Communication method, communication node, storage medium and program product

Abstract

The invention provides a communication method, a communication node, a storage medium and a program product. The method comprises the steps that a training request sent by second equipment is received, the training request comprises training demand information, and the training demand information is used for indicating the demand of a training task; performing model training according to the training request; the problem that model training cannot meet actual requirements is solved, the second device sends the training request, the training request carries the training requirement information to indicate the requirement of the training task of the first device, the first device performs targeted training according to the training requirement information, the requirement of the training task is met, and the training efficiency is improved. The actual training task requirement is considered in time when the first equipment carries out model training, the trained model can meet the requirement, the effectiveness of model training is improved, and resources are saved.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a communication method, a communication node, a storage medium, and a program product. Background Art

[0002] 3GPP SA5 is studying and finalizing the AI / machine learning (ML) management specifications for Rel-18. For example, these specifications will enable and manage AI / ML capabilities across various areas of the 5G system. The AI / ML management functions defined in SA5 Rel-18 include the management and operation of ML training, ML testing, AI / ML simulation, ML entity deployment, and AI / ML inference.

[0003] While AI / ML offers numerous benefits for 5G, there are also challenges. For example, the requirements for AI / ML solutions and capabilities vary significantly depending on the complexity of the ML model's tasks and structure. Existing technologies often fail to consider actual requirements during model training, resulting in models that fail to meet these requirements. Summary of the Invention

[0004] The present application provides a communication method, a communication node, a storage medium, and a program product to solve the problem that actual needs cannot be met during model training.

[0005] To achieve the above objectives, an embodiment of the present application provides a communication method, applied to a first device, comprising:

[0006] receiving a training request sent by a second device, the training request including training requirement information, the training requirement information being used to indicate requirements for the current training task;

[0007] Perform model training according to the training request.

[0008] To achieve the above-mentioned object, an embodiment of the present application provides another communication method, which is applied to a second device and includes:

[0009] Generate a training request, wherein the training request includes training requirement information, and the training requirement information is used to indicate the requirements of the current training task;

[0010] The training request is sent to the first device, so that the first device performs model training according to the training request.

[0011] To achieve the above-mentioned purpose, an embodiment of the present application provides a communication node, comprising: a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for realizing connection communication between the processor and the memory. When the program is executed by the processor, the steps of the communication method described in any one of the embodiments of the present application are implemented.

[0012] To achieve the above-mentioned purpose, an embodiment of the present application provides a storage medium for computer-readable storage, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the communication method described in any one of the embodiments of the present application.

[0013] To achieve the above-mentioned purpose, an embodiment of the present application provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the communication method described in any one of the embodiments of the present application.

[0014] The communication method, communication node, storage medium and program product provided in the embodiments of the present application, the first device receives a training request sent by the second device, the training request includes training requirement information, the training requirement information is used to indicate the requirements of this training task, and model training is performed according to the training request, which solves the problem that model training cannot meet actual needs. The second device sends a training request and carries the training requirement information in the training request to indicate the requirements of this training task of the first device. The first device performs targeted training according to the training requirement information to meet the requirements of the training task. The first device considers the actual training task requirements in a timely manner when performing model training. The trained model can meet the requirements, thereby improving the effectiveness of model training, avoiding repeated training, and saving time.

[0015] With respect to the above embodiments and other aspects of the present application and their implementation, further description is provided in the accompanying drawings, detailed description and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a service-oriented management architecture provided by an embodiment;

[0017] Figure 2 A flowchart of a communication method provided by an embodiment;

[0018] Figure 3 A flowchart of another communication method provided by an embodiment;

[0019] Figure 4 A timing diagram of communication between a first device and a second device provided in one embodiment;

[0020] Figure 5 Another timing diagram of communication between a first device and a second device provided in an embodiment;

[0021] Figure 6 A schematic structural diagram of a communication device provided by an embodiment;

[0022] Figure 7 A schematic structural diagram of another communication device provided by an embodiment;

[0023] Figure 8 A schematic structural diagram of a communication node provided by an embodiment. DETAILED DESCRIPTION

[0024] To make the purpose, technical solutions and advantages of this application more clear, the embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of this application can be combined with each other in any way.

[0025] Figure 1 This is a schematic diagram of a service-based management architecture provided in one embodiment, wherein the service-based management architecture (Service Based Management Architecture) includes a business support system (BSS), a cross-domain management function unit (Cross Domain Management Function, CD-MnF), a domain management function unit (Domain Management Function, Domain-MnF) and a network element (NE). The cross-domain management function unit is used to manage one or more domain management function units. The domain management function unit can be used to manage one or more network elements. The following is a brief introduction to each unit.

[0026] Business support systems (BSSs) are oriented towards communication services and provide functions and management services such as billing, settlement, accounting, customer service, sales, network monitoring, communication service lifecycle management, and business intent translation. BSSs can be either carrier operations systems or vertical industry operations systems (OTSystems).

[0027] The cross-domain management function unit is also called the Network Management Function (NMF), which can be a network management entity such as the Network Management System (NMS), the Network Management Service Producer (MnS Producer), the Network Management Service Consumer (MnS Consumer), and the Network Function Management Service Consumer (NFMS_C). Among them, the cross-domain management function unit provides one or more of the following management functions or management services: network lifecycle management, network deployment, network fault management, network performance management, network configuration management, network assurance, network optimization functions, and translation of the network intent (intent from communication service provider, Intent-CSP) of the service producer. The network referred to in the above-mentioned management functions or management services may include one or more network elements or subnetworks, or it may be a network slice. That is to say, the network management function unit can be a network slice management function unit (Network Slice Management Function, NSMF), or a cross-domain management data analysis function unit (Management Data Analytical Function, MDAF), or a cross-domain self-organizing network function (Self-Organization Network function, SON Function) or a cross-domain intent management function unit (Intent-Driven Management Service, Intent Driven MnS).

[0028] The domain management functional unit is also called the Network Subnet Management Function (NSMF) or the network element management functional unit. It can be a network element management entity such as the wireless automation engine (MBB automation engine, MAE), the element management system (Element Management System, EMS), the network function management service provider (NFMS_P), the network slice subnet management functional unit (NSSMF), the domain management data analysis functional unit (Domain MDAF), the domain self-organization network function (SON Function), the domain intent management functional unit, the MnS Producer, and the MnS Consumer. Domain management function units can be categorized as follows: by network type, they can be divided into: radio access network (RAN) domain management function unit (RAN domain MnF), core network domain management function unit (CN domain MnF), transport network domain management function unit (TN domain MnF), etc. It should be noted that the domain management function unit can also be a domain network management system that can manage one or more of the access network, core network, or transport network; by administrative region, they can be divided into: domain management function unit for a certain region, such as the domain management function unit of city A, the domain management function unit of city B, etc. The domain management function unit provides one or more of the following functions or management services: lifecycle management of subnetworks or network elements, deployment of subnetworks or network elements, fault management of subnetworks or network elements, performance management of subnetworks or network elements, assurance of subnetworks or network elements, optimization of subnetworks or network elements, and translation of intents from network operators (Intent-NOPs) of subnetworks or network elements. The subnetwork here includes one or more network elements.A subnetwork can also include subnetworks, that is, one or more subnetworks form a larger subnetwork. The subnetwork here can also be a network slice subnetwork.

[0029] A network element is an entity that provides network services. Network elements may include core network elements, radio access network elements, or transport network elements. Specifically, core network elements may include, but are not limited to, Access and Mobility Management Function (AMF) entities, Session Management Function (SMF) entities, Policy Control Function (PCF) entities, Network Data Analysis Function (NWDAF) entities, Network Repository Function (NRF) entities, gateways, and the like. The radio access network network element may include but is not limited to: various base stations (such as the next generation base station (Generation Node B, gNB), evolved base station (Evolved Node B, eNB), Central Unit Control Panel (Central Unit Control Panel, CUCP), Central Unit (Central Unit, CU), Distributed Unit (Distributed Unit, DU), Centralized User Plane Unit (Central Unit User Panel, CUUP), etc. In this application, the network function NF is also referred to as the network element NE. The network element can provide one or more of the following management functions or management services: network element lifecycle management, network element deployment, network element fault management, network element performance management, network element assurance, network element optimization function, and network element intent translation.

[0030] The cross-domain management functional unit can be used for model training of AI / ML models and model reasoning of AI / ML models; the domain management functional unit can be used for model training of AI / ML models and model reasoning of AI / ML models; the network element is an entity that provides network services, including core network network elements and access network network elements, which can provide at least one of model training of AI / ML models and model reasoning of AI / ML models.

[0031] The AI / ML management function and service framework involves the following:

[0032] ML Entity: A manageable artifact of an ML model. An ML entity may contain metadata related to the model, which may include, for example, the applicable runtime context of the ML model.

[0033] ML models: Mathematical algorithms that can be “trained” using data and human expert input as examples to replicate the decisions that experts would make when provided with the same information.

[0034] ML model training: The process performed by the ML training function to obtain training data, run that data through the ML model, infer the associated loss, and adjust the parameterization of the ML model based on the calculated loss.

[0035] ML initial training: ML model training to generate the initial version of the ML entity.

[0036] ML Retraining: The process of retraining a previously trained ML model.

[0037] The new version of the trained ML entity supports the same type of inference as the previous version of the ML entity, that is, the data type of the inference input and the data type of the inference output remain unchanged between the two versions of the ML entity, but the parameter values ​​of the retrained model may be different.

[0038] ML federated training: ML training is performed on a set of ML models that are trained and targeted for inference.

[0039] ML Training: Refers to the end-to-end process that enables the ML training function to perform initial training or retraining of ML models (as described above). ML training may include interactions with other parties to collect and format the data required for ML model training.

[0040] ML training function: A logical function with ML model training capabilities; also known as an MLT function.

[0041] AI / ML inference: Refers to the process of running a set of input data through a trained ML entity to generate a set of output data (e.g., predictions).

[0042] AI / ML reasoning function: logical functions that use ML models for reasoning.

[0043] The AI / ML workflow involves four main phases: training, simulation, deployment, and inference.

[0044] Training phase:

[0045] ML training: The training of one or a set of ML models, including initial training and retraining. It also includes validation of the ML entity to evaluate its performance when executed on training and validation data. If the validation results do not meet expectations (for example, the variance is unacceptable), the ML model associated with the entity needs to be retrained. ML model training is the initial stage of the workflow.

[0046] ML testing: Test the validated ML entity to evaluate the performance of the trained ML model when executed on the test data. If the test results meet expectations, the ML entity can move to the next stage, otherwise the ML model associated with the entity may need to be retrained.

[0047] Simulation phase:

[0048] ML simulation: Runs ML entities in a simulation environment for inference. The goal is to evaluate the inference performance of ML entities in the simulation environment before applying them to the target network or system. The simulation phase is considered optional and can be skipped in the AI / ML operational workflow.

[0049] Deployment phase:

[0050] ML entity loading: The process of making trained ML entities available to the target AI / ML inference function (also known as atomic operation serialization). In some cases, the deployment phase may not be required, such as when the training function and the inference function are co-located.

[0051] Reasoning stage:

[0052] AI / ML Inference: Perform inference using trained ML entities through the AI / ML inference feature.

[0053] In one embodiment, the first device is an ML training producer, and the second device is an ML training consumer; the management service consumer can be an ML training consumer, and the management service producer can be an ML training producer. Figure 1 Cross-domain management functional unit, domain management functional unit, network element, etc.

[0054] Figure 2 A flow chart of a communication method provided in one embodiment is shown in FIG. Figure 2 As shown, the communication method described in the embodiment of the present application is applied to the first device, and the method includes S110-S120:

[0055] S110: Receive a training request sent by the second device, where the training request includes training requirement information, and the training requirement information is used to indicate the requirements of this training task.

[0056] Among them, the training request can be understood as a communication request information, requesting the first device to perform model training; the training requirement information can be understood as a type of information containing instructions for the requirements of this training task, which is used to indicate the requirements of this training task. The training requirement information can be the requirement information of the model, or it can be a strategy indicating how to perform model training, and so on.

[0057] When the second device needs to train a model, it determines training requirement information and uses the training requirement information to indicate the requirements of the current training task. For example, the second device requests to train a model for base station energy conservation. A training request is generated based on the training requirement information. For example, a field in the training request carries the training requirement information. The training request may also include other training-related information. The second device sends the training request to the first device via a pre-agreed communication method. The first device receives the training request sent by the second device and parses the training request to obtain the training requirement information. When parsing the training request, the training request may be parsed according to a pre-agreed protocol specification. The obtained training requirement information may indicate the requirements of the current training task.

[0058] S120: Perform model training according to the training request.

[0059] The first device performs model training according to the information indicated by the training request, for example, performing model training according to the training requirement information in the training request, or performing model training according to the training requirement information and other training-related information in the training request. Before performing model training, the first device may determine how to carry out model training based on the information in the training request, for example, which node will perform model training, when to start model training, etc., and start model training according to the determined model training method.

[0060] In the communication method provided in the embodiment of the present application, a first device receives a training request sent by a second device, where the training request includes training requirement information, where the training requirement information is used to indicate the requirements of this training task. Model training is performed according to the training request, thereby solving the problem that model training cannot meet actual requirements. The second device indicates the requirements of this training task to the first device by sending a training request and carrying the training requirement information in the training request. The first device performs targeted training according to the training requirement information to meet the requirements of the training task. The first device considers the actual training task requirements in a timely manner when performing model training. The trained model can meet the requirements, thereby improving the effectiveness of model training, avoiding repeated training, and saving time.

[0061] In some embodiments, the training requirement information includes one or more of the following:

[0062] Model requirement information;

[0063] training strategies;

[0064] Among them, the model requirement information is used to indicate the requirements of this training, and the training strategy is used to arrange the training plan.

[0065] Model requirement information can be understood as the requirements placed on the model during training, such as accuracy and complexity. This information indicates the requirements for this training session. It can also refer to the requirements for this training session, such as the energy consumption required. A training strategy can be understood as the strategy for executing model training, such as how to conduct model training. This strategy is used to orchestrate training plans, including the time, nodes, resources, and whether a single model should be used for training, or whether multiple models should be used for collaborative training.

[0066] The training requirement information may include model requirement information, or a training strategy, or both model requirement information and a training strategy. Both the model requirement information and the training strategy may instruct the first device on how to perform model training this time.

[0067] In some embodiments, the model requirement information includes one or more of the following:

[0068] Model complexity requirement information;

[0069] Model performance requirement information;

[0070] Energy consumption demand information;

[0071] Among them, the model complexity requirement information is used to indicate the requirement of this training for model complexity, the model performance requirement information is used to indicate the requirement of this training for model performance, and the energy consumption requirement information is used to indicate the energy consumption requirement of this training.

[0072] Model complexity requirement information can be understood as information describing the complexity requirements of the model, and is used to indicate the complexity requirements of the model being trained. It can include the number of model parameters, model computational effort (e.g., floating-point operations per second (FLOPs), multiply-add operations (MACs), multiply-add operations (Madds), etc.), model complexity level, etc. Exemplarily, the model complexity requirement information indicates the FLOPs threshold for the model being trained.

[0073] Model performance requirement information can be understood as information that describes the performance requirements of the model, and is used to indicate the performance requirements of the model for this training. For example, it indicates the performance parameters and their values ​​that the model for this training needs to support. For example, the performance parameters can be one or more of the following: accuracy, precision, recall rate, F1, MSE (mean square error), MAE (mean absolute error), RMSE (root mean square error).

[0074] Energy consumption demand information can be understood as information that describes the energy consumption required for model training, and is used to indicate the energy consumption demand for this training. For example, the energy consumption demand indicated in the energy consumption demand information is a threshold or a threshold range, such as the energy consumption of this training is less than x KWh, etc. The energy consumption demand information can also include relevant energy consumption levels, corresponding to different energy consumptions, and so on.

[0075] In some embodiments, the number of model parameters refers to the number of learnable parameters in a machine learning model, which are typically weights and biases used to represent the features and relationships of the model. The number of model parameters directly affects the complexity and capacity of the model, and is generally closely related to the expressiveness and performance of the model. A larger number of parameters may mean a more complex model, but it may also increase the risk of overfitting. Model computational capacity: that is, the computing resources required by the model when performing inference or training. Computational capacity is usually measured by indicators such as FLOPs, MACs, and MAdds. FLOPs, MACs, and MAdds are metrics used to quantify the computational complexity of algorithms, especially in the fields of machine learning and signal processing. FLOPs represents the number of floating-point operations (such as addition, subtraction, multiplication, and division) performed by a computing device in one second, measuring the raw computing power of the system; FLOPs can also be used to measure the computational complexity of an algorithm, indicating the number of floating-point operations required to perform a specific task. MACs specifically refers to the number of multiplication-accumulation operations performed by an algorithm or hardware component. In many machine learning models (such as convolutional neural networks (CNNs), the majority of the computational work comes from MAC operations, which involve multiplying two numbers and then adding the result to an accumulator. MAdds is similar to MACs, but counts the number of multiplication and addition operations separately. Therefore, for each MAC operation, two operations (one multiplication and one addition) are calculated, providing a more detailed breakdown of computational complexity compared to MACs. These metrics are important for understanding and comparing the efficiency of different algorithms, architectures, and hardware platforms, and can help optimize algorithms in resource-constrained environments (such as mobile devices or edge computing devices) to improve speed and resource consumption efficiency.

[0076] In some embodiments, the training strategy includes one or more of the following:

[0077] Energy consumption-based training strategies;

[0078] Training strategies based on model complexity;

[0079] Training strategies based on model performance;

[0080] Cost-based training strategies;

[0081] Among them, the energy consumption-based training strategy is used to indicate the scheduling of training plans based on energy consumption, the model complexity-based training strategy is used to indicate the scheduling of training plans based on model complexity, the model performance-based training strategy is used to indicate the scheduling of training plans based on model performance, and the cost-based training strategy is used to indicate the scheduling of training plans based on cost.

[0082] When the energy consumption-based training strategy indicates that the training plan should be arranged based on energy consumption, the energy consumption of the model training is given priority. For example, the arranged training plan can ensure the least energy consumption during the model training process. When the model complexity-based training strategy indicates that the training plan should be arranged based on model complexity, the model complexity is given priority. For example, the arranged training plan can ensure the lowest complexity during the model training process. When the model performance-based training strategy indicates that the training plan should be arranged based on model performance, the model performance is given priority. For example, the arranged training plan can ensure the best performance during the model training process, or achieve the best performance on a certain model performance indicator. When the cost-based training strategy indicates that the training plan should be arranged based on model performance, the cost of the resources used in the training process is given priority. For example, the arranged training plan can ensure the least cost during the model training process.

[0083] In some embodiments, the training requirement information further includes:

[0084] Priority information.

[0085] Priority information can be used to indicate the execution priority of this model training. For example, in conflict resolution, higher-priority tasks can be executed first (preempting the resources of lower-priority tasks). When a task conflict occurs, lower-priority tasks need to be paused and wait for the higher-priority task to complete before executing or directly canceling the task, etc. Priority information is used to indicate the execution priority of different training tasks for conflict resolution.

[0086] In some embodiments, the priority information includes a priority indication or priority level of whether the task can be preempted.

[0087] When the priority information is a priority indication of whether the task can be preempted, it indicates whether the task of this training can be preempted; for example, if the priority information is that the task can be preempted, the priority of this training task is low, and during the training process, this model training task can be preempted by other training tasks; if the priority information is that the task cannot be preempted, the priority of this training task is high, and during the training process, this model training task cannot be preempted by other training tasks.

[0088] When the priority information is a priority level, the execution priority of the task can be determined by comparing the priority levels of two training tasks. For example, the priority level of the currently executing training task A is low. At this time, a new training task B is received. The priority level of training task B is medium, which is higher than the level of training task A. The training of training task A is paused or interrupted, and training task B is executed for model training.

[0089] In some embodiments, performing model training according to a training request includes:

[0090] Determine a training plan based on training requirements. The training plan includes one or more of the following: training execution time, training execution nodes, training resources, training environment, and model collaborative training strategy.

[0091] Train the model according to the training plan.

[0092] The training requirement information in the training request is analyzed to determine the requirements, and the requirements corresponding to different training schemes are pre-determined. An appropriate training scheme is selected based on the requirements. In this case, each training scheme includes one or more of the determined training execution time, training execution node, training resources, training environment, and model collaborative training strategy. Alternatively, the training execution nodes, training execution time, training resources, training environment, etc. corresponding to different requirements are pre-set. After analyzing the training requirement information, the appropriate training execution node, training execution time, training resources, training environment, model collaborative training strategy, etc. are selected based on the analyzed requirements, and a training scheme is orchestrated and generated. Model training according to the training scheme can include executing model training tasks through training execution nodes, starting model training tasks according to training execution time, performing model training through training resources, and deploying the environment according to the training environment, such as setting software parameters, setting hardware parameters, and deploying a virtual environment. The collaborative training of multiple models is orchestrated according to the model collaborative training strategy, such as determining the order in which multiple models are trained.

[0093] The training plan includes, but is not limited to, one or more of the following: training execution time, training execution node, training resources, training environment, and model collaborative training strategy. The training execution node may be the first device or another device. If the training execution node is not the first device, the first device may send the training plan to the corresponding training execution node, instructing the training execution node to perform model training according to the training plan.

[0094] In some embodiments, the training resources are resources reserved for this training.

[0095] Training resources can be pre-reserved. Different resources can be pre-reserved for model training based on different training requirements, or the same resources can be reserved for model training. Training resources can also be set based on the device sending the training request. For example, different training resources can be reserved for different or different types of second devices.

[0096] In some embodiments, the model collaborative training strategy indicates a strategy for collaborative training of multiple models.

[0097] The model collaborative training strategy can indicate how to train multiple models during the training process, such as the training order of each model and the input-output relationship of the model. For example, the output of model A serves as the input of model B and model C, and the output of model B and model C serves as the input of model D at the same time, and so on.

[0098] In some embodiments, the first device sends the training regimen to the second device.

[0099] After the training plan is arranged, it is sent to the second device so that the second device can save the training plan. The second device can use the training plan to determine whether the model training meets the requirements, retain information related to the model training, etc.

[0100] In some embodiments, the method further comprises:

[0101] After the model training is started, the first device sends a first indication message to the second device, where the first indication message is used to indicate that the model training has been started.

[0102] The first indication information can be understood as a communication message used to indicate that model training has been initiated. After initiating model training, the first device generates the first indication information. The first indication information can carry agreed information in agreed fields. For example, carrying 1 in the first field indicates that model training has been initiated. After generating the first indication information, the first device sends the first indication information to the second device. After receiving the first indication information, the second device can determine that model training has been initiated based on the first indication information.

[0103] In some embodiments, before starting model training, the method further includes:

[0104] The first device performs a feasibility check on the training task, generates a feasibility check report on the training task, and sends the report to the second device.

[0105] Among them, the training task feasibility check report can be understood as a report generated by checking whether the training task is feasible. Analyze the training request, for example, analyze whether the training requirement information in the training request can be met, and complete the feasibility check of the training task. Analyzing whether the training requirement information in the training request can be met can be analyzing whether the model requirement information, training strategy, etc. can be met. For example, after analysis, it is determined that the energy consumption requirements and model complexity requirements of this training cannot be met due to network changes. Determine the result of the feasibility check, generate a training task feasibility check report based on the result, and send the training task feasibility check report to the second device so that the second device can determine whether the training task is feasible.

[0106] In some embodiments, the training task feasibility check report includes one or more of the following information:

[0107] Results of feasibility checks on training missions;

[0108] Auxiliary information.

[0109] Among them, auxiliary information can be understood as information that assists the second device in understanding or analyzing the results of the training task feasibility check. When performing a training task feasibility check on a training request, after determining the results of the training task feasibility check, auxiliary information can also be generated accordingly. For example, if the training task feasibility check fails, auxiliary information can be generated based on the reason for the failure. Corresponding modification suggestions can also be given. For example, if energy consumption cannot meet energy consumption requirements, it can be suggested that the second device modify the energy consumption requirements, and it can be suggested to which value or range the energy consumption requirements should be modified.

[0110] In some embodiments, the auxiliary information includes one or more of the following:

[0111] Reasons for training request recommendations;

[0112] Training request suggestion information.

[0113] Among them, the reason for the training request suggestion can be understood as the reason for providing suggestions for the training request, for example, the reason for the failure of the feasibility check of the training task is directly used as the reason for the training request suggestion; the training request suggestion information can be understood as the suggestion information for the training request, for example, it is recommended to modify the training request, how to modify it, etc. Exemplarily, when the feasibility check of the training task is performed, the energy consumption requirement cannot be met, resulting in the feasibility check result of the training task failing, the reason for the training request suggestion is that the energy consumption requirement cannot be met, and the training request suggestion information is to reduce the energy consumption requirement or adjust the energy consumption requirement information to 1KWh. When the feasibility check of the training task passes, the auxiliary information may not be included in the feasibility check report of the training task.

[0114] In some embodiments, the method further comprises:

[0115] After completing the model training, the first device generates a training report and sends it to the second device;

[0116] The training report includes one or more of the following information:

[0117] Training needs report;

[0118] Model information.

[0119] A training report can be understood as a report describing information related to model training, such as the training process and results. A training requirements report can be understood as a report that expresses training requirements. Model information can be understood as information describing the model, such as its energy consumption, complexity, training duration, and parameters.

[0120] After the model training is completed, relevant information in the training process is obtained and written into the report to generate a training report. The information type included in the training report can be pre-set. When generating the training report, the corresponding information is obtained according to the pre-set information type.

[0121] In some embodiments, the training needs report includes at least one of the following:

[0122] Model complexity requirement report;

[0123] Model performance requirements report;

[0124] Energy consumption demand report;

[0125] Among them, the model complexity requirement report is used to indicate the model complexity of this training, the model performance requirement report is used to indicate the model performance of this training, and the energy consumption requirement report is used to indicate the energy consumption of this training.

[0126] In some embodiments, the method further comprises:

[0127] After completing the model training, the first device saves or updates the model information.

[0128] After completing the model training, the model information needs to be saved. If the model has not been trained, the model information can be saved directly. If the model has been trained before this training, historical information may have been saved. At this time, the historically saved model information is updated according to the model information obtained from this training, replacing the historically saved model information, and the model information obtained from this training is saved as the latest information.

[0129] In some embodiments, the method further comprises:

[0130] After completing the model training, the first device sends a model information saving request to the second device, where the model information saving request is used to request whether to save or update the model information;

[0131] After receiving the save instruction sent by the second device, the first device determines whether to save or update the model information according to the save instruction, where the save instruction is used to instruct the first device whether to save or update the model information.

[0132] Among them, the model information save request can be understood as a request for saving or updating the model information; the save indication can be understood as indication information indicating whether to save or update the model information, which is used to instruct the first device whether to save or update the model information.

[0133] After completing model training, the first device generates a model information saving instruction and sends it to the second device, requesting the second device whether the model information needs to (or can) be saved or updated. After receiving the model information saving request, the second device can generate a saving instruction based on the importance, type, and other information of the model, and instruct the first device through the saving instruction whether to save or update the model information. The first device receives the saving instruction fed back by the second device, parses the saving instruction, and determines whether to save or update the model information based on the information carried in the saving instruction. For example, if the data stored in the specified field in the saving instruction is 1, it is determined that the model information is saved or updated; if the data stored in the specified field in the saving instruction is 0, it is determined that the model information is not saved or updated.

[0134] In some embodiments, the method further includes: after the first device completes the model training, saving or updating the model. After the model is saved or updated, the model information can also be saved accordingly.

[0135] In some embodiments, the method further includes: after completing model training, the first device sends a model save request to the second device, and the model save request is used to request whether to save or update the model; after receiving the third indication information sent by the second device, determining whether to save or update the model according to the third indication information, and the third indication information is used to instruct the first device whether to save or update the model.

[0136] In some embodiments, the model information includes one or more of the following:

[0137] Model energy consumption information;

[0138] Model complexity information;

[0139] Model interoperability information;

[0140] Training background information;

[0141] Among them, the model energy consumption information is used to indicate the energy consumption required to run the model, the model complexity information is used to indicate the complexity of the model, the model interoperability information is used to indicate the information required for the model to interoperate, and the training background information is used to indicate the background of the model training.

[0142] Model energy consumption information indicates the energy consumption required to run the model, which can be x KWh. Model complexity can include the number of model parameters, model computational complexity, model complexity level, etc. Model interoperability information can include the model type (e.g., random forest, neural network, etc.), input information, operating environment, etc. Training background information can include training execution time, training execution nodes, etc. Training requirements information can also be used directly as training background information, or part of the training requirements information can be used as training background information, etc.

[0143] After the model training is completed, some parameter information of the model, information involved in the training process, etc. are determined accordingly. For example, after the model is trained, the model energy consumption information, model complexity information, model interoperability information, training background information, model parameters, etc. are determined accordingly, and one or more of the model energy consumption information, model complexity information, model interoperability information, and training background information are obtained as model information.

[0144] In some embodiments, the method further comprises:

[0145] During the model training process, the first device detects or predicts a conflict, generates a conflict report, and sends it to the second device.

[0146] Among them, the conflict report can be understood as information indicating that there is a conflict in model training. During the execution of this model training task (i.e., during this model training process), if other model training tasks are received and the priority information of the two model training tasks is compared, if the priority of this model training task is lower than the priority of other model training tasks, or if this model training task can be preempted, it is determined that there is a conflict, and a conflict report is generated and sent to the second device. During the model training process, the first device may also determine that there is a conflict if it detects an abnormality in the training execution node, etc. During the model training process, the relevant data involved in the training process may also be analyzed to predict whether there may be a conflict. For example, if the accuracy of the model is not compatible with the software and hardware parameters of the training execution node through data analysis, it can be considered that a conflict is predicted; or, during the model training process, a new training request is received and a training plan for the new training request is arranged. The new training plan has the same training execution node as the training plan of the current training, but the training execution time is different. Based on the progress of this training, the end time of this training can be predicted. If the training execution time of the new training plan is before the end time of this training, it can be considered that a conflict is predicted, etc.

[0147] In some embodiments, the conflict report includes one or more of the following information:

[0148] Causes of conflict;

[0149] Training request suggestion information.

[0150] The conflict reason may be that a high-priority training request preempts or an abnormality occurs in a training execution node, etc. The training request suggestion information may be a conflict resolution method, such as pausing, canceling, or continuing the training task.

[0151] In some embodiments, the method further comprises:

[0152] The first device receives second indication information sent by the second device, and processes the training task conflict according to the second indication information, where the second indication information is used to instruct the first device to process the training task conflict; or

[0153] The first device processes the training task conflict according to a predefined processing strategy.

[0154] Among them, the second indication information can be understood as an indication information indicating how to handle the training task conflict. After receiving the conflict report, the second device generates the second indication information based on the information in the conflict report, the model type and other information and sends it to the first device, and instructs the first device to handle the training task conflict through the second indication information. After the first device receives the second indication information sent by the second device, it parses the second indication information, determines the conflict handling method specifically indicated by the second indication information, and handles the training task conflict according to the conflict handling method indicated. For example, if the second indication information indicates 0, the current model training is suspended; if the second indication information indicates 1, the current model training is canceled; if the second indication information indicates 2, the first device decides how to handle the current model training. The first device can determine whether to suspend or continue the model training based on the preset processing strategy, device operation status, model training progress, types of the two models and other information.

[0155] Alternatively, a processing strategy is predefined, such as pausing model training, continuing model training, or canceling model training. After detecting a conflict, the first device handles the training task conflict according to the predefined processing strategy. Handling the training task conflict using the predefined processing strategy can be performed after the conflict is detected, and there is no strict order of execution with the generation and transmission of the conflict report; it can be performed simultaneously or sequentially.

[0156] The communication method provided in the embodiment of the present application further explains the training requirement information, which includes one or more of the model requirement information and the training strategy. The model requirement information is used to indicate the requirements of the model for this training, and the training strategy is used to arrange the training plan. The first device performs targeted training based on the training requirement information to meet the requirements of different training tasks. The first device considers the actual training task requirements in a timely manner when performing model training, and can perform model training based on energy consumption requirements, model accuracy requirements, model complexity requirements, etc., which can achieve the effects of saving energy consumption, ensuring model accuracy and complexity, etc., thereby improving the effectiveness of model training, avoiding repeated training, and saving time.

[0157] Figure 3 A flowchart of another communication method provided in an embodiment, such as Figure 3 As shown, the communication method described in the embodiment of the present application is applied to the second device, and the method includes S210-S220:

[0158] S210: Generate a training request, where the training request includes training requirement information, and the training requirement information is used to indicate the requirements of this training task.

[0159] S220: Send the training request to the first device, so that the first device performs model training according to the training request.

[0160] In the communication method provided in the embodiment of the present application, the second device generates and sends a training request to the first device. The training request includes training requirement information. The training requirement information is used to indicate the requirements of this training task. The first device performs model training according to the training request, which solves the problem that model training cannot meet actual needs. The second device sends a training request and carries the training requirement information in the training request to indicate the requirements of this training task to the first device. The first device performs targeted training according to the training requirement information to meet the requirements of the training task. The first device considers the actual training task requirements in a timely manner when performing model training. The trained model can meet the requirements, thereby improving the effectiveness of model training, avoiding repeated training, and saving time.

[0161] In some embodiments, the training requirement information includes one or more of the following:

[0162] Model requirement information;

[0163] training strategies;

[0164] The model requirement information is used to indicate the requirements of this training, and the training strategy is used to arrange the training plan.

[0165] In some embodiments, the model requirement information includes one or more of the following:

[0166] Model complexity requirement information;

[0167] Model performance requirement information;

[0168] Energy consumption demand information;

[0169] Among them, the model complexity requirement information is used to indicate the requirement of this training for model complexity, the model performance requirement information is used to indicate the requirement of this training for model performance, and the energy consumption requirement information is used to indicate the energy consumption requirement of this training.

[0170] In some embodiments, the training strategy includes one or more of the following:

[0171] Energy consumption-based training strategies;

[0172] Training strategies based on model complexity;

[0173] Training strategies based on model performance;

[0174] Cost-based training strategies;

[0175] Among them, the energy consumption-based training strategy is used to indicate the training plan arranged based on energy consumption, the model complexity-based training strategy is used to indicate the training plan arranged based on model complexity, the model performance-based training strategy is used to indicate the training plan arranged based on model performance, and the cost-based training strategy is used to indicate the training strategy arranged based on cost.

[0176] In some embodiments, the training requirement information further includes:

[0177] Priority information.

[0178] In some embodiments, the priority information includes a priority indication or priority level of whether the task can be preempted.

[0179] In some embodiments, the method further comprises:

[0180] The second device receives a training plan sent by the first device, where the training plan includes one or more of the following: training execution time, training execution node, training resources, training environment, and model collaborative training strategy;

[0181] Among them, training resources are the resources reserved for this training;

[0182] The model co-training strategy indicates the strategy for co-training multiple models.

[0183] In some embodiments, the method further comprises:

[0184] The second device receives first indication information sent by the first device after the model training is started, where the first indication information is used to indicate that the model training has been started.

[0185] In some embodiments, the method further comprises:

[0186] The second device receives the training task feasibility check report sent by the first device.

[0187] In some embodiments, the training task feasibility check report includes one or more of the following information:

[0188] Results of feasibility checks on training missions;

[0189] Auxiliary information.

[0190] In some embodiments, the auxiliary information includes one or more of the following:

[0191] Reasons for training request recommendations;

[0192] Training request suggestion information.

[0193] In some embodiments, the method further comprises:

[0194] The second device receives the training report sent by the first device;

[0195] The training report includes one or more of the following information:

[0196] Training needs report;

[0197] Model information.

[0198] In some embodiments, the training needs report includes at least one of the following:

[0199] Model complexity requirement report;

[0200] Model performance requirements report;

[0201] Energy consumption demand report;

[0202] Among them, the model complexity requirement report is used to indicate the model complexity of this training, the model performance requirement report is used to indicate the model performance of this training, and the energy consumption requirement report is used to indicate the energy consumption of this training.

[0203] In some embodiments, the method further comprises:

[0204] The second device saves or updates the model information.

[0205] In some embodiments, the method further comprises:

[0206] The second device receives a model information saving request sent by the first device, where the model information saving request is used to request whether to save or update the model information;

[0207] The second device generates a save instruction and sends it to the first device, where the save instruction is used to instruct the first device whether to save or update the model information.

[0208] In some embodiments, the model information includes one or more of the following:

[0209] Model energy consumption information;

[0210] Model complexity information;

[0211] Model interoperability information;

[0212] Training background information;

[0213] Among them, the model energy consumption information is used to indicate the energy consumption required to run the model, the model complexity information is used to indicate the complexity of the model, the model interoperability information is used to indicate the information required for the model to interoperate, and the training background information is used to indicate the background of the model training.

[0214] In some embodiments, the method further comprises:

[0215] The second device receives the conflict report sent by the first device.

[0216] In some embodiments, the conflict report includes at least one of the following information:

[0217] Causes of conflict;

[0218] Training request suggestion information.

[0219] In some embodiments, the method further comprises:

[0220] The second device generates second indication information and sends it to the first device, where the second indication information is used to instruct the first device to handle the training task conflict.

[0221] The communication process is described through the following examples:

[0222] Example 1

[0223] Figure 4 A timing diagram of communication between a first device and a second device is provided, using model training based on energy consumption strategy as an example to illustrate the communication process:

[0224] Step 1: The second device sends a training request to the first device, requesting the creation of a training instance. The training request includes training requirement information. In this embodiment, the second device requests the training of a model for base station energy conservation. The training requirement information includes model requirement information and a training strategy. The model requirement information includes: model complexity requirement information and model performance requirement information for this training. The model complexity requirement information indicates the FLOPs threshold of the model for this training; the model performance requirement information indicates the performance parameters and their values ​​that the model for this training must support, such as accuracy; and the training strategy is an energy-based training strategy that minimizes energy consumption for this training.

[0225] Step 2: The first device creates a training instance based on the received training request.

[0226] Step 3: The first device sends an instance creation response to the second device.

[0227] Step 4: The first device arranges a training plan or a multi-model collaborative training plan (applicable to retraining scenarios) according to the training strategy in the received training request and the energy consumption-based training strategy. The above-mentioned plan includes one or more of the following: training execution time and training execution node responsible for this training, training resources, training environment, and model collaborative training strategy. The model collaborative training strategy is used to indicate the strategy for collaborative training of multiple models, such as the order of training multiple models. The first device starts model training based on the training plan.

[0228] Step 5: The first device notifies the second device that training has started.

[0229] Step 6: Training is complete.

[0230] Step 7: The first device sends a training report to the second device. The training report includes at least one of a training requirements report and model information. The training requirements report includes a model complexity requirements report, a model performance requirements report, and an energy consumption requirements report. The model complexity requirements report reports the FLOPs value of the trained model, and the model performance requirements report indicates the accuracy of the trained model. The model information includes one or more of the following: model energy consumption information, model complexity information, model interoperability information, and training background information (e.g., training requirements information).

[0231] Step 8: The first device requests the physical warehouse to register the model obtained in this training.

[0232] Step 9: The entity warehouse stores the model, i.e., the entity, including model information. The model information includes one or more of the following: model energy consumption information, model complexity information, model interoperability information, and training background information.

[0233] Step 10: The physical warehouse notifies the first device that the physical registration is successful.

[0234] Example 2

[0235] Figure 5 Another timing diagram for communication between a first device and a second device is provided, illustrating the communication process using energy consumption requirement-based training and conflict handling, as well as feasibility check of training tasks.

[0236] Step 1: The second device sends a training request to the first device, requesting the creation of an ML training instance. The training request includes training requirement information. In this embodiment, the second device requests the training of a model for base station energy conservation. The training requirement information includes model requirement information and priority information. The model requirement information includes energy consumption requirement information, model complexity requirement information, and model performance requirement information for this training. The energy consumption requirement information indicates that the energy consumption for this training must be less than 10kWh; the model complexity requirement information indicates the lower limit of FLOPs for the model being trained; the model performance requirement information indicates the performance parameters and their values ​​that the model must support, such as accuracy; and the priority information is priority 2.

[0237] Step 2: The first device creates a training instance based on the received training request.

[0238] Step 3: The first device sends an instance creation response to the second device.

[0239] Step 4: The first device performs a feasibility check on the training task for the training request and generates a feasibility check report for the training task.

[0240] Step 5: The first device sends a training task feasibility check report to the second device, where the feasibility report includes a training task feasibility check result and auxiliary information, wherein the auxiliary information includes at least one of a reason for the training request suggestion and training request suggestion information.

[0241] Note: Figure 5 Taking the training task feasibility check result as failed as an example, the first device notifies the second device to modify the training request through the training task feasibility check report. If the training task feasibility check result is passed, directly execute step 8.

[0242] Figure 5 The reason for the training request suggestion is that the energy consumption demand cannot be met, and the training request suggestion information is "The energy consumption demand this time is less than 15kWh."

[0243] Step 6: The second device modifies the training request according to the training task feasibility check report.

[0244] Step 7: The second device sends the modified training request to the first device.

[0245] Step 8: The first device arranges a training plan or a multi-model collaborative training plan (applicable to retraining scenarios) based on the energy consumption of this training indicated by the energy consumption requirement information in the received training request. The above plan includes one or more of the following: training execution time and training execution node responsible for this training, training resources, training environment, and model collaborative training strategy. The model collaborative training strategy is used to indicate the strategy for collaborative training of multiple models, such as the order in which multiple models are trained. The second device starts model training based on the training plan.

[0246] Step 9: The first device notifies the second device that training has started.

[0247] Step 10: The first device detects or predicts a conflict.

[0248] For example, a training task with a training priority of 1 is triggered.

[0249] Step 11: The first device sends a conflict report to the second device, where the conflict report includes the cause of the conflict and training request suggestion information.

[0250] For example, the conflict reason may be: a high-priority training request preempts the request; the training request suggestion information is a conflict resolution method, such as pausing the current training task, modifying the priority information of the current training task, etc. In this example, the priority of the current training task is increased.

[0251] Step 12: The second device instructs the first device to continue the training task through the second instruction information.

[0252] Step 13: Training is complete.

[0253] Step 14: The first device sends a training report to the second device. The training report includes a training requirements report and model information. The training requirements report includes a model complexity requirement report, a model performance requirement report, and an energy consumption requirement report. The model complexity requirement report reports the FLOPs value of the trained model, the model performance requirement report indicates the accuracy of the trained model, and the energy consumption requirement report indicates that the training consumed 14 kWh. Model information includes one or more of the following: model energy consumption information, model complexity information, model interoperability information, and training background information (e.g., training requirements information).

[0254] Step 15: The first device requests the physical warehouse to register the model obtained in this training.

[0255] Step 16: The entity warehouse stores the model, i.e., the entity, including model information. The model information includes one or more of the following: model energy consumption information, model complexity information, model interoperability information, and training background information.

[0256] Step 17: The physical warehouse notifies the first device that the physical registration is successful.

[0257] When the second device requests the first device to suspend or cancel the training task through the second indication information, the first device can suspend or cancel the training task according to the second indication information and send a request response to the second device.

[0258] Figure 6 A schematic diagram of the structure of a communication device provided in an embodiment, wherein the device is applied to a first device, such as Figure 6 As shown, the device includes: a training request receiving module 310 and a model training module 320.

[0259] A training request receiving module 310 is configured to receive a training request sent by a second device, wherein the training request includes training requirement information indicating requirements for the current training task;

[0260] The model training module 320 is used to perform model training according to the training request.

[0261] In the communication device provided in the embodiment of the present application, a first device receives a training request sent by a second device, the training request includes training requirement information, the training requirement information is used to indicate the requirements of this training task, and model training is performed according to the training request, thereby solving the problem that model training cannot meet actual needs. The second device indicates the requirements of this training task of the first device by sending a training request and carrying the training requirement information in the training request. The first device performs targeted training according to the training requirement information to meet the requirements of the training task. The first device considers the actual training task requirements in a timely manner when performing model training. The trained model can meet the requirements, thereby improving the effectiveness of model training, avoiding repeated training, and saving time.

[0262] In some embodiments, the training requirement information includes one or more of the following:

[0263] Model requirement information;

[0264] training strategies;

[0265] The model requirement information is used to indicate the requirements of this training, and the training strategy is used to arrange the training plan.

[0266] In some embodiments, the model requirement information includes one or more of the following:

[0267] Model complexity requirement information;

[0268] Model performance requirement information;

[0269] Energy consumption demand information;

[0270] Among them, the model complexity requirement information is used to indicate the requirement of this training for model complexity, the model performance requirement information is used to indicate the requirement of this training for model performance, and the energy consumption requirement information is used to indicate the energy consumption requirement of this training.

[0271] In some embodiments, the training strategy includes one or more of the following:

[0272] Energy consumption-based training strategies;

[0273] Training strategies based on model complexity;

[0274] Training strategies based on model performance;

[0275] Cost-based training strategies;

[0276] Among them, the energy consumption-based training strategy is used to indicate the scheduling of training plans based on energy consumption, the model complexity-based training strategy is used to indicate the scheduling of training plans based on model complexity, the model performance-based training strategy is used to indicate the scheduling of training plans based on model performance, and the cost-based training strategy is used to indicate the scheduling of training plans based on cost.

[0277] In some embodiments, the training requirement information further includes:

[0278] Priority information.

[0279] In some embodiments, the priority information includes a priority indication or a priority level of whether the task can be preempted.

[0280] In some embodiments, performing model training according to the training request includes:

[0281] Determine a training plan based on the training requirement information, wherein the training plan includes one or more of the following: training execution time, training execution nodes, training resources, training environment, and model collaborative training strategy;

[0282] The model is trained according to the training scheme.

[0283] In some embodiments, the training resources are resources reserved for this training.

[0284] The model collaborative training strategy indicates a strategy for collaborative training of multiple models.

[0285] In some embodiments, the apparatus further comprises:

[0286] The training scheme sending module is used to send the training scheme to the second device.

[0287] In some embodiments, the apparatus further comprises:

[0288] The first indication sending module is used to send first indication information to the second device after the model training is started, and the first indication information is used to indicate that the model training has been started.

[0289] In some embodiments, the apparatus further comprises:

[0290] The feasibility check report sending module is used to perform a training task feasibility check on the training request before starting model training, generate a training task feasibility check report and send it to the second device.

[0291] In some embodiments, the training task feasibility check report includes one or more of the following information:

[0292] Results of feasibility checks on training missions;

[0293] Auxiliary information.

[0294] In some embodiments, the auxiliary information includes one or more of the following:

[0295] Reasons for training request recommendations;

[0296] Training request suggestion information.

[0297] In some embodiments, the apparatus further comprises:

[0298] A training report sending module, configured to generate a training report and send the report to the second device after completing the model training;

[0299] The training report may include one or more of the following information:

[0300] Training needs report;

[0301] Model information.

[0302] In some embodiments, the training needs report includes at least one of the following:

[0303] Model complexity requirement report;

[0304] Model performance requirements report;

[0305] Energy consumption demand report;

[0306] Among them, the model complexity requirement report is used to indicate the model complexity of this training, the model performance requirement report is used to indicate the model performance of this training, and the energy consumption requirement report is used to indicate the energy consumption of this training.

[0307] In some embodiments, the apparatus further comprises:

[0308] The first model updating module is used to save or update the model information after completing the model training.

[0309] In some embodiments, the apparatus further comprises:

[0310] A save request sending module, configured to send a model information save request to the second device after completing model training, wherein the model information save request is used to request whether to save or update the model information;

[0311] The second model updating module is used to determine whether to save or update the model information according to the save instruction after receiving the save instruction sent by the second device, and the save instruction is used to instruct the first device whether to save or update the model information.

[0312] In some embodiments, the model information includes one or more of the following:

[0313] Model energy consumption information;

[0314] Model complexity information;

[0315] Model interoperability information;

[0316] Training background information;

[0317] Among them, the model energy consumption information is used to indicate the energy consumption required to run the model, the model complexity information is used to indicate the complexity of the model, the model interoperability information is used to indicate the information required for the model to interoperate, and the training background information is used to indicate the background of model training.

[0318] In some embodiments, the apparatus further comprises:

[0319] The conflict report sending module is used to detect or predict a conflict during the model training process, generate a conflict report and send it to the second device.

[0320] In some embodiments, the conflict report includes one or more of the following information:

[0321] Causes of conflict;

[0322] Training request suggestion information.

[0323] In some embodiments, the apparatus further comprises:

[0324] A conflict handling module is used to receive the second indication information sent by the second device, and handle the training task conflict according to the second indication information, wherein the second indication information is used to instruct the first device to handle the training task conflict; or, handle the training task conflict according to a predefined processing strategy.

[0325] The communication device proposed in this embodiment and the communication method proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to any of the above embodiments, and this embodiment has the same beneficial effects as executing the communication method.

[0326] Figure 7 This is a schematic structural diagram of another communication device provided in an embodiment, which is applied to a second device, such as Figure 7 As shown, the apparatus includes: a training request generating module 410 and a training request sending module 420 .

[0327] A training request generating module 410 is configured to generate a training request, wherein the training request includes training requirement information, and the training requirement information is configured to indicate the requirements of the current training task;

[0328] The training request sending module 420 is used to send the training request to the first device, so that the first device performs model training according to the training request.

[0329] In the communication device provided in the embodiment of the present application, the second device generates and sends a training request to the first device. The training request includes training requirement information. The training requirement information is used to indicate the requirements of this training task. The first device performs model training according to the training request, which solves the problem that model training cannot meet actual needs. The second device sends a training request and carries the training requirement information in the training request to indicate the requirements of this training task of the first device. The first device performs targeted training according to the training requirement information to meet the requirements of the training task. The first device considers the actual training task requirements in a timely manner when performing model training. The trained model can meet the requirements, thereby improving the effectiveness of model training, avoiding repeated training, and saving time.

[0330] In some embodiments, the training requirement information includes one or more of the following:

[0331] Model requirement information;

[0332] training strategies;

[0333] The model requirement information is used to indicate the requirements of this training, and the training strategy is used to arrange the training plan.

[0334] In some embodiments, the model requirement information includes one or more of the following:

[0335] Model complexity requirement information;

[0336] Model performance requirement information;

[0337] Energy consumption demand information;

[0338] Among them, the model complexity requirement information is used to indicate the requirement of this training for model complexity, the model performance requirement information is used to indicate the requirement of this training for model performance, and the energy consumption requirement information is used to indicate the energy consumption requirement of this training.

[0339] In some embodiments, the training strategy includes one or more of the following:

[0340] Energy consumption-based training strategies;

[0341] Training strategies based on model complexity;

[0342] Training strategies based on model performance;

[0343] Cost-based training strategies;

[0344] Among them, the energy consumption-based training strategy is used to indicate the training plan arranged based on energy consumption, the model complexity-based training strategy is used to indicate the training plan arranged based on model complexity, the model performance-based training strategy is used to indicate the training plan arranged based on model performance, and the cost-based training strategy is used to indicate the training strategy arranged based on cost.

[0345] In some embodiments, the training requirement information further includes:

[0346] Priority information.

[0347] In some embodiments, the priority information includes a priority indication or priority level of whether the task can be preempted.

[0348] In some embodiments, the apparatus further comprises:

[0349] A training scheme receiving module is configured to receive a training scheme sent by the first device, wherein the training scheme includes one or more of the following: training execution time, training execution node, training resources, training environment, and model collaborative training strategy;

[0350] Wherein, the training resources are the resources reserved for this training;

[0351] The model collaborative training strategy indicates a strategy for collaborative training of multiple models.

[0352] In some embodiments, the apparatus further comprises:

[0353] The first indication receiving module is used to receive first indication information sent by the first device after the model training is started, and the first indication information is used to indicate that the model training has been started.

[0354] In some embodiments, the apparatus further comprises:

[0355] The feasibility check report receiving module is used to receive the training task feasibility check report sent by the first device.

[0356] In some embodiments, the training task feasibility check report includes one or more of the following information:

[0357] Results of feasibility checks on training missions;

[0358] Auxiliary information.

[0359] In some embodiments, the auxiliary information includes one or more of the following:

[0360] Reasons for training request recommendations;

[0361] Training request suggestion information.

[0362] In some embodiments, the apparatus further comprises:

[0363] A training report receiving module, configured to receive a training report sent by the first device;

[0364] The training report includes one or more of the following information:

[0365] Training needs report;

[0366] Model information.

[0367] In some embodiments, the training needs report includes at least one of the following:

[0368] Model complexity requirement report;

[0369] Model performance requirements report;

[0370] Energy consumption demand report;

[0371] Among them, the model complexity requirement report is used to indicate the model complexity of this training, the model performance requirement report is used to indicate the model performance of this training, and the energy consumption requirement report is used to indicate the energy consumption of this training.

[0372] In some embodiments, the apparatus further comprises:

[0373] The second model updating module is used to save or update the model information.

[0374] In some embodiments, the apparatus further comprises:

[0375] A save request receiving module, configured to receive a model information save request sent by the first device, where the model information save request is used to request whether to save or update the model information;

[0376] The save instruction sending module is used to generate a save instruction and send it to the first device, where the save instruction is used to instruct the first device whether to save or update the model information.

[0377] In some embodiments, the model information includes one or more of the following:

[0378] Model energy consumption information;

[0379] Model complexity information;

[0380] Model interoperability information;

[0381] Training background information;

[0382] Among them, the model energy consumption information is used to indicate the energy consumption required to run the model, the model complexity information is used to indicate the complexity of the model, the model interoperability information is used to indicate the information required for the model to interoperate, and the training background information is used to indicate the background of the model training.

[0383] In some embodiments, the apparatus further comprises:

[0384] The conflict report receiving module is configured to receive a conflict report sent by the first device.

[0385] In some embodiments, the conflict report includes at least one of the following information:

[0386] Causes of conflict;

[0387] Training request suggestion information.

[0388] In some embodiments, the apparatus further comprises:

[0389] The second indication sending module is used to generate second indication information and send it to the first device, where the second indication information is used to instruct the first device to handle the training task conflict.

[0390] The communication device proposed in this embodiment and the communication method proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to any of the above embodiments, and this embodiment has the same beneficial effects as executing the communication method.

[0391] The embodiment of the present application also provides a communication node, Figure 8 A schematic diagram of a communication node structure provided by an embodiment is shown in FIG. Figure 8 As shown, the communication node provided by the present application includes a memory 520, a processor 510, and a computer program stored in the memory and executable on the processor. When the processor 510 executes the program, the above-mentioned communication method is implemented.

[0392] The communication node may further include a memory 520; the processor 510 in the communication node may be one or more, Figure 8 Take a processor 510 as an example; the memory 520 is used to store one or more programs; the one or more programs are executed by the one or more processors 510, so that the one or more processors 510 implement the communication method as described in the embodiment of the present application.

[0393] The communication node further includes: a communication module 530 , an input device 540 and an output device 550 .

[0394] The processor 510, memory 520, communication module 530, input device 540 and output device 550 in the communication node may be connected via a bus or other means. Figure 8 The bus connection is taken as an example.

[0395] The input device 540 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the communication node. The output device 550 may include a display device such as a display screen.

[0396] The communication module 530 may include a receiver and a transmitter. The communication module 530 is configured to perform information transmission and reception communication according to the control of the processor 510.

[0397] The memory 520, as a computer-readable storage medium, can be configured to store software programs, computer executable programs, and modules, such as program instructions / modules corresponding to the communication method described in the embodiments of the present application (for example, the training request receiving module 310 and the model training module 320 in the communication device, or the training request generating module 410 and the training request sending module 420 in the communication device). The memory 520 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; and the data storage area may store data created according to the use of the communication node, etc. In addition, the memory 520 may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 520 may further include a memory remotely located relative to the processor 510, and these remote memories may be connected to the communication node via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0398] An embodiment of the present application further provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, it implements any communication method described in the embodiments of the present application.

[0399] Optionally, the communication method, applied to the first device, includes: receiving a training request sent by the second device, the training request including training requirement information, the training requirement information being used to indicate the requirements of this training task; and performing model training according to the training request.

[0400] Optionally, the communication method, applied to the second device, includes: generating a training request, the training request including training requirement information, the training requirement information being used to indicate the requirements of this training task; sending the training request to the first device so that the first device performs model training according to the training request.

[0401] The computer storage medium of the embodiment of the present application can adopt any combination of one or more computer-readable media.Computer-readable media can be computer-readable signal media or computer-readable storage media.Computer-readable storage media can be, for example, but not limited to: electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above.More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM), flash memories, optical fibers, portable CD-ROMs, optical storage devices, magnetic storage devices, or any suitable combination of the above.Computer-readable storage media can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0402] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0403] The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0404] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the communication method provided in any embodiment of the present application.

[0405] The computer program code for performing the operations of the present application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and also conventional procedural programming languages ​​such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect via the Internet).

[0406] The above description is merely an exemplary embodiment of the present application and is not intended to limit the scope of protection of the present application.

[0407] It will be appreciated by those skilled in the art that the term user terminal covers any suitable type of wireless user equipment, such as a mobile phone, a portable data processing device, a portable web browser or a vehicle-mounted mobile station.

[0408] In general, various embodiments of the present application may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, although the present application is not limited thereto.

[0409] Embodiments of the present application may be implemented by executing computer program instructions by a data processor of a mobile device, for example, in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages.

[0410] The block diagram of any logic flow in the drawings of this application may represent program steps, or may represent interconnected logic circuits, modules and functions, or may represent a combination of program steps and logic circuits, modules and functions. A computer program may be stored on a memory. The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as but not limited to read-only memory (ROM), random access memory (RAM), optical storage devices and systems (digital versatile discs (DVD) or compact disks (CD), etc.). Computer-readable media may include non-transitory storage media. The data processor may be of any type suitable for the local technical environment, such as but not limited to a general-purpose computer, a special-purpose computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), and a processor based on a multi-core processor architecture.

[0411] The above description of exemplary embodiments of the present application has been provided by way of exemplary and non-limiting examples. However, various modifications and adaptations of the above embodiments will be apparent to those skilled in the art, when considered in conjunction with the accompanying drawings and the appended claims, without departing from the scope of the present application. Therefore, the proper scope of the present application will be determined by reference to the appended claims.

Claims

1. A communication method, characterized in that: Applied to a first device, comprising: receiving a training request sent by a second device, the training request including training requirement information, where the training requirement information is used to indicate requirements for the current training task; Perform model training according to the training request.

2. The communication method according to claim 1, wherein: The training requirement information includes one or more of the following: Model requirement information; training strategies; The model requirement information is used to indicate the requirements of this training, and the training strategy is used to arrange the training plan.

3. The communication method according to claim 2, wherein: The model requirement information includes one or more of the following: Model complexity requirement information; Model performance requirement information; Energy consumption demand information; The model complexity requirement information is used to indicate the requirement of this training for model complexity, the model performance requirement information is used to indicate the requirement of this training for model performance, and the energy consumption requirement information is used to indicate the energy consumption requirement of this training; The training strategies include one or more of the following: Energy consumption-based training strategies; Training strategies based on model complexity; Training strategies based on model performance; Cost-based training strategies; Among them, the energy consumption-based training strategy is used to indicate the scheduling of training plans based on energy consumption, the model complexity-based training strategy is used to indicate the scheduling of training plans based on model complexity, the model performance-based training strategy is used to indicate the scheduling of training plans based on model performance, and the cost-based training strategy is used to indicate the scheduling of training plans based on cost.

4. The communication method according to claim 2, wherein: The training requirement information also includes: Priority information; The priority information includes a priority indication or a priority level indicating whether the task can be preempted.

5. The communication method according to claim 2, wherein: The performing model training according to the training request includes: Determine a training plan based on the training requirement information, wherein the training plan includes one or more of the following: training execution time, training execution nodes, training resources, training environment, and model collaborative training strategy; Performing model training according to the training scheme; Wherein, the training resources are the resources reserved for this training; The model collaborative training strategy indicates a strategy for collaborative training of multiple models. The communication method according to claim 5 , wherein: Also includes: The first device sends a training scheme to the second device.

7. The communication method according to claim 1, wherein: Also includes: After the model training is started, the first device sends a first indication message to the second device, where the first indication message is used to indicate that the model training has been started.

8. The communication method according to claim 1, wherein: Before starting model training, also include: The first device performs a feasibility check on the training task, generates a feasibility check report on the training task, and sends the report to the second device; The training task feasibility check report includes one or more of the following information: Results of feasibility checks on training missions; Auxiliary information, the auxiliary information includes one or more of the following: a reason for the training request suggestion, and training request suggestion information.

9. The communication method according to claim 1, wherein: Also includes: After completing the model training, the first device generates a training report and sends it to the second device; The training report may include one or more of the following information: Training needs report; Model information; The training needs report includes at least one of the following: Model complexity requirement report; Model performance requirements report; Energy consumption demand report; Among them, the model complexity requirement report is used to indicate the model complexity of this training, the model performance requirement report is used to indicate the model performance of this training, and the energy consumption requirement report is used to indicate the energy consumption of this training.

10. The communication method according to claim 1, wherein: Also includes: After completing the model training, the first device saves or updates the model information.

11. The communication method according to claim 1, wherein: Also includes: After completing the model training, the first device sends a model information saving request to the second device, where the model information saving request is used to request whether to save or update the model information; After receiving the save instruction sent by the second device, the first device determines whether to save or update the model information according to the save instruction, where the save instruction is used to instruct the first device whether to save or update the model information.

12. The communication method according to any one of claims 9 to 11, characterized in that: The model information includes one or more of the following: Model energy consumption information; Model complexity information; Model interoperability information; Training background information; Among them, the model energy consumption information is used to indicate the energy consumption required to run the model, the model complexity information is used to indicate the complexity of the model, the model interoperability information is used to indicate the information required for the model to interoperate, and the training background information is used to indicate the background of model training.

13. The communication method according to claim 1, wherein: Also includes: The first device detects or predicts a conflict during model training, generates a conflict report, and sends the report to the second device; The conflict report may include one or more of the following information: Causes of conflict; Training request suggestion information.

14. The communication method according to claim 13, wherein: Also includes: The first device receives second indication information sent by the second device, and processes the training task conflict according to the second indication information, where the second indication information is used to instruct the first device to process the training task conflict; or The first device processes the training task conflict according to a predefined processing strategy.

15. A communication method, characterized in that: Applied to the second device, comprising: Generate a training request, wherein the training request includes training requirement information, and the training requirement information is used to indicate the requirements of the current training task; The training request is sent to the first device, so that the first device performs model training according to the training request.

16. The communication method according to claim 15, characterized in that: The training requirement information includes one or more of the following: Model requirement information; training strategies; The model requirement information is used to indicate the requirements of this training, and the training strategy is used to arrange the training plan.

17. The communication method according to claim 16, wherein: The model requirement information includes one or more of the following: Model complexity requirement information; Model performance requirement information; Energy consumption demand information; The model complexity requirement information is used to indicate the requirement of this training for model complexity, the model performance requirement information is used to indicate the requirement of this training for model performance, and the energy consumption requirement information is used to indicate the energy consumption requirement of this training; The training strategies include one or more of the following: Energy consumption-based training strategies; Training strategies based on model complexity; Training strategies based on model performance; Cost-based training strategies; Among them, the energy consumption-based training strategy is used to indicate the training plan arranged based on energy consumption, the model complexity-based training strategy is used to indicate the training plan arranged based on model complexity, the model performance-based training strategy is used to indicate the training plan arranged based on model performance, and the cost-based training strategy is used to indicate the training strategy arranged based on cost.

18. The communication method according to claim 16, wherein: The training requirement information also includes: Priority information; The priority information includes a priority indication or a priority level indicating whether the task can be preempted.

19. The communication method according to claim 15, wherein: Also includes: The second device receives a training plan sent by the first device, where the training plan includes one or more of the following: training execution time, training execution node, training resources, training environment, and model collaborative training strategy; Wherein, the training resources are the resources reserved for this training; The model collaborative training strategy indicates a strategy for collaborative training of multiple models.

20. The communication method according to claim 15, wherein: Also includes: The second device receives first indication information sent by the first device after model training is started, where the first indication information is used to indicate that model training has been started.

21. The communication method according to claim 15, wherein: Also includes: The second device receives the training task feasibility check report sent by the first device; The training task feasibility check report includes one or more of the following information: Results of feasibility checks on training missions; Auxiliary information, the auxiliary information includes one or more of the following: a reason for the training request suggestion, and training request suggestion information.

22. The communication method according to claim 15, wherein: Also includes: The second device receives the training report sent by the first device; The training report may include one or more of the following information: Training needs report; Model information; The training needs report includes at least one of the following: Model complexity requirement report; Model performance requirements report; Energy consumption demand report; Among them, the model complexity requirement report is used to indicate the model complexity of this training, the model performance requirement report is used to indicate the model performance of this training, and the energy consumption requirement report is used to indicate the energy consumption of this training.

23. The communication method according to claim 22, wherein: Also includes: The second device saves or updates the model information.

24. The communication method according to claim 15, wherein: Also includes: The second device receives a model information saving request sent by the first device, where the model information saving request is used to request whether to save or update the model information; The second device generates a save instruction and sends it to the first device, where the save instruction is used to instruct the first device whether to save or update the model information.

25. The communication method according to any one of claims 22 to 24, characterized in that: The model information includes one or more of the following: Model energy consumption information; Model complexity information; Model interoperability information; Training background information; Among them, the model energy consumption information is used to indicate the energy consumption required to run the model, the model complexity information is used to indicate the complexity of the model, the model interoperability information is used to indicate the information required for the model to interoperate, and the training background information is used to indicate the background of model training.

26. The communication method according to claim 15, wherein: Also includes: The second device receives the conflict report sent by the first device; The conflict report includes at least one of the following information: Causes of conflict; Training request suggestion information.

27. The communication method according to claim 26, wherein: Also includes: The second device generates second indication information and sends it to the first device, where the second indication information is used to instruct the first device to handle the training task conflict.

28. A communication node, characterized in that include: A memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection communication between the processor and the memory, wherein the program, when executed by the processor, realizes the steps of the communication method according to any one of claims 1 to 27.

29. A storage medium for computer-readable storage, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the communication method according to any one of claims 1 to 27.

30. A computer program product, characterized in that The computer program product comprises a computer program which, when executed by a processor, implements the steps of the communication method according to any one of claims 1 to 27.

Images