the complex feature extraction module, sub-model 2 including the complex feature extraction module and the abstract feature extraction module, and sub-model 3 including a feature fusion module, a classification processing module, and a portion of the abstract feature extraction module. Sub-model 1 and sub-model 2 both include the four network layers (i.e., layer c1, layer a5, layer a6, and layer a7) of the complex feature extraction module, and sub-model 2 and sub-model 3 both include network layers c2 and b3 of the abstract feature extraction module. In this embodiment, since the two connected sub-models have the same network layer, when the two sub-models are deployed on different computing nodes, the management node can enable the network layer of the sub-models based on the resource status of the computing nodes where the two sub-models are deployed, thereby improving the flexibility of sub-model deployment on computing nodes.

[0098] It should also be noted that when the management node splits the target model into at least two sub-models, the sub-models determined by the management node may not include redundant network layers, that is, different sub-models do not contain the same network layers, so as to save signaling overhead for transmitting sub-model attribute information. For example, as shown in FIG3B , the management node determines that the target model for processing the first task includes a low-level feature extraction module, a complex feature extraction module, an abstract feature extraction module, a feature fusion processing module, and a classification processing module. The management node then splits the target model into three sub-models: sub-model 1, which includes a portion of the network layer of the low-level feature extraction module; sub-model 2, which includes the complex feature extraction module and the abstract feature extraction module; and sub-model 3, which includes the feature fusion module and the classification processing module. Specifically, layer a3 in sub-model 1 is connected to layer a4 in sub-model 2, and layer b3 in sub-model 2 is connected to layer d1 in sub-model 3. Sub-models 1, 2, and 3 do not share the same network layers. In this embodiment, there is no identical network layer between the two sub-models with a connection relationship. When the management node sends the aforementioned two sub-models and the attribute information of the sub-models to at least two computing nodes, the management node can save the signaling overhead of sending the attribute information of the sub-models and the sub-models.

[0099] Optionally, the attribute information of the sub-model also includes the network layer position of the sub-model relative to the target model. It can be understood as which network layers of the target model the various network layers contained in the sub-model are; it can also be understood as which network layers in the target model the network layers contained in the sub-model are connected to. Exemplarily, taking the sub-model 2 shown in FIG3A as an example, the network layer position of the sub-model 2 relative to the target model is {a4, c1; b3, d1}, indicating that the layer c1 in the sub-model 2 is connected to the layer a4 in the target model, that is, the output of the layer a4 in the target model can be used as the input of the layer c1 in the sub-model 2; the layer b3 in the sub-model 2 is connected to the layer d1 in the target model, that is, the output of the layer b3 in the sub-model 2 can be used as the input of the layer d1 in the target model.

[0100] It should be noted that in traditional technology, management nodes generally store backbone models, that is, complete models for executing a certain AI task. Since the backbone models stored by management nodes in traditional technology are limited, the backbone models that the management node can select for the currently received AI task are also limited. Therefore, traditional technology may consider the matching of model functions, but ignore the matching of the performance requirements of the task and the performance requirements of the model. In this embodiment, the management node can screen out functional modules that can meet the task requirements of the first task based on the task requirements of the first task and the performance requirements of the functional modules, and then ensure that the target model composed of multiple functional modules can also meet the task requirements of the first task, thereby improving the matching rate of the management node configuration tasks and improving the performance of the AI ​​processing service.

[0101] In addition, in addition to determining at least two sub-models used to constitute the target model, the management node also needs to determine in which computing nodes the at least two sub-models are deployed, so as to ensure that the computing node that receives the sub-model and the attribute information of the sub-model has sufficient AI resources to deploy the sub-model and establish a connection relationship with the sub-models deployed by other computing nodes.

[0102] In one possible implementation, the management node obtains resource status information of at least one computing node, and the resource status information is used to indicate the usage status of the AI ​​resources of the computing node. Among them, AI resources include model resources (for example, AI models deployed by computing nodes), computing resources (for example, computing resources that are idle at computing nodes), and data resources (for example, data used for training or reasoning in computing nodes). Then, the management node determines the sub-model to be deployed for each computing node based on the resource status information of at least one computing node. Optionally, the AI ​​resources of the computing node for deploying the sub-model determined by the management node must meet the AI ​​resource requirements of the sub-model to be deployed. Exemplarily, if the management node determines to deploy sub-model 1 to computing node 1, computing node 1 can provide model resources for deploying sub-model 1, for example, the AI ​​model deployed in computing node 1 is compatible with sub-model 1; or, computing node 1 can provide computing resources for deploying sub-model 1, for example, the computing node for deploying sub-model 1 requires 1G of computing resources, then computing node 1 has at least 1G of idle computing resources for deploying sub-model 1; or, computing node 1 can provide data resources for deploying sub-model 1, for example, sub-model 1 requires training data of a specific size or a specific type of training data during operation, then computing node 1 stores the aforementioned training data of a specific size or a specific type of training data.

[0103] After the management node determines the sub-model to be deployed on each computing node and the attribute information of the sub-model, the management node will send the sub-model and the attribute information of the sub-model corresponding to the computing node to each computing node. Specifically, the management node will execute step 203.

[0104] In step 203, the management node sends the sub-models to be deployed and the attribute information of the sub-models to at least two computing nodes; accordingly, at least two computing nodes receive the sub-models to be deployed and the attribute information of the sub-models from the management node.

[0105] For example, the management node determines the first sub-model and the second sub-model, and determines the first computing node to deploy the first sub-model and the second computing node to deploy the second sub-model. The management node sends the first sub-model and the attribute information of the first sub-model to the first computing node, and sends the second sub-model and the attribute information of the second sub-model to the second computing node.

[0106] In step 204 , at least two computing nodes respectively deploy the sub-models they received.

[0107] Specifically, at least two computing nodes deploy the received sub-models based on the received attribute information of the sub-models. For example, after a first computing node receives the first sub-model and the attribute information of the first sub-model, the first computing node deploys the first sub-model based on the attribute information of the first sub-model, and the second computing node deploys the second sub-model based on the attribute information of the second sub-model.

[0108] It should be noted that after a computing node deploys a sub-model, the sub-model needs to be further configured so that it can be combined with sub-models deployed on other computing nodes to achieve distributed processing of the first task.

[0109] Step 205: The management node sends configuration information of the sub-model deployed by the computing nodes to at least two computing nodes; correspondingly, at least two computing nodes receive the configuration information of the sub-model deployed by the computing nodes from the management node.

[0110] The sub-model configuration information is used to configure the model parameters used by the sub-model when participating in processing the first task and to establish connections between the sub-model and sub-models deployed on different computing nodes. Optionally, the sub-model configuration information is also used to configure the interaction rules and transmission methods for AI data between nodes.

[0111] Optionally, the configuration information of the submodel includes parameters of an enabled network layer, and the parameters of the enabled network layer are used to indicate the network layer that the submodel deployed on the computing node enables when participating in processing the first task. The network layers enabled by at least two submodels constitute the target model. The parameters of the enabled network layer include: information of the starting network layer of the submodel (i.e., the network layer at the data input end of the multiple network layers enabled by the submodel), information of the exit network layer of the submodel (i.e., the network layer at the data output end of the multiple network layers enabled by the submodel), and information connecting the starting network layer and the exit network layer in the submodel.

[0112] Exemplarily, taking FIG3A as an example, the management node determines that submodel 1, submodel 2, and submodel 3 are deployed in computing node 1, computing node 2, and computing node 3, respectively, and the network layers enabled by submodel 1, submodel 2, and submodel 3 can just constitute the target model shown in FIG3A. In one example, if the management node determines that submodel 1 enables layers a1, a2, b1, a3, and a4, and determines that submodel 2 enables layers c1, a5, a6, a7, b2, a8, and a9, and determines that submodel 3 enables layers c2, b3, d1, d2, e1, and e2, then the configuration information of submodel 1 includes the parameters of each network layer enabled by submodel 1, namely, the parameters of layers a1, a2, b1, a3, and a4, and indicates that layer a1 is the starting network layer of submodel 1, and layer a4 is the starting network layer of submodel 1. The configuration information of submodel 2 includes the parameters of each network layer enabled by submodel 2, namely, the parameters of layer c1, layer a5, layer a6, layer a7, layer b2, layer a8 and layer a9, and indicates that layer c1 is the starting network layer of submodel 2, and indicates that layer a9 is the exit network layer of submodel 2; the configuration information of submodel 3 includes the parameters of each network layer enabled by submodel 3, namely, layer c2, layer b3, layer d1, layer d2, layer e1 and layer e2, and indicates that layer c2 is the starting network layer of submodel 3, and indicates that layer e2 is the exit network layer of submodel 3.

[0113] Optionally, the configuration information of the sub-model also includes the connection relationship between the sub-model and the sub-models deployed on other computing nodes. Exemplarily, the connection relationship between the sub-model and the sub-models deployed on other computing nodes includes: information about the next network layer (i.e., the network layer connected to the exit network layer of the sub-model and the computing node where the network layer is located), and / or information about the previous network layer (i.e., the network layer connected to the starting network layer of the sub-model and the computing node where the network layer is located).

[0114] For example, still taking Figure 3A as an example, submodel 1 enables {a1, layer a2, layer b1, layer a3, layer a4}, submodel 2 enables {layer c1, layer a5, layer a6, layer a7, layer b2, layer a8, layer a9}, and submodel 3 enables {layer c2, layer b3, layer d1, layer d2, layer e1, layer e2}. Submodel 1 deployed on computing node 1 is connected to submodel 2 deployed on computing node 2, and submodel 2 deployed on computing node 2 is connected to submodel 3 deployed on computing node 3. In this example, the configuration information of submodel 1 also includes information about computing node 2 and information about layer c1 in submodel 2, indicating that the exit network layer of submodel 1 (i.e., layer a4) is connected to the next network layer (i.e., layer c1). The configuration information for submodel 2 also includes information about computing node 1 and layer a4 in submodel 1, indicating that the starting network layer of submodel 2 (i.e., layer c1) is connected to the previous network layer (i.e., layer a4); and information about computing node 3 and layer c2 in submodel 3, indicating that the exit network layer of submodel 2 (i.e., layer a9) is connected to the next network layer (i.e., layer c2). The configuration information for submodel 3 also includes information about computing node 2 and layer a9 in submodel 2, indicating that the starting network layer of submodel 3 (i.e., layer c2) is connected to the previous network layer (i.e., layer a9).

[0115] Optionally, the sub-model configuration information is also used to configure the task processing strategy when the sub-model participates in processing the first task. This can be understood as the task processing strategy of the sub-model jointly processing the first task with other sub-models (for example, sub-models deployed on other computing nodes in at least two computing nodes determined by the management node).

[0116] Exemplarily, the configuration information of the sub-model includes at least one of the following task processing strategies:

[0117] The submodel strategy is used to indicate the model parameters used by the submodel when participating in processing the first task. Exemplarily, the submodel strategy includes a network layer enable parameter indicating which network layers of the submodel are enabled; an exit point parameter indicating which network layers of the submodel perform exit operations, and the task function or granularity corresponding to each exit point; and a submodel data dimension parameter indicating the data dimension of the input and / or output network layers of the submodel.

[0118] The data screening strategy is used to indicate the sub-model's strategy for filtering received data. The received data includes auxiliary data provided by the access network device and / or terminal device, training data or inference data pre-configured by the computing node, and task data for the first task. Exemplarily, the data screening strategy includes a screening strategy for auxiliary data provided by the base station and / or terminal device, sampling rules for training data, and filtering and pre-processing rules for auxiliary inference data.

[0119] Sub-model interaction strategies are used to indicate the interaction strategies between a sub-model and sub-models deployed on other computing nodes. For example, sub-model interaction strategies include model interaction strategies between sub-models, data interaction strategies between sub-models, regularization term interaction strategies between sub-models, and the proportional configuration of various interaction strategies between sub-models.

[0120] In this embodiment, the management node may determine the configuration information of the sub-model deployed by the computing node by any of the following implementation methods:

[0121] In one embodiment, the management node determines the configuration information of the sub-model deployed by each computing node based only on the task information of the first task.

[0122] In another type of implementation, the management node is able to obtain the network status information of the initiator of the first task (for example, an access network device or a terminal device), and then the management node determines the configuration information of the sub-model deployed by the computing node based on the task information of the first task and the network status information of the initiator of the first task.

[0123] In one implementation, a management node receives first network status information from an access network device, where the first network status information indicates an air interface status of the access network device connected to the computing node. The management node then determines configuration information for a sub-model to be deployed on the computing node based on resource status information of at least one computing node and the first network status information.

[0124] In another implementation, the management node receives second network status information from an access network device, where the second network status information indicates an air interface status of a terminal device that sends the first task request. The management node then determines configuration information for a sub-model to be deployed on the computing node based on resource status information of at least one computing node, the first network status information, and the second network status information.

[0125] It should be understood that the configuration information of different sub-models determined by the management node is not exactly the same. For example, if the management node determines two sub-models, namely the first sub-model and the second sub-model, and the two sub-models need to be deployed on the first computing node and the second computing node respectively, the management node sends the configuration information of the first sub-model deployed by the first computing node to the first computing node, and the configuration information of the first sub-model enables at least one network layer in the first sub-model; the management node sends the configuration information of the second sub-model deployed by the second computing node to the second computing node, and the configuration information of the second sub-model enables at least one network layer in the second sub-model. Since the first sub-model and the second sub-model are different sub-models, the configuration information of the first sub-model is not exactly the same as the configuration information of the second sub-model, and the network layer enabled in the first sub-model is different from the network layer enabled in the second sub-model, but there is a connection relationship between the network layer enabled in the first sub-model and the network layer enabled in the second sub-model.

[0126] Step 206: The computing node configures the sub-model and establishes a connection relationship between the sub-model and the sub-models deployed on different computing nodes.

[0127] Specifically, the computing node configures model parameters used by the sub-model when participating in processing the first task based on the received configuration information of the sub-model and establishes a connection relationship between the sub-model and sub-models deployed on different computing nodes.

[0128] For example, the computing node enables the network layer used by the sub-model when participating in processing the first task based on the parameters of the network layer enabled in the configuration information. Then, the computing node establishes a connection relationship between the sub-model and the sub-model deployed on other computing nodes based on the information of the next network layer and / or the information of the previous network layer of the sub-model, so that the sub-models deployed between different computing nodes can interact with each other.

[0129] Optionally, the computing node may further configure a task processing strategy for the sub-model when participating in processing the first task based on the sub-model's configuration information. For example, the computing node may configure a strategy for filtering received data by the sub-model based on a data filtering strategy. In another example, the computing node may configure an interaction strategy between the sub-model and sub-models deployed on other computing nodes based on a sub-model interaction strategy.

[0130] In this embodiment, the management node can determine the sub-models deployed to each computing node and the attribute information used to deploy each sub-model based on the task information and the information of the multiple functional modules stored by the management node, and the management node can also send the configuration information of the sub-model to the computing node based on the task information to configure the task processing strategy when the sub-model participates in processing the first task. Since the multiple functional modules stored by the management node can determine the sub-models used to constitute the target model, and the management node can send the configuration information of the sub-model for configuring the task processing strategy when the sub-model participates in processing the first task to the computing node where each sub-model is deployed based on the task information, the management node realizes the generalization of the AI ​​model through the stored functional modules, so that the sub-models deployed by the management node for each computing node according to the task can constitute a target model suitable for processing the task, thereby improving the performance of each distributed computing node in providing AI services.

[0131] The following describes the scenarios in which the management node configures distributed AI tasks initiated by access network devices and distributed AI tasks initiated by third-party applications.

[0132] Figure 4 is a flowchart of an embodiment of a method for configuring a distributed AI task provided by this application. In this embodiment, the first task is an AI task initiated by an access network device. In this case, the management node, computing node, and access network device will perform the following steps:

[0133] In step 401 , the access network device sends a first task request to the management node; correspondingly, the management node receives the first task request from the access network device.

[0134] The first task request includes task information of the first task. The task information of the first task includes the task type of the first task and the task requirements of the first task. For an introduction to the task type and task requirements of the first task, please refer to step 201 above and will not be repeated here.

[0135] In this embodiment, the first task is an AI task initiated by the access network device. For example, the first task can be any type of physical-layer AI task, such as beam management, channel prediction, or resource allocation. It can also be any type of high-layer AI task, such as terminal trajectory prediction, terminal load balancing, or terminal energy saving. It should be understood that in actual applications, the first task can also be other types of AI tasks, which are not detailed here.

[0136] Step 402: The management node determines a target model based on the task type, task requirements, and attribute information of multiple functional modules stored in the management node.

[0137] In one possible implementation, the management node determines a target model that supports processing the first task based on the task type, task requirements, and attribute information of multiple functional modules. The target model is composed of at least one functional module among the multiple functional modules, the task type of the first task is used to determine the model type of the target model, the performance requirements of the target model meet the task requirements of the first task, and the performance requirements of the target model are determined based on the requirements of at least one functional module. For example, the management node determines the model type of the target model based on the task type of the first task, and then the management node determines the multiple at least one functional module that can meet the task requirements of the first task based on the task requirements and the requirements of multiple functional modules, and splices them into a target model belonging to the model type. In this implementation, the management node can determine the target model that meets the task requirements of the first task based on the task type, task requirements, and functional module requirements of the first task, which is conducive to ensuring that the target determined by the management node needs to meet the AI ​​service requirements of the first task. In addition, the management node can form different backbone models by splicing multiple functional modules for different tasks, which is conducive to providing AI services with different performance requirements for different tasks.

[0138] It should be noted that the multiple functional modules stored in the management node can be pre-configured or generated by the management node based on the pre-configured backbone model. The following describes them separately:

[0139] In one possible implementation, the management node receives at least one functional module and attribute information of the functional module. The attribute information of the functional module is used to describe information such as the functional type and performance requirements of the functional module. For an introduction to the functional module and its attribute information, please refer to the relevant description in step 202 above and will not be repeated here.

[0140] In another possible embodiment, the management node receives at least one backbone model and attribute information of the backbone model. Then, the management node determines multiple functional modules based on the at least one backbone model, and determines the attribute information of each functional module based on the attribute information of the backbone model. The backbone model is a pre-trained artificial intelligence / machine learning (AI / ML) model, or an initial AI / ML model. A backbone model can be divided into multiple functional modules according to its function. A functional module is used to implement a sub-function of the backbone model, and different functional modules are used to implement different sub-functions of the backbone model or different precision / granularity of the same sub-function. In addition, the attribute information of the backbone model is used to indicate the function and structure of the backbone model. In this embodiment, the management node can divide the received backbone model into functional modules according to its function for storage, rather than directly storing the backbone model. This not only helps save the storage resources of the management node, but also makes it easier for the management node to generate sub-models using the functional modules, thereby increasing the diversity of sub-models that the management node can generate. For deployment nodes, through the interconnection between functional modules, multiple types of distributed AI paradigms can be flexibly supported.

[0141] Step 403: The management node obtains resource status information of at least two computing nodes.

[0142] The resource status information of the computing node is used to indicate the usage status of the AI ​​resources of the computing node. For an introduction to the resource status information of the computing node, please refer to step 202 above and will not be repeated here.

[0143] In an example, the management node sends status reporting indication information to at least two computing nodes managed by the management node, where the status reporting indication information is used to instruct the computing node to send resource status information of the computing node to the management node.

[0144] In another example, the computing node is pre-configured with a resource status reporting rule, and the computing node can periodically send its resource status information to the management node; or, when the resource status changes, the computing node sends updated resource status information to the management node.

[0145] Step 404: The management node determines at least two sub-models, attribute information of the sub-models, and at least two computing nodes for deploying the at least two sub-models based on the resource status information of the computing node and the target model.

[0146] Specifically, the management node determines how to split the target model into at least two sub-models based on the resource status information received from at least two computing nodes, on which computing nodes the at least two sub-models need to be deployed, and the attribute information of each of the at least two sub-models. The AI ​​resources of the computing nodes on which the sub-models are to be deployed, as determined by the management node, must meet the AI ​​resource requirements of the sub-models to be deployed.

[0147] In one implementation, the management node first splits the target model into at least two sub-models, and then, based on the resource status information of the at least two computing nodes received, the management node determines at least two computing nodes that can deploy the at least two sub-models. For example, the management node determines to split the target model into three sub-models, and then, based on the resource status information of the at least two computing nodes, the management node selects three computing nodes from the at least two computing nodes for respectively deploying the aforementioned three sub-models. If the management node determines to deploy sub-model 1 to computing node 1, sub-model 2 to computing node 2, and sub-model 3 to computing node 3, then the AI ​​resources of computing node 1 must meet the AI ​​resource requirements of sub-model 1, the AI ​​resources of computing node 2 must meet the AI ​​resource requirements of sub-model 2, and the AI ​​resources of computing node 3 must meet the AI ​​resource requirements of sub-model 3.

[0148] In another implementation, the management node first filters out multiple idle computing nodes from at least two computing nodes based on the resource status information of at least two computing nodes received, and then determines to split the target model into at least two sub-models based on the number of idle computing nodes, and the number of sub-models is less than or equal to the number of idle computing nodes. For example, the management node filters out two idle computing nodes (for example, computing node 1 and computing node 2) from at least two computing nodes based on the resource status information of at least two computing nodes received, and then the management node splits the target model into two sub-models (for example, sub-model 1 and sub-model 2). If the management node determines to deploy sub-model 1 to computing node 1 and sub-model 2 to computing node 2, the AI ​​resources of computing node 1 must meet the AI ​​resource requirements of sub-model 1, and the AI ​​resources of computing node 2 must meet the AI ​​resource requirements of sub-model 2.

[0149] It should be noted that after the management node determines at least two sub-models based on the target model, it also determines the attribute information of the sub-models based on the attribute information of the functional modules that make up the target model. For an introduction to the attribute information of the functional modules and the attribute information of the sub-models, please refer to step 202 above and will not be repeated here.

[0150] The following example uses the case where the management node determines to split the target model into two sub-models, a first sub-model and a second sub-model. These two sub-models need to be deployed on the first and second compute nodes, respectively. The management node then sends the sub-models to be deployed on each compute node to the two compute nodes. Specifically, the management node executes steps 405a and 405b.

[0151] Step 405a: The management node sends the first sub-model and the attribute information of the first sub-model to the first computing node; correspondingly, the first computing node receives the first sub-model and the attribute information of the first sub-model from the management node.

[0152] Step 405b: The management node sends the second sub-model and the attribute information of the second sub-model to the second computing node; correspondingly, the second computing node receives the attribute information of the second sub-model and the second sub-model from the management node.

[0153] Optionally, the management node sends the target model and the target model's attribute information to the first computing node. For example, the management node sends the target model, the target model's attribute information, the first sub-model, and the first sub-model's attribute information in a single message to the first computing node. For another example, the management node sends the target model, the target model's attribute information, the second sub-model, and the second sub-model's attribute information in a single message to the second computing node.

[0154] The attribute information of the target model includes information about the network layers contained in each of the at least two sub-models, and the connection relationships between the network layers of different sub-models in the at least two sub-models. When the first computing node obtains the attribute information of the target model and the attribute information of the first sub-model, the first computing node can determine which network layers in the target model the first sub-model is composed of. Similarly, when the second computing node obtains the attribute information of the target model and the attribute information of the second sub-model, the second computing node can determine which network layers in the target model the second sub-model is composed of.

[0155] Step 406a: The first computing node deploys the first sub-model.

[0156] Specifically, the first computing node deploys the first sub-model based on the attribute information of the first sub-model.

[0157] Step 406b: The second computing node deploys the second sub-model.

[0158] Specifically, the second computing node deploys the second sub-model based on the attribute information of the second sub-model.

[0159] Step 407: The access network device sends the first network status information to the management node; correspondingly, the management node receives the first network status information from the access network device.

[0160] The first network status information is used to indicate the air interface status of the access network device connected to the computing node, for example, the air interface status of the first access network device connected to the first computing node and the air interface status of the second access network device connected to the second computing node.

[0161] Step 408: The management node determines configuration information of the sub-model deployed on each computing node based on the resource status information of the at least two computing nodes and the first network status information.

[0162] Wherein, the at least two computing nodes include the aforementioned first computing node and the second computing node. For example, the management node determines the configuration information of the first sub-model deployed by the first computing node and the configuration information of the second sub-model deployed by the second computing node based on the resource status information of the first computing node, the resource status information of the second computing node and the first network status information. Wherein, the configuration information of the first sub-model is used to configure the model parameters used by the first sub-model when participating in the processing of the first task, and the configuration information of the second sub-model is used to configure the model parameters used by the second sub-model when participating in the processing of the first task. In addition, the configuration information of the first sub-model is also used to indicate the establishment of a connection relationship between the first sub-model and the second sub-model deployed on the second computing node, and the configuration information of the second sub-model is also used to indicate the establishment of a connection relationship between the second sub-model and the first sub-model deployed on the first computing node.

[0163] For an explanation of the configuration information of the sub-model, please refer to the relevant introduction in the previous step 205, which will not be repeated here.

[0164] Optionally, the management node will also determine the configuration information of at least one participating node. The participating node is a node that has a connection relationship with the computing node that deploys the sub-model. For example, the participating node is a node that has a connection relationship with the first computing node that deploys the first sub-model and / or the second computing node that deploys the second sub-model. The participating node does not deploy a sub-model, or the sub-model deployed by the participating node does not participate in processing the first task, and only forwards data between the first computing node and / or the second computing node. The configuration information of the participating node is used to indicate the data interaction strategy between the participating node and the computing node.

[0165] Step 409a: The management node sends the configuration information of the first sub-model to the first computing node; correspondingly, the first computing node receives the configuration information of the first sub-model from the management node.

[0166] In one possible implementation, the configuration information of the first sub-model includes first configuration information of the first sub-model. The first configuration information includes a first model parameter, and the first model parameter is used to indicate whether to enable the first network layer in the first sub-model, and the first network layer is at least one network layer among the multiple network layers indicated by the attribute information of the first sub-model. It can be understood that the first network layer is at least one network layer enabled among the multiple network layers deployed on the first computing node. In the initial configuration process, the management node sends the first configuration information of the first sub-model to the first computing node to indicate which network layers and which connection relationships are enabled by the first sub-model.

[0167] Optionally, the management node sends configuration information of the participating node to at least one participating node, where the configuration information of the participating node is used to indicate a data interaction strategy between the participating node and the computing node.

[0168] Step 409b: The management node sends the configuration information of the second sub-model to the second computing node; correspondingly, the second computing node receives the configuration information of the second sub-model from the management node.

[0169] Step 409b is similar to step 409a. Please refer to the relevant introduction in the previous step 409a for details, which will not be repeated here.

[0170] Step 410a: The first computing node configures a task processing strategy for the first sub-model.

[0171] After the first computing node receives the configuration information of the first sub-model, the first computing node configures the model parameters used when participating in processing the first task based on the configuration information of the first sub-model, and establishes a connection relationship between the first sub-model and the second sub-model deployed on the second computing node.

[0172] Step 410b: The second computing node configures a task processing strategy for the second sub-model.

[0173] After the second computing node receives the configuration information of the second sub-model, the second computing node configures the model parameters used when participating in processing the first task based on the configuration information of the second sub-model, and establishes a connection relationship between the second sub-model and the first sub-model deployed on the first computing node.

[0174] Step 411a: The access network device sends task data of the first task to the first computing node.

[0175] Step 411b: The access network device sends the task data of the first task to the second computing node.

[0176] It should be noted that the access network device can send first data requiring processing by the first sub-model to the first computing node and second data requiring processing by the second sub-model to the second computing node, respectively, where both the first data and the second data are task data for the first task. The access network device can also send the first data and the second data to one computing node (e.g., the first computing node), which then forwards the data to another computing node (e.g., the second computing node).

[0177] In step 412 , the first computing node and the second computing node jointly process the first task through the first sub-model and the second sub-model, and output a processing result of the first task.

[0178] For example, the first task can be physical layer beam management, channel prediction, and resource allocation, or high-level terminal trajectory prediction, load balancing, and energy saving. Taking the physical layer channel prediction as an example, the first computing node and the second computing node process the received task data of the first task using the first sub-model and the second sub-model, respectively, to obtain a processing result of the first task (e.g., a physical layer channel prediction result).

[0179] It should be noted that when the sub-model deployed by the computing node is insufficient to meet the requirements of the first task, the computing node can enhance the function or capability of the sub-model under the instruction of the management node, or the computing node can independently determine to enhance the function or capability of the sub-model. The following are introduced respectively:

[0180] In one possible implementation, the management node may send new configuration information to the computing node so that the computing node enhances the function or capability of the sub-model. Taking the first computing node as an example, if the software resources and / or hardware resources of the first computing node are in short supply after the sub-model is deployed, resulting in the first computing node being unable to meet the task requirements of the first task when processing the task data of the first task through the first sub-model, the management node sends the second configuration information to the first computing node. The second configuration information includes a second model parameter, and the second model parameter is used to indicate the enabling of the second network layer in the first sub-model. The second network layer is at least one network layer in the first computing node, and the second network layer is different from the first network layer. The management node modifies the configuration of the first sub-model by sending the second configuration information of the first sub-model to the first computing node, so that the first computing node uses the second configuration information to configure the first sub-model.

[0181] In one example, the second model parameter indicates all network layers that need to be enabled for the first sub-model, that is, the second network layer includes the first network layer. Exemplarily, taking FIG3A as an example, the management node indicates the first model parameter through the first configuration information, and the first model parameter indicates enabling layers a1, a2, b1, a3, a4, c1, and a5 in sub-model 1. When the management node determines that the first computing node needs to enhance sub-model 1, the management node indicates the second model parameter through the second configuration information, and the second model parameter indicates enabling layers a1, a2, b1, a3, a4, c1, a5, a6, and a7 in sub-model 1.

[0182] In another example, the second model parameter indicates a newly added network layer that needs to be enabled in the first sub-model, that is, the sum of the first network layer and the second network layer is the network layer ultimately enabled by the first sub-model. Exemplarily, taking FIG3A as an example, the management node indicates the first model parameter through the first configuration information, and the first model parameter indicates enabling layers a1, a2, b1, a3, a4, c1, and a5 in sub-model 1. When the management node determines that the first computing node needs to enhance sub-model 1, the management node indicates the second model parameter through the second configuration information, and the second model parameter indicates enabling layers a6 and a7 in sub-model 1.

[0183] In this embodiment, the computing node can modify the network layer enabled by the first sub-model under the instruction of the management node, thereby enhancing the sub-model, which is conducive to improving the flexibility and accuracy of the management node in configuring the sub-model for each computing node.

[0184] In another possible implementation, if the computing node stores the attribute information of the target model and the target model, the computing node can determine the function or capability of the enhanced sub-model based on the attribute information of the target model. Taking the first computing node as an example, if the software resources and / or hardware resources of the first computing node are in short supply after the sub-model is deployed, and the first computing node stores the target model and the attribute information of the target model, the first computing node determines the second network layer based on the attribute information of the target model and the first configuration information of the first sub-model, and enables the second network layer to enhance the first sub-model. For an explanation of the second network layer, please refer to the previous introduction and will not be repeated here.

[0185] In this embodiment, the computing node can autonomously modify the network layer enabled by the first sub-model based on the attribute information of the target model, thereby enhancing the sub-model, which is conducive to improving the sub-model in the computing node and improving the service quality of the AI ​​service.

[0186] Step 413: The first computing node sends a first task response to the access network device.

[0187] The first task response includes the processing result of the first task.

[0188] Optionally, the processing result of the first task can be sent to the access network device by only one computing node (for example, the first computing node) among at least two computing nodes, or can be sent to the access network device separately by at least two computing nodes, which is not limited here.

[0189] This embodiment provides a distributed AI configuration process in a RAN AI scenario. In this process, a management node can determine at least two sub-models that can meet the requirements of the first task based on the task requirements of the first task and the requirements of the functional modules, and deploy the at least two sub-models to at least two compute nodes. Furthermore, the management node also configures the at least two sub-models and orchestrates the tasks within the at least two sub-models based on information such as the task requirements of the first task. Because the aforementioned process fully considers the requirements and characteristics of the AI ​​task itself, the AI ​​resources of the compute nodes, the network status of the access network devices, and the requirements of the functional modules that the management node can provide, the management node achieves more reasonable distributed AI configuration and task allocation. Furthermore, this embodiment provides an architecture for implementing distributed intelligence endogenously in mobile networks. Through multi-level interconnection of models based on functional modules, a unified deployment framework for various distributed AI paradigms is provided. Multiple distributed AI paradigms can be integrated by setting the ratio of different interaction modes, thereby combining the advantages of multiple paradigms and improving the quality of AI services.

[0190] Figure 5 is a flowchart of an embodiment of a method for configuring a distributed AI task provided by this application. In this embodiment, the first task is an AI task initiated by a third-party application in the terminal device. In this case, the management node, computing node, access network device, and terminal device will perform the following steps:

[0191] Step 501: The terminal device sends a first task request; accordingly, the management node receives the first task request.

[0192] For example, the terminal device sends a first task request to the management node through the access network device; correspondingly, the management node receives the first task request sent by the terminal device through the access network device.

[0193] The first task request includes task information of the first task. The task information of the first task includes the task type of the first task and the task requirements of the first task. For an introduction to the task type and task requirements of the first task, please refer to step 201 above and will not be repeated here.

[0194] In this embodiment, the first task is an AI task initiated by a third-party application on the terminal device. For example, the first task is an AI task that instructs navigation path planning, congestion prediction, etc. It should be understood that in actual applications, the first task can also be other types of AI tasks, which are not detailed here.

[0195] Optionally, the first task request further includes resource status information of the terminal device, which is used to indicate the size of idle resources of the terminal device, so that the management node can determine whether a sub-model needs to be deployed in the terminal device.

[0196] Step 502: The management node determines a target model based on the task type, task requirements, and attribute information of multiple functional modules stored in the management node.

[0197] Step 503: At least two computing nodes send resource status information of the computing nodes to the management node; correspondingly, the management node receives the resource status information of the computing nodes from the at least two computing nodes.

[0198] Step 504: The management node determines at least two sub-models, attribute information of the sub-models, and at least two computing nodes for deploying the at least two sub-models based on the resource status information of the at least two computing nodes and the target model.

[0199] In this embodiment, steps 502 to 504 are similar to steps 402 to 404 described above. For details, please refer to the relevant descriptions in steps 402 to 404 described above, which will not be repeated here.

[0200] Optionally, if the terminal device is capable of supporting the deployment of sub-models and the management node can obtain the size of the terminal device's idle resources, the at least two computing nodes determined by the management node may include the terminal device. That is, the terminal device can serve as a computing node for deploying a certain sub-model (hereinafter referred to as the third sub-model). In this case, the size of the terminal device's idle resources must meet the resource requirements for deploying the third sub-model.

[0201] In this embodiment, the management node determines the first sub-model deployed on the first computing node and the second sub-model deployed on the second computing node based on the target model as an example. Optionally, the management node can also determine the third sub-model deployed on the terminal device.

[0202] Step 505a: The management node sends the first sub-model and the attribute information of the first sub-model to the first computing node; correspondingly, the first computing node receives the first sub-model and the attribute information of the first sub-model from the management node.

[0203] Step 505b: The management node sends the second sub-model and the attribute information of the second sub-model to the second computing node; correspondingly, the second computing node receives the attribute information of the second sub-model and the second sub-model from the management node.

[0204] In this embodiment, step 505a and step 505b are similar to steps 405a to 405b above. For details, please refer to the relevant descriptions in steps 405a to 405b above, which will not be repeated here.

[0205] Step 505c: The management node sends the third sub-model and the attribute information of the third sub-model to the terminal device; correspondingly, the terminal device receives the third sub-model and the attribute information of the third sub-model from the management node.

[0206] In this embodiment, step 505c is an optional step.

[0207] Step 506a: The first computing node deploys the first sub-model.

[0208] Specifically, the first computing node deploys the first sub-model based on the attribute information of the first sub-model.

[0209] Step 506b: The second computing node deploys the second sub-model.

[0210] Specifically, the second computing node deploys the second sub-model based on the attribute information of the second sub-model.

[0211] Step 506c: The terminal device deploys the sub-model.

[0212] In this embodiment, step 506c is an optional step.

[0213] Step 507: The access network device sends the first network status information to the management node; correspondingly, the management node receives the first network status information from the access network device.

[0214] In this embodiment, step 507 is similar to step 407 described above. For details, please refer to the relevant introduction in step 407 described above, which will not be repeated here.

[0215] Step 508: The terminal device sends the second network status information to the management node; correspondingly, the management node receives the second network status information from the terminal device.

[0216] The second network status information is used to indicate the air interface status of the terminal device that sends the first task request, for example, indicators such as the channel quality between the terminal device and the access network device.

[0217] Step 509: The management node determines configuration information of each sub-model based on the resource status information of at least two computing nodes, the first network status information, and the second network status information.

[0218] Wherein, the at least two computing nodes include the aforementioned first computing node and the second computing node. For example, the management node determines the configuration information of the first sub-model deployed by the first computing node and the configuration information of the second sub-model deployed by the second computing node based on the resource status information of the first computing node, the resource status information of the second computing node, and the first network status information and the second network status information. Wherein, the configuration information of the first sub-model is used to configure the model parameters used by the first sub-model when participating in the processing of the first task, and the configuration information of the second sub-model is used to configure the model parameters used by the second sub-model when participating in the processing of the first task. In addition, the configuration information of the first sub-model is also used to indicate the establishment of a connection relationship between the first sub-model and the second sub-model deployed on the second computing node, and the configuration information of the second sub-model is also used to indicate the establishment of a connection relationship between the second sub-model and the first sub-model deployed on the first computing node.

[0219] For an explanation of the configuration information of the sub-model, please refer to the relevant introduction in the previous step 205, which will not be repeated here.

[0220] Optionally, the management node will also determine the configuration information of at least one participating node. For an explanation of the configuration information of the participating nodes, please refer to the relevant introduction in step 409 above, which will not be repeated here.

[0221] Step 510a: The management node sends the configuration information of the first sub-model to the first computing node; correspondingly, the first computing node receives the configuration information of the first sub-model from the management node.

[0222] Step 510b: The management node sends the configuration information of the second sub-model to the second computing node; correspondingly, the second computing node receives the configuration information of the second sub-model from the management node.

[0223] Step 510c: The management node sends the configuration information of the third sub-model to the terminal device; correspondingly, the terminal device receives the configuration information of the third sub-model from the management node.

[0224] In this embodiment, step 510c is an optional step.

[0225] Optionally, the access network device sends auxiliary information of the first task to the first computing node and / or the second computing node; accordingly, the first computing node and / or the second computing node receives the auxiliary information of the first task sent by the access network device.

[0226] Among them, the auxiliary information is related to the task type of the first task. When the task type of the first task is different, the content of the auxiliary information is also different. Optionally, the auxiliary information includes the access network device's prediction information on the network environment and / or user behavior, and / or the access network device's perception information on the network environment and / or user behavior. Exemplarily, the auxiliary information is the user / air interface auxiliary information required for the third-party AI task. For example, the auxiliary information includes user profiles, behavior predictions, location / environmental perception and other information.

[0227] In this embodiment, the access network device provides auxiliary information of the first task to the first computing node and / or the second computing node, so that the first computing node and / or the second computing node uses the auxiliary information as input data of the AI ​​model, which is conducive to enhancing the performance of the sub-model in processing the first task, improving the capabilities of the sub-model, and providing personalized AI services based on user preferences.

[0228] Step 511a: The first computing node configures a task processing strategy for the first sub-model.

[0229] After receiving the configuration information of the first sub-model, the first computing node configures model parameters for use in processing the first task based on the configuration information of the first sub-model, and establishes a connection between the first sub-model and a second sub-model deployed on the second computing node. Optionally, the first computing node also establishes a connection between the first sub-model and a third sub-model deployed on the terminal device.

[0230] Step 511b: The second computing node configures a task processing strategy for the second sub-model.

[0231] After receiving the configuration information of the second sub-model, the second computing node configures model parameters for use when processing the first task based on the configuration information of the second sub-model, and establishes a connection between the second sub-model and the first sub-model deployed on the first computing node. Optionally, the second computing node also establishes a connection between the second sub-model and a third sub-model deployed on the terminal device.

[0232] Step 511c: The terminal device configures the task processing strategy of the third sub-model.

[0233] In this embodiment, step 511c is an optional step.

[0234] After the terminal device receives the configuration information of the third sub-model, the terminal device configures the model parameters used when participating in the processing of the first task based on the configuration information of the third sub-model, and establishes a connection relationship between the terminal device and the first sub-model deployed on the first computing node, and establishes a connection relationship between the terminal device and the second sub-model deployed on the second computing node.

[0235] Step 512a: The terminal device sends task data of the first task to the first computing node.

[0236] Step 512b: The terminal device sends the task data of the first task to the first computing node.

[0237] It should be noted that the terminal device can send first data that requires processing by the first sub-model to the first computing node, and send second data that requires processing by the second sub-model to the second computing node, where both the first data and the second data are task data for the first task. The terminal device can also send the first data and the second data to one computing node (for example, the first computing node), which will forward the data to another computing node (for example, the second computing node).

[0238] In step 513a, the first computing node and the second computing node jointly process the first task through the first sub-model and the second sub-model, and output the processing result of the first task.

[0239] For example, the first task may be an AI task such as navigation path planning or congestion prediction. Taking navigation path planning as an example, the first computing node and the second computing node process the received task data of the first task using the first sub-model and the second sub-model, respectively, to obtain a processing result of the first task (e.g., a navigation path planning result).

[0240] It should be noted that when the sub-model deployed by the computing node is insufficient to meet the requirements of the first task, the computing node may, under the instruction of the management node, enhance the function or capability of the sub-model, or the computing node may autonomously determine to enhance the function or capability of the sub-model. For details, please refer to the relevant description in step 412 above, which will not be repeated here.

[0241] In step 513b, the first computing node, the second computing node, and the terminal device jointly process the first task through the first sub-model, the second sub-model, and the third sub-model, and output the processing result of the first task.

[0242] In this embodiment, step 513b is optional. The terminal device participates in processing the first task only if the management node instructs the terminal device to deploy the third sub-model. Because the management node specifies the data interaction rules between the terminal device and the computing node (e.g., the first computing node or the second computing node) in the configuration information of the third sub-model, the terminal device can obtain intermediate data or the processing results of the first task.

[0243] Step 514: The first computing node sends a first task response to the terminal device.

[0244] The first task response includes the processing result of the first task.

[0245] In this embodiment, step 514 is an optional step. For example, when the terminal device has obtained the processing result of the first task while participating in the processing of the first task, the first computing node may not execute step 514.

[0246] In this embodiment, a distributed AI configuration process in a third-party AI scenario is provided. This process fully takes into account the needs and characteristics of the AI ​​task itself, the AI ​​resources of the computing node, the network status of the access network device, and the requirements of the functional modules that the management node can provide. Therefore, the management node implements a more reasonable distributed AI setting and task allocation. In addition, this embodiment supports the deployment of sub-models for terminal devices, so that the terminal and the computing node in the network can jointly complete the collaborative processing of third-party AI services, which helps to improve the flexibility and privacy protection of distributed AI applications. In addition, the computing node can also obtain network-perceived auxiliary information from the access network device, thereby improving the personalized service quality of third-party AI applications.

[0247] Figure 6 shows a schematic diagram of the structure of another apparatus 60 provided herein. It should be understood that the management node or computing node in the method embodiments corresponding to Figures 2, 4, or 5 described above can be based on the structure of the apparatus 60 shown in Figure 6 of this embodiment. As shown in Figure 6, the apparatus 60 may include a processor 601, a memory 603, and a communication interface 602. The processor 601 is coupled to the memory 603, which is in turn coupled to the communication interface 602.

[0248] The aforementioned communication interface 602 is connected to other devices via a communication link. For example, when the device 60 implements the function of a management node, the communication interface 602 may include an interface between the management node and an access network device, as well as an interface between the management node and at least one computing node. For another example, when the device 60 implements the function of a computing node, the communication interface 602 may include an interface between the computing node and an access network device, as well as an interface between the computing node and other computing nodes.

[0249] The processor 601 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or any combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof. The processor 601 may be a single processor or may include multiple processors, which is not specifically limited herein.

[0250] In addition, the aforementioned memory 603 is mainly used to store software programs and data. The memory 603 can exist independently and be connected to the processor 601. Optionally, the memory 603 can be integrated with the processor 601, for example, integrated into one or more chips. Among them, the memory 603 can store program codes for executing the technical solutions of the embodiments of the present application, and is controlled and executed by the processor 601. The various types of computer program codes executed can also be regarded as drivers for the processor 601. The memory 603 may include volatile memory (volatile memory), such as random-access memory (RAM); the memory may also include non-volatile memory (non-volatile memory), such as read-only memory (ROM), flash memory, hard disk drive (HDD) or solid-state drive (SSD); the memory 603 may also include a combination of the above types of memory. The memory 603 may refer to a single memory or may include multiple memories. Exemplarily, the memory 603 is used to store various data. For example, when the apparatus 60 implements the function of a management node, the memory 603 stores at least one functional module and attribute information of each functional module. For another example, when the apparatus 60 implements the function of a computing node, the memory 603 is used to store received sub-models, attribute information of the sub-models, and configuration information of the sub-models.

[0251] In one design, an apparatus 60 is used to execute the method for managing nodes in the embodiment corresponding to FIG2 . In the apparatus 60 , a communication interface 602 is used to receive a first task request, the first task request including task information of the first task; a processor 601 determines at least two sub-models and attribute information of the sub-models based on the task information and attribute information of multiple functional modules stored in the management node, the sub-model including at least one functional module and / or a portion of a functional module, the at least two sub-models being distributedly deployed in at least two computing nodes, the attribute information of the sub-model being determined based on the attribute information of the functional modules constituting the sub-model, and the attribute information of the sub-model being used for computing node deployment of the sub-model; the communication interface 602 sends the sub-models to be deployed by each computing node and the attribute information of the sub-model to the at least two computing nodes; and, based on the task information, sends configuration information of the sub-model deployed by the computing node to the at least two computing nodes, the configuration information of the sub-model being used to configure model parameters used by the sub-model when participating in processing the first task and to establish connection relationships between the sub-model and sub-models deployed on different computing nodes.

[0252] In a possible implementation, the attribute information of the functional module includes the functional type of the functional module and the requirements of the functional module. The requirements of the functional module are used to indicate the requirements for data processing and / or data transmission when implementing the functions of the functional module.

[0253] In one possible implementation, the task information of the first task includes the task type of the first task and the task requirement of the first task, where the task requirement of the first task is used to indicate the requirements of the distributed AI processing service requested by the first task for data processing and / or data transmission.

[0254] In one possible embodiment, the processor 601 is specifically used to determine a target model that supports processing a first task based on task information and attribute information of multiple functional modules, where the target model is composed of at least one functional module among the multiple functional modules, the task type of the first task is used to determine the model type of the target model, the performance requirements of the target model meet the task requirements of the first task, and the performance requirements of the target model are determined based on the requirements of at least one functional module; and, based on the target model, at least two sub-models and attribute information of each sub-model are determined.

[0255] In one possible implementation, the processor 601 is further configured to obtain resource status information of at least one computing node, where the resource status information indicates the usage status of the AI ​​resources of the computing node, where the AI ​​resources include model resources, computing resources, and data resources; and to determine at least two computing nodes based on the resource status information of the at least one computing node, and determine a sub-model to be deployed for each of the at least two computing nodes, where the AI ​​resources of the computing nodes meet the AI ​​resource requirements of the sub-model to be deployed by the computing nodes.

[0256] In one possible embodiment, the communication interface 602 is also used to receive network status information from an access network device connected to the computing node, and the network status information is used to indicate the network status of the communication device applying for the first task; the processor 601 is also used to determine the configuration information of the sub-model to be deployed by the computing node based on the task information, the resource status information of the computing node and the network status information; wherein the network status information includes first network status information and / or second network status information, the first network status information is used to indicate the air interface status of the access network device connected to the computing node, and the second network status information is used to indicate the air interface status of the terminal device sending the first task request.

[0257] In one possible implementation, the attribute information of the functional module also includes information of multiple network layers contained in the functional module, the input dimension of the functional module, and the output dimension of the functional module; the attribute information of the sub-model includes information of multiple network layers contained in the sub-model.

[0258] In a possible implementation, the configuration information of the sub-model includes parameters of an enabled network layer, where the parameters of the enabled network layer are used to indicate the network layer that the sub-model deployed on the computing node enables when participating in processing the first task.

[0259] In a possible implementation, the configuration information of the sub-model further includes a connection relationship between the sub-model and sub-models deployed on other computing nodes.

[0260] In one possible embodiment, the communication interface 602 is specifically used to send first configuration information of the first sub-model to the first computing node, the first configuration information includes a first model parameter, the first model parameter is used to indicate enabling of the first network layer in the first sub-model, and the first network layer is at least one network layer among the multiple network layers indicated by the attribute information of the first sub-model.

[0261] In one possible embodiment, the communication interface 602 is also used to send second configuration information of the first sub-model to the first computing node, the second configuration information includes second model parameters, the second model parameters are used to indicate enabling of the second network layer in the first sub-model, the second network layer is at least one network layer in the first computing node, and the second network layer is different from the first network layer.

[0262] In one possible embodiment, the communication interface 602 is also used to send the target model and attribute information of the target model to the first computing node, where the attribute information of the target model includes information of the network layers contained in each sub-model in at least two sub-models, and the connection relationship between the network layers of different sub-models in at least two sub-models.

[0263] In a possible implementation, the communication interface 602 is further configured to send configuration information of the participating node to at least one participating node, where the configuration information of the participating node is used to indicate a data interaction strategy between the participating node and the computing node.

[0264] In a possible implementation, the communication interface 602 is further configured to receive auxiliary information of the first task, where the auxiliary information includes prediction information of the access network device on the network environment and / or user behavior, and the auxiliary information is related to the task type of the first task.

[0265] It should be noted that the specific implementation and beneficial effects of this embodiment can refer to the method for managing nodes in the above embodiment, which will not be repeated here.

[0266] In another design, the device 60 is used to execute the method for the first computing node in the embodiment corresponding to Figure 4 or Figure 5. In the device 60, the communication interface 602 is used to receive the first sub-model of the first task and the attribute information of the first sub-model, the attribute information of the first sub-model is used to indicate the structure and function of the first sub-model; the processor 601 is used to deploy the first sub-model based on the attribute information of the first sub-model; the communication interface 602 is also used to receive the configuration information of the first sub-model, the configuration information of the first sub-model is used to configure the model parameters used by the first sub-model when participating in processing the first task and the connection relationship between the sub-models deployed on different computing nodes; the processor 601 is also used to configure the first sub-model based on the configuration information of the first sub-model, and establish the connection relationship between the first sub-model and the sub-models deployed on different computing nodes.

[0267] In a possible implementation, the attribute information of the first sub-model includes information of multiple network layers included in the first sub-model.

[0268] In a possible implementation, the configuration information of the first sub-model includes parameters of an enabled network layer, where the parameters of the enabled network layer are used to indicate the network layer that the first sub-model deployed on the first computing node enables when participating in processing the first task.

[0269] In a possible implementation, the configuration information of the first sub-model further includes a connection relationship between the first sub-model and sub-models deployed on other computing nodes.

[0270] In one possible implementation, the configuration information of the first sub-model includes first configuration information, the first configuration information includes a first model parameter, the first model parameter is used to indicate enabling of a first network layer in the first sub-model, the first network layer being at least one network layer among multiple network layers indicated by the attribute information of the first sub-model. The processor 601 is specifically configured to enable the first network layer in the first sub-model based on the first model parameter.

[0271] In one possible embodiment, the communication interface 602 is used to receive second configuration information of the first sub-model, the second configuration information includes second model parameters, the second model parameters are used to indicate enabling of the second network layer in the first sub-model, the second network layer is at least one network layer in the first computing node, and the second network layer is different from the first network layer; the processor 601 is used to enable the second network layer in the first sub-model based on the second model parameters.

[0272] In one possible embodiment, the communication interface 602 is also used to receive the target model and attribute information of the target model, where the attribute information of the target model includes information about the network layers contained in each sub-model in at least two sub-models, and the connection relationship between the network layers of different sub-models in at least two sub-models.

[0273] In a possible implementation, the communication interface 602 is further used to send resource status information of the first computing node, where the resource status information is used to indicate the usage status of the AI ​​resources of the first computing node, where the AI ​​resources include model resources, computing resources, and data resources.

[0274] It should be noted that the specific implementation and beneficial effects of this embodiment can refer to the method of the first computing node in the above embodiment, which will not be repeated here.

[0275] As shown in FIG7 , the present application further provides an apparatus 70. The apparatus 70 may be a management node or a computing node, or a component of a management node or a computing node (e.g., an integrated circuit, a chip, etc.). The apparatus 70 may also be other functional modules for implementing the methods in the method embodiments of the present application.

[0276] The apparatus 70 may include a processing module 701 (or processing unit). Optionally, it may also include an interface module 702 (or transceiver unit or transceiver module) and a storage module 703 (or storage unit). The interface module 702 is used to communicate with other devices. For example, the interface module 702 may be a transceiver module or an input / output module.

[0277] In one possible design, one or more modules in FIG7 may be implemented by one or more processors, or by one or more processors and memory, or by one or more processors and transceivers, or by one or more processors, memory, and transceivers, which are not limited in this embodiment of the present application. The processor, memory, and transceiver may be provided separately or integrated.

[0278] The device 70 has the function of implementing the management node described in the embodiment of the present application. For example, the device 70 includes a module or unit or means (means) corresponding to the management node execution step described in the embodiment of the present application. The function or unit or means can be implemented by software, or by hardware, or by hardware executing the corresponding software implementation, or by a combination of software and hardware. For details, please refer to the corresponding description in the aforementioned corresponding method embodiment. Please refer to the relevant description in the aforementioned method embodiment for details, which will not be repeated here.

[0279] Alternatively, the device 70 has the function of implementing the computing node described in the embodiment of the present application. For example, the device 70 includes a computing node that executes the computing node described in the embodiment of the present application. The module or unit or means (means) corresponding to the steps involved in the computing node, the function or unit or means can be implemented by software, or by hardware, or by hardware executing the corresponding software implementation, or by a combination of software and hardware. For details, please refer to the corresponding description in the aforementioned corresponding method embodiment. Please refer to the relevant description in the foregoing method embodiment for details, which will not be repeated here.

[0280] In addition, the present application provides a computer program product comprising one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. For example, the method related to the management node in Figure 2, Figure 4 or Figure 5 is implemented. For another example, the method related to the computing node in Figure 2, Figure 4 or Figure 5 is implemented. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be stored by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital versatile disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0281] In addition, the present application also provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the management node-related method as shown in Figure 2, Figure 4 or Figure 5 above.

[0282] In addition, the present application also provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method related to the computing node as shown in Figure 2, Figure 4 or Figure 5 above.

[0283] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0284] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

Claims

1. A method for configuring distributed AI tasks, applied to management nodes, characterized in that: include: receiving a first task request, wherein the first task request includes task information of the first task; determining at least two sub-models and attribute information of the sub-models based on the task information and attribute information of a plurality of functional modules stored in the management node, wherein the sub-models include at least one of the functional modules and / or a portion of one of the functional modules, the at least two sub-models are configured to be distributedly deployed in at least two computing nodes, the attribute information of the sub-models is determined based on attribute information of the functional modules constituting the sub-models, and the attribute information of the sub-models is used for deploying the sub-models on the computing nodes; Sending the sub-models to be deployed by the respective computing nodes and the attribute information of the sub-models to the at least two computing nodes; Based on the task information, configuration information of the sub-model deployed by the computing node is sent to the at least two computing nodes, where the configuration information of the sub-model is used to configure model parameters used by the sub-model when participating in processing the first task and to establish a connection relationship between the sub-model and sub-models deployed on different computing nodes.

2. The method according to claim 1, characterized in that The attribute information of the functional module includes the function type of the functional module and the requirements of the functional module. The requirements of the functional module are used to indicate the requirements for data processing and / or data transmission when implementing the functions of the functional module.

3. The method according to claim 1 or 2, characterized in that The task information of the first task includes a task type of the first task and a task requirement of the first task. The task requirement of the first task is used to indicate the requirements of the distributed AI processing service requested by the first task for data processing and / or data transmission.

4. The method according to claim 3, characterized in that The determining of at least two sub-models and the attribute information of the sub-models based on the task information and the attribute information of the plurality of functional modules stored in the management node includes: Determining a target model that supports processing the first task based on the task information and the attribute information of the multiple functional modules, the target model being composed of at least one functional module among the multiple functional modules, the task type of the first task being used to determine a model type of the target model, the performance requirement of the target model satisfying the task requirement of the first task, the performance requirement of the target model being determined based on the requirement of the at least one functional module; The at least two sub-models and attribute information of each of the sub-models are determined based on the target model.

5. The method according to any one of claims 1 to 4, characterized in that The attribute information of the sub-model includes the AI ​​resource requirements of the sub-model; Before sending the sub-models to be deployed and the attribute information of the sub-models of the respective computing nodes to the at least two computing nodes, the method further includes: Obtaining resource status information of at least one computing node, where the resource status information indicates a usage status of AI resources of the computing node, where the AI ​​resources include model resources, computing resources, and data resources; Based on the resource status information of the at least one computing node, the at least two computing nodes are determined, and the sub-model to be deployed of each of the at least two computing nodes is determined, and the AI ​​resources of the computing nodes meet the AI ​​resource requirements of the sub-model to be deployed on the computing nodes.

6. The method according to claim 5, characterized in that The method further comprises: receiving network status information from an access network device connected to the computing node, the network status information being used to indicate a network status of a communication device applying for the first task; Determining configuration information of a sub-model to be deployed on the computing node based on the task information, the resource status information of the computing node, and the network status information; In which, the network status information includes first network status information and / or second network status information, the first network status information is used to indicate the air interface status of the access network device connected to the computing node, and the second network status information is used to indicate the air interface status of the terminal device that sends the first task request.

7. The method according to claim 6, characterized in that The attribute information of the functional module also includes information of multiple network layers contained in the functional module, the input dimension of the functional module and the output dimension of the functional module; the attribute information of the sub-model includes information of multiple network layers contained in the sub-model.

8. The method according to any one of claims 1 to 7, characterized in that The configuration information of the sub-model includes parameters of an enabled network layer, and the parameters of the enabled network layer are used to indicate the network layer that the sub-model deployed on the computing node enables when participating in processing the first task.

9. The method according to claim 8, characterized in that The configuration information of the sub-model also includes the connection relationship between the sub-model and the sub-models deployed on other computing nodes.

10. The method according to any one of claims 1 to 9, characterized in that The sending, based on the task information, to the at least two computing nodes, configuration information of the sub-model deployed by the computing nodes, includes: Send first configuration information of the first sub-model to the first computing node, the first configuration information includes a first model parameter, the first model parameter is used to indicate enabling a first network layer in the first sub-model, and the first network layer is at least one network layer among multiple network layers indicated by the attribute information of the first sub-model.

11. The method according to claim 10, characterized in that The method further comprises: Send second configuration information of the first sub-model to the first computing node, the second configuration information includes second model parameters, the second model parameters are used to indicate enabling of the second network layer in the first sub-model, the second network layer is at least one network layer in the first computing node, and the second network layer is different from the first network layer.

12. The method according to any one of claims 4 to 11, characterized in that The method further comprises: The target model and attribute information of the target model are sent to the first computing node, wherein the attribute information of the target model includes information of the network layer contained in each sub-model of the at least two sub-models, and the connection relationship between the network layers of different sub-models of the at least two sub-models.

13. The method according to any one of claims 1 to 12, characterized in that The method further comprises: The configuration information of the participating node is sent to at least one participating node, where the configuration information of the participating node is used to indicate a data interaction strategy between the participating node and the computing node.

14. The method according to any one of claims 1 to 13, characterized in that The method further comprises: Auxiliary information of the first task is received, where the auxiliary information includes prediction information of the access network device on the network environment and / or user behavior, and the auxiliary information is related to the task type of the first task.

15. A method for configuring a distributed AI task, applied to a first computing node, characterized in that: include: receiving a first sub-model of a first task and attribute information of the first sub-model, where the attribute information of the first sub-model is used to indicate a structure and a function of the first sub-model; deploying the first sub-model based on the attribute information of the first sub-model; receiving configuration information of the first sub-model, where the configuration information of the first sub-model is used to configure model parameters used by the first sub-model when participating in processing the first task and connection relationships between sub-models deployed on different computing nodes; The first sub-model is configured based on the configuration information of the first sub-model, and a connection relationship is established between the first sub-model and sub-models deployed on different computing nodes.

16. The method according to claim 15, characterized in that The attribute information of the first sub-model includes information of multiple network layers included in the first sub-model.

17. The method according to claim 15 or 16, characterized in that The configuration information of the first sub-model includes parameters of an enabled network layer, where the parameters of the enabled network layer are used to indicate the network layer that the first sub-model deployed on the first computing node enables when participating in processing the first task.

18. The method according to claim 17, characterized in that The configuration information of the first sub-model also includes a connection relationship between the first sub-model and sub-models deployed on other computing nodes.

19. The method according to any one of claims 15 to 18, characterized in that The configuration information of the first sub-model includes first configuration information, the first configuration information includes a first model parameter, the first model parameter is used to indicate enabling of a first network layer in the first sub-model, the first network layer being at least one network layer among the multiple network layers indicated by the attribute information of the first sub-model; The deploying the first sub-model based on the attribute information of the first sub-model includes: The first network layer in the first sub-model is enabled based on the first model parameters.

20. The method according to any one of claims 15 to 19, characterized in that The method further comprises: Receiving second configuration information of the first sub-model, the second configuration information including a second model parameter, the second model parameter being used to indicate enabling a second network layer in the first sub-model, the second network layer being at least one network layer in the first computing node, and the second network layer being different from the first network layer; The second network layer in the first sub-model is enabled based on the second model parameters.

21. The method according to any one of claims 15 to 20, characterized in that The method further comprises: A target model and attribute information of the target model are received, wherein the attribute information of the target model includes information of a network layer included in each of at least two sub-models, and a connection relationship between network layers of different sub-models in the at least two sub-models.

22. The method according to any one of claims 15 to 21, characterized in that The method further comprises: Send resource status information of the first computing node, where the resource status information is used to indicate the usage status of AI resources of the first computing node, where the AI ​​resources include model resources, computing resources, and data resources.

23. A management node, characterized in that: including processor and memory; wherein the memory stores a computer program; The processor calls the computer program to enable the management node to perform the method according to any one of claims 1 to 14.

24. A computing node, characterized in that: including processor and memory; wherein the memory stores a computer program; The processor calls the computer program to cause the computing node to execute the method according to any one of claims 15 to 22.

25. A computer-readable storage medium storing instructions, which, when executed on a computer, causes the computer to execute the method according to any one of claims 1 to 14; or to execute the method according to any one of claims 15 to 22.

26. A network system, characterized in that: include: A management node that performs the method according to any one of claims 1 to 14, and a management node that performs the method according to any one of claims 15 to 22.

Images