Method and device for carrying out resource configuration on multi-mode network, equipment and medium

Abstract

The invention provides a method, a device, equipment and a medium for carrying out resource configuration on a multi-mode network, and the method comprises the steps: obtaining the network state information of each network mode in the multi-mode network, the network state information comprising network request information and network environment index information; based on the network state information, running a preset configuration strategy code to generate a configuration strategy for the multi-modal network, and completing optimization of resource configuration of the multi-modal network according to the configuration strategy; wherein the configuration strategy code is generated by performing nested iterative optimization on the basis of a pre-constructed database comprising a plurality of data samples by utilizing a large language model; the network state information and network state input data of a code predictor used for screening codes in the configuration strategy code generation process are from the same network environment. Therefore, the corresponding configuration strategy can be adaptively generated according to the change of the network environment so as to optimize the resource configuration of the multi-modal network, and the network performance and the resource utilization efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a method, apparatus, device, and medium for configuring resources in a multimodal network. Background Art

[0002] A multimodal network (Polymorphic intelligent Network) is a technology that integrates multiple network technologies into the network environment in the form of network modes through programmable technology, striving to integrate the advantages of multiple networks and enhance the network's ability to carry multiple services.

[0003] In current multimodal network environments, network resource allocation is primarily accomplished through manual configuration or fixed-rule allocation strategies. While this approach can achieve resource allocation, it suffers from a coarse control granularity and reliance on user experience, resulting in poor real-time control. This leads to misalignment between the generated configuration strategies and actual network requirements, making them difficult to adapt to dynamically changing network environments. Summary of the Invention

[0004] In view of this, the present application provides a method, apparatus, device, and storage medium for resource configuration in a multimodal network, which can adaptively generate corresponding configuration strategies according to changes in the network environment to optimize the resource configuration of the multimodal network, thereby helping to improve network performance and resource utilization efficiency.

[0005] According to a first aspect of the present application, a method for configuring resources in a multimodal network is provided. The method is applied to a network control agent that configures resources in the multimodal network. The method comprises:

[0006] Acquiring network status information of each network modality in the multimodal network, the network status information including network request information and network environment indicator information;

[0007] Based on the network status information, running a preset configuration policy code to generate a configuration policy for the multimodal network, and optimizing resource configuration of the multimodal network according to the configuration policy;

[0008] The configuration strategy code is generated by nested iterative optimization using a large language model based on a pre-built database including multiple data samples; the network status information and the network status input data of the code predictor used to screen the code during the configuration strategy code generation process come from the same network environment.

[0009] In an optional implementation, the configuration policy code is generated by the following steps:

[0010] Acquire at least one data sample from the database, each data sample including a problem description prompt word and a corresponding reference code example;

[0011] Processing the problem description prompt words in the at least one data sample using a large language model to obtain candidate problem description prompt words, and generating a first number of initial candidate codes based on the candidate problem description prompt words and the reference code examples;

[0012] The initial candidate codes are screened using a code predictor, and when the result of this round of iteration meets a first preset condition, a target candidate data sample is obtained; the target candidate data sample includes a target candidate question description prompt word and a corresponding target candidate reference code;

[0013] generating a second number of intermediate candidate codes based on the target candidate data sample using the large language model;

[0014] Using the code predictor to screen the intermediate candidate codes, and obtaining the optimal code for this round if the result of this round of iteration meets the second preset condition;

[0015] In response to the iterative round not reaching the preset round threshold, the optimal code of this round and the target candidate question description prompt word form a new data sample and add it to the database, repeating the above steps until the iterative round reaches the preset round threshold to obtain the optimal configuration strategy code.

[0016] In an optional embodiment, the second number is greater than the first number.

[0017] In an optional embodiment, the processing of the question description prompt words in the at least one data sample using a large language model to obtain candidate question description prompt words includes:

[0018] The large language model is used to reflect on the problem description ideas according to positive and negative examples, and the problem description prompt words are processed to obtain the candidate problem description prompt words.

[0019] In an optional embodiment, the method further includes:

[0020] If the iteration result does not meet the first preset condition, a new question description prompt word is generated, and the process returns to the step of using the large language model to process the question description prompt words in the at least one data sample to obtain candidate question description prompt words.

[0021] In an optional embodiment, the method further includes:

[0022] When the number of generated question description prompt words reaches a first preset number threshold, it is determined that the iteration result meets the first preset condition.

[0023] In an optional implementation, generating a second number of intermediate candidate codes based on the target candidate data sample using the large language model includes:

[0024] The large language model is used to construct positive and negative example candidate codes according to the target candidate reference codes, and a second number of intermediate candidate codes are generated according to the target candidate question description prompt words and the positive and negative example candidate codes.

[0025] In an optional embodiment, the method further includes:

[0026] If the iteration result does not meet the second preset condition, at least one target candidate reference code is generated, and the process returns to the step of constructing positive and negative example candidate codes based on the target candidate reference code using the large language model.

[0027] In an optional implementation, when the number of generated target candidate reference codes reaches a second preset number threshold, it is determined that the iteration result meets the second preset condition.

[0028] According to a second aspect of the present application, there is provided an apparatus for configuring resources for a multimodal network, comprising:

[0029] An information acquisition module, configured to acquire network status information of each network mode in the multimodal network, wherein the network status information includes network request information and network environment indicator information;

[0030] A resource configuration module, configured to run a preset configuration policy code to generate a configuration policy for the multimodal network based on the network status information, and optimize the resource configuration of the multimodal network according to the configuration policy;

[0031] The configuration strategy code is generated by nested iterative optimization using a large language model based on a pre-built database including multiple data samples; the network status information and the network status input data of the code predictor used to screen the code during the configuration strategy code generation process come from the same network environment.

[0032] In an optional embodiment, the device further includes:

[0033] A data acquisition module, configured to acquire at least one data sample from the database, each data sample including a problem description prompt word and a corresponding reference code example;

[0034] The prompt word processing module is configured to process the question description prompt word in the at least one data sample by using the large language model to obtain a candidate question description prompt word, and generate a first number of initial candidate codes according to the candidate question description prompt word and the reference code example.

[0035] The first code screening module is configured to screen the initial candidate codes by using a code predictor, and obtain a target candidate data sample in a case where a result of a current iteration meets a first preset condition; the target candidate data sample includes a target candidate question description prompt word and a corresponding target candidate reference code.

[0036] The intermediate code generation module is configured to generate a second number of intermediate candidate codes from the target candidate data sample by using the large language model.

[0037] The second code screening module is configured to screen the intermediate candidate codes by using the code predictor, and obtain a current optimal code in a case where a result of a current iteration meets a second preset condition.

[0038] The loop iteration module is configured to add the current optimal code and a target candidate question description prompt word into the database to form a new data sample in response to an iteration round not reaching a preset round threshold, and repeat the above steps until the iteration round reaches the preset round threshold, to obtain an optimal configuration strategy code.

[0039] In an optional implementation, the second number is greater than the first number.

[0040] In an optional implementation, the prompt word processing module is specifically configured to:

[0041] The large language model is used to process the question description prompt word according to a positive and negative example reflection question description idea to obtain the candidate question description prompt word.

[0042] In an optional implementation, the first code screening module is further configured to:

[0043] In a case where the iteration result does not meet the first preset condition, a new question description prompt word is generated, and the step of processing the question description prompt word in the at least one data sample by using the large language model to obtain a candidate question description prompt word is returned to.

[0044] In an optional implementation, the first code screening module is further configured to:

[0045] In a case where a number of generated question description prompt words reaches a first preset number threshold, it is determined that the iteration result meets the first preset condition.

[0046] In an optional implementation, the intermediate code generation module is specifically configured to:

[0047] The large language model is used to construct positive and negative example candidate codes according to the target candidate reference codes, and a second number of intermediate candidate codes are generated according to the target candidate question description prompt words and the positive and negative example candidate codes.

[0048] In an optional implementation, the second code screening module is further configured to:

[0049] If the iteration result does not meet the second preset condition, at least one target candidate reference code is generated, and the process returns to the step of constructing positive and negative example candidate codes based on the target candidate reference code using the large language model.

[0050] In an optional implementation, the second code screening module is further configured to:

[0051] When the number of generated target candidate reference codes reaches a second preset number threshold, it is determined that the iteration result meets the second preset condition.

[0052] According to a third aspect of the present application, an electronic device is provided, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor; when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the method for configuring resources for a multimodal network as described in the first aspect above is performed.

[0053] According to a fourth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for configuring resources for a multimodal network described in the first aspect is executed.

[0054] The method, apparatus, electronic device, and storage medium for resource configuration in a multimodal network provided herein utilize a large language model, generating configuration policy code through nested iterative optimization based on a pre-built database containing multiple data samples. Furthermore, the network state information and the network state input data for the code predictor used to filter the code during the configuration policy code generation process are derived from the same network environment. This allows adaptive generation of configuration policy code to meet dynamic constraints in changing scenarios. Furthermore, adaptive generation of corresponding configuration policies based on changes in the network environment can optimize resource configuration in the multimodal network, thereby improving network performance and resource utilization efficiency.

[0055] In addition, the embodiment of the present application uses a code predictor to perform code screening, which can screen out low-quality candidates at an early stage, reduce computing overhead, and maintain the quality of the generated code.

[0056] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments. The drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without inventive effort.

[0058] Figure 1 A schematic diagram of a multimodal network environment is shown;

[0059] Figure 2a It is a schematic diagram of resource configuration using a fixed rule control strategy in related technologies;

[0060] Figure 2b It is a schematic diagram of resource allocation using the adaptive control method;

[0061] Figure 3 This is a schematic diagram of the structure of a decision system for network modal evolution shown in an exemplary embodiment of the present application;

[0062] Figure 4 This is a flow chart of a method for configuring resources in a multimodal network, shown in an exemplary embodiment of the present application;

[0063] Figure 5 This is a flow chart of a method for generating a configuration policy code shown in an exemplary embodiment of the present application;

[0064] Figure 6 is a flowchart of another method for generating a configuration policy code according to an exemplary embodiment of the present application;

[0065] Figure 7a is a schematic diagram showing the time consumption of TSP problem instances of different scales according to an exemplary embodiment of the present application;

[0066] Figure 7b 1 is a schematic diagram showing a comparison of standardized errors based on an MKP100 instance, as shown in an exemplary embodiment of the present application;

[0067] Figure 7c is a comparative diagram of a standardization error based on a TSP 200 instance, according to an example embodiment of the present application;

[0068] Figure 8 is a functional module diagram of an apparatus for resource configuration of a multi-modal network, according to an example embodiment of the present application;

[0069] Figure 9 is a functional module diagram of another apparatus for resource configuration of a multi-modal network, according to an example embodiment of the present application;

[0070] Figure 10 is a structural diagram of an electronic device, according to an example embodiment of the present application. DETAILED DESCRIPTION

[0071] The example embodiments will now be described in detail with reference to the accompanying drawings. If the description of the example embodiments refers to accompanying drawings, then the description is illustrative of the example embodiments and does not limit the present application. Thus, it should be understood that various modifications and changes can be made to the implementation of the present application without departing from the scope of the application. Furthermore, the present application covers all such modifications and variations as fall within the scope of the disclosure. It is to be understood that the following description is only exemplary of the various aspects of the present application, and is intended to provide an overview or framework for understanding the nature and character of the application as it is claimed.

[0072] The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in the description of the application and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It also will be understood that the term “and / or” as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0073] It is to be understood that the singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise. It will be further understood that the terms “comprises” and / or “comprising,” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is to be understood that the terms “including,” “comprising,” “consisting” and “consisting essentially of,” when used herein, specify the presence of stated features, integers, steps, operations, elements, and / or components), but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0074] The term "and / or" herein simply describes an association relationship, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, and the existence of B alone. In addition, the term "at least one" herein refers to any combination of at least two of any one or more of a plurality of items. For example, "at least one of A, B, and C" can represent any one or more elements selected from the set consisting of A, B, and C.

[0075] See Figure 1 , Figure 1 A schematic diagram of a multimodal network environment is shown. Figure 1 As shown, the network multimodal symbiosis and evolution target model, based on the full lifecycle scalability of infrastructure (multimodal network elements), introduces the network modality diversity (V) metric in addition to the traditional network service quality commitment (S) and resource multiplexing (M) metrics. It also proposes a three-aspect architecture for network modalities: evolution, generation, and symbiosis. Specifically, network modalities within the same infrastructure share and collaborate on underlying resources, forming a symbiotic state that continuously evolves with service and resource conditions. For example, the evolutionary aspect focuses on the dynamic adjustment of network modalities to changes in service and resources, including resource allocation (scaling / scaling) and survival (generation / optimization / extinction). The generation aspect focuses on the creation mechanism of new network modalities, which must be compatible with underlying resources. The symbiosis aspect emphasizes the sharing of resources and collaborative operation of multiple modalities within the same infrastructure, forming a collaborative state.

[0076] Furthermore, a two-dimensional design approach is implemented for model evolution. At the micro level, evolutionary timing and strategies can be determined through cooperative game deduction. For example, when the resource utilization of a particular modality exceeds a threshold, a resource reallocation game between modalities is triggered to balance the performance requirements of each modality. At the macro level, the evolutionary path can be regulated through quantitative assessment (e.g., establishing a mathematical model for SMV indicators) with the goal of achieving "SMV optimization in the joint spatiotemporal dimension." For example, resource allocation can be dynamically adjusted in different time periods (spatiotemporal dimensions) to ensure the optimal combination of S (quality), M (reuse), and V (diversity) throughout the entire lifecycle. Furthermore, the combination of these two approaches can provide a strong guarantee for achieving a complete SMV intersection across the entire lifecycle of multimodal networks. For example, this can address resource waste or performance bottlenecks in traditional single-modality networks, improve overall efficiency through multimodal collaboration, adapt to dynamic business needs (such as hybrid cloud and edge computing scenarios), and achieve network self-optimization and self-adaptation through evolutionary mechanisms.

[0077] Research has found that in a multimodal network environment, faced with complex constraints posed by heterogeneous networks and dynamic business demands, there is a lack of control algorithms that can adapt to the huge search space composed of multiple constraint variable combinations, resulting in waste of resources and poor service quality.

[0078] Specifically, see Figure 2a and Figure 2b ,in, Figure 2a A schematic diagram of resource allocation using a fixed rule control strategy in related technologies; Figure 2b This diagram illustrates resource allocation using adaptive control methods. Related network control methods primarily focus on static networks, often relying on manual configuration or fixed-rule deployment strategies. These methods exhibit coarse control granularity and reliance on user experience, resulting in poor real-time performance. This leads to misalignment between the generated configuration policies and actual network requirements, making them difficult to adapt to dynamically changing network environments. Furthermore, the evaluation of candidate policies relies on actual deployment feedback, requiring significant computing resources and impacting network service quality.

[0079] That is to say, the technical problems in the related art are mainly the following:

[0080] 1) Modality misconfiguration under complex network constraints: Network control methods in related technologies are based on static rules or manual experience. They are difficult to adapt to complex constraint scenarios in multimodal networks where heterogeneous modalities coexist and service demands change dynamically (such as cross-modal resource competition and inter-modal service dependency conflicts). This leads to significant deviations between modality configuration strategies and actual network status, resulting in resource fragmentation and service quality fluctuations.

[0081] 2) Inefficient optimization caused by blind search and verification during the evolution process: Related technologies evaluate candidate strategies through exhaustive verification, requiring complete deployment and testing of all modal combinations. The evaluation overhead increases exponentially with the number of modalities. During the evolution of modal states (e.g., generation / optimization / extinction), there is a lack of an early screening mechanism for inefficient strategies, resulting in a blind search process and difficulty meeting real-time requirements.

[0082] 3) The evolution process lacks closed-loop feedback and adaptive capabilities: Related technologies use an open-loop strategy generation model and fail to establish a mechanism to link historical experience with current evolution. As a result, the modal configuration strategy is prone to falling into local optimality in dynamic networks and cannot adaptively adjust the evolution path based on changes in network status, which restricts the full life cycle optimization capability of multimodal networks.

[0083] Based on the above research, the embodiment of the present application provides a nested evolution mechanism (NeRM, Nested-Refinement Metamorphosis), which solves the technical problems of poor adaptability, low efficiency, unstable quality and the like in the related art in a complex constraint scene such as a multi-modal network scene by constructing a closed-loop optimization system including dynamic constraint modeling, intelligent evaluation screening and structured evolution.

[0084] Specifically, the nested evolution mechanism provided by the embodiment of the present application can improve the solving of a complex constraint combination optimization problem, and is used for evolving a network modal configuration strategy and dynamically regulating and controlling the configuration of network modal resources, such as the resource ratio of each network modal on a network element and the survival state of the network modal in a network. The mechanism forms a self-improving feedback loop by iteratively refining the description of the evolution task and the constraint problem and the algorithm solution, and realizes efficient optimization of the network modal evolution. Meanwhile, a predictor is introduced to imitate natural selection, filter out weak candidate evolution strategies at an early stage, reduce the calculation cost, improve the evolution efficiency, and finally provide protection for efficient use of network resources and service quality.

[0085] It should be noted that the nested evolution mechanism described above can be understood as an optimization process for generating code of a configuration strategy for a multi-modal network. Referring to Figure 3 as shown, Figure 3 A structural schematic diagram of a decision system for network modal evolution provided by the embodiment of the present application is shown. Exemplarily, the configuration strategy generation method can be applied to a network control intelligent agent in a multi-modal network evolution aspect. Specifically, based on the service information of the environment service and the dynamic change information of the network state, the scheduling algorithm (also referred to as code) currently used to generate the network configuration strategy can be self-adapted to the dynamic environment constraints and self-improved through SMV index analysis.

[0086] It can be understood that there are various identification systems in a multi-modal network, such as content identification, identity identification, and geographic space identification, and the network architecture model formed based on each identification system is a network modal, such as an IP network, a named data network, and a geographic location network.

[0087] This configuration policy code generation method can be applied to any environment that supports multimodal networks, such as geographic identification network modality, industrial control identification network modality, identity identification network modality, content identification network modality, IP (Internet Protocol) identification network modality, etc. By upgrading the network elements of these networks to intelligent network elements with multi-network modality processing capabilities and implanting a network modality evolution decision framework (a model for optimizing code) in the top-level controller, changes in the network environment can be perceived in real time, and the solution algorithm can be adaptively generated and evolved according to the dynamic changes in the constraint relationship to improve the quality of the evolution strategy. In this way, precise regulation and long-term evolution management of network modalities can be achieved, network performance and resource utilization efficiency can be improved, and operating costs can be reduced. In addition, users can also obtain better service quality guarantees.

[0088] Among them, the geographic identification network modality can use the actual geographic longitude and latitude as the message forwarding identifier to address and communicate with all devices within the specified range, thereby realizing business needs based on geographic location addressing; the content identification network modality can use the content name as the forwarding identifier and cache the content through switching nodes, thereby making data transmission faster and improving content retrieval efficiency; the identity identification network modality can design the network architecture based on the idea of ​​separating identity and location, fundamentally solving the problem of poor mobility of traditional IP networks; the industrial control identification network modality can address and forward based on the industrial control identity, providing low-latency and low-jitter real-time communication capabilities for industrial control equipment and services; the IP identification network modality can address and route communications based on traditional IP addresses.

[0089] The following describes in detail the method for configuring resources in a multimodal network provided by the embodiment of the present application with reference to the accompanying drawings.

[0090] See also Figure 4 FIG. 1 is a flowchart of a method for configuring resources for a multimodal network provided in an embodiment of the present application. The method can be applied to a network control agent that configures resources for the multimodal network. The method includes the following steps S101 to S102:

[0091] S101: Acquire network status information of each network modality in the multimodal network, where the network status information includes network request information and network environment indicator information.

[0092] For example, network status information may include network request information (service input) and network environment indicator information. Network request information refers to information about service requests that require services and are actually deployed, such as the service source, source and destination nodes, network modality to which the service belongs, bandwidth requirements, computing, storage, and forwarding resource requirements, and quality of service requirements such as latency thresholds. Network environment indicator information refers to the current network operation status, such as the total node resource volume and occupancy rate, link bandwidth occupancy and actual latency, topological relationships, and modality service status, such as quantity and resource usage.

[0093] It is understandable that the network request information and network environment indicator information may vary according to the actual environment and are not specifically limited. For example, the network request information may also include service input and service quality requirements under each of the network modes.

[0094] S102: Based on the network state information, execute preset configuration strategy code to generate a configuration strategy for the multimodal network, and optimize the resource configuration of the multimodal network according to the configuration strategy. The configuration strategy code is generated by performing nested iterative optimization using a large language model based on a pre-built database comprising multiple data samples. The network state information and the network state input data of a code predictor used to screen code during the configuration strategy code generation process are derived from the same network environment.

[0095] Among them, the configuration strategy code can be deployed in Figure 3 In this way, a configuration strategy for the multimodal network can be generated based on the received network status information. Specifically, the configuration strategy refers to the specific ratio of network modes at each node, which can be a specific numerical result. For example, the white configuration strategy may include: the three modal resource allocation ratios on node 1 are [0.2, 0.2, 0.6].

[0096] The method for configuring resources in a multimodal network provided in this application utilizes a large language model and nested iterative optimization to generate configuration policy code based on a pre-built database containing multiple data samples. Furthermore, the network state information and the network state input data for the code predictor used to filter the code during the configuration policy code generation process are derived from the same network environment. This allows adaptive generation of configuration policy code to meet dynamic constraints in changing scenarios. Furthermore, adaptive generation of corresponding configuration policies based on changes in the network environment can optimize resource configuration in the multimodal network, thereby improving network performance and resource utilization efficiency.

[0097] The following is a detailed description of the generation process of the configuration policy code.

[0098] See also Figure 5 FIG. 2 is a flow chart of a method for generating a configuration policy code according to an embodiment of the present application, wherein the method includes the following steps S201 to S206:

[0099] S201 , obtaining at least one data sample from the database, each of the data samples including a problem description prompt word and a corresponding reference code example.

[0100] For example, a problem description prompt can be constructed based on user-preconfigured task description content. The task description content describes the target problem to be solved, such as the content included in the optimization goal and the content included in the constraint goal. For example, feedback information such as focusing on computing / storage / forwarding resources on a specific node, focusing on modal service fairness, and paying attention to high-traffic services to be deployed near a node can be used.

[0101] For example, the task description might be, "Given an edge distance matrix and an approximate local optimal route, please help design a strategy for updating the distance matrix to avoid falling into the local optimum, with the ultimate goal of finding a candidate path with the shortest distance. Create a heuristic method to update the edge distance matrix." Furthermore, corresponding reference code examples must be constructed based on the task description.

[0102] S202: Process the problem description prompt words in the at least one data sample using a large language model to obtain candidate problem description prompt words, and generate a first number of initial candidate codes based on the candidate problem description prompt words and the reference code examples.

[0103] Among them, Large Language Model (LLM) refers to a deep learning model trained using a large amount of text data, which can generate natural language text or understand the meaning of language text.

[0104] Here, a large language model can be used to improve the problem description prompt words in any data sample to obtain candidate problem description prompt words, and a first number of initial candidate codes can be pre-generated based on the candidate problem description prompt words and reference code examples. The first number can be a relatively small number. For example, the first number can be 8, 10, or 15.

[0105] In some embodiments, when using a large language model to process question description prompt words in a data sample to obtain candidate question description prompt words, the following steps may be included:

[0106] The large language model is used to reflect on the problem description ideas according to positive and negative examples, and the problem description prompt words are improved to obtain the candidate problem description prompt words.

[0107] Here, the large language model is used to improve the problem description prompt words according to the positive and negative examples to reflect on the problem description ideas, so that better and worse samples can be obtained. This enables the model to better distinguish between good and bad, analyze improvement ideas, and help improve the effect of improvement.

[0108] S203, using a code predictor to screen the initial candidate codes, and obtaining a target candidate data sample when the result of this round of iteration meets a first preset condition; the target candidate data sample includes a target candidate question description prompt word and a corresponding target candidate reference code.

[0109] Specifically, a code predictor can be used to embed the coding and score the initial candidate codes. Initial candidate codes with lower scores will be eliminated, while initial candidate codes with higher scores can continue to be evaluated for their performance indicators for further screening.

[0110] Here, to alleviate the computational load brought by large-scale candidate evaluation, a code predictor is used for code screening. This can screen out low-quality initial candidate codes, achieve early elimination of inferior candidates, and concentrate limited computing resources on the evaluation of high-potential codes, thereby reducing computational overhead while maintaining the quality of the subsequently generated configuration policy codes.

[0111] Exemplarily, when the iteration result does not meet the first preset condition, a new question description prompt word is generated, and the process returns to the step of using the large language model to process the question description prompt words in the at least one data sample to obtain candidate question description prompt words.

[0112] When the number of generated question description prompt words reaches a first preset threshold, the iteration result is determined to meet the first preset condition. For example, the first preset threshold can be set to 10, 12, 15, 20, etc., without specific limitation.

[0113] It should be noted that if a new problem description prompt word is generated in each iteration, then generating 15 problem description words means that 15 iterations are required. When the number of iterations reaches 15, the first preset condition is determined to be satisfied. In other words, the first preset condition is associated with the number of iterations and the number of problem description words generated each time.

[0114] S204: Generate a second number of intermediate candidate codes according to the target candidate data samples using the large language model.

[0115] Here, when generating a second number of intermediate candidate codes according to the target candidate data sample using the large language model, the method may include:

[0116] The large language model is used to construct positive and negative example candidate codes according to the target candidate reference codes, and a second number of intermediate candidate codes are generated according to the target candidate question description prompt words and the positive and negative example candidate codes.

[0117] The second number is greater than the first number. For example, the second number may be 30, 35, 40, etc., without limitation.

[0118] Similarly, by constructing positive and negative samples here, we can obtain better and worse samples, which can enable the model to better distinguish between good and bad, analyze improvement ideas, and help improve the effect of improvement.

[0119] S205 , using the code predictor to screen the intermediate candidate codes, and obtaining the optimal code for this round if the result of this round of iteration meets a second preset condition.

[0120] Here, the process of using the code predictor to screen the intermediate candidate codes is similar to the process of using the code predictor to screen the initial candidate codes, and will not be repeated here.

[0121] In some embodiments, when the iteration result does not meet the second preset condition, multiple target candidate reference codes are generated, and the process returns to the step of using the large language model to construct positive and negative example candidate codes based on the target candidate reference codes.

[0122] Exemplarily, the iteration result may be determined to satisfy the second preset condition when the number of generated target candidate reference codes reaches a second preset number threshold. For example, the second preset number threshold may be 260, 280, or 300, etc., without specific limitation. Similarly, similar to the first preset condition, the second preset condition may be related to the number of target candidate reference codes generated each time and the number of iterations.

[0123] In step S206, in response to the iterative round not reaching the preset round threshold, a new data sample consisting of the optimal code of this round and the target candidate question description prompt word is added to the database, and the above steps are repeated until the iterative round reaches the preset round threshold, thereby obtaining the optimal configuration strategy code.

[0124] It can be understood that in the iteration process, multiple rounds of iterations can be set, and the corresponding number of iterations can be set for each round of iteration. Through iteration, the performance of the configuration strategy code can be improved. In addition, after the iteration round reaches the preset round threshold, not only the optimal configuration strategy code can be obtained, but also the corresponding optimal problem description prompt word can be obtained, and the optimal problem description prompt word and the optimal configuration strategy code can be used as the starting point for the next optimization code, thereby improving the code quality of the next optimization and improving the optimization efficiency.

[0125] For a better understanding of the relationship of iteration, please refer to Figure 6 , wherein, Figure 6 A flowchart of a code generation method according to another embodiment of the present application is provided, and the method comprises the following S301-S315:

[0126] S301, at least one data sample is obtained from the database, and each data sample comprises a problem description prompt word and a corresponding reference code example.

[0127] This step is similar to step S201, which will not be repeated here.

[0128] S302, using a large language model to process the problem description prompt word in at least one data sample to obtain a candidate problem description prompt word.

[0129] S303, using a large language model to generate a first number of initial candidate codes according to the candidate problem description prompt word and the reference code example.

[0130] Steps S302 and S303 are similar to step S202, which will not be repeated here.

[0131] S304, using a code predictor to screen the initial candidate code.

[0132] S305, determining whether the current iteration result meets the first preset condition; if yes, executing step S307; if no, executing step S306.

[0133] S306, generating a new problem description prompt word.

[0134] S307, obtaining a target candidate data sample; the target candidate data sample comprises a target candidate problem description prompt word and a corresponding target candidate reference code.

[0135] The process of steps S303-S307 is similar to that of step S203, and the similarities can be referred to the foregoing description.

[0136] S308, using the large language model to construct positive and negative example candidate codes according to the target candidate reference code.

[0137] S309: Generate a second number of intermediate candidate codes using the large language model according to the target candidate question description prompt words and the positive and negative example candidate codes.

[0138] The process of steps S308 to S309 is similar to the process of step S204 , and the similarities can be referred to the above description.

[0139] S310: Using the code predictor to screen the intermediate candidate codes.

[0140] S311, determine whether the result of this round of iteration meets the second preset condition; if so, execute step S313; if not, execute step S312.

[0141] S312: Generate multiple target candidate reference codes.

[0142] S313, determine whether the iteration round reaches the preset round threshold; if so, execute step S314; if not, execute step S315.

[0143] S314, obtaining the optimal configuration strategy code.

[0144] The process of steps S310 to S314 is similar to the process of step S205 , and the similarities can be referred to the above description.

[0145] S315: Add the optimal code of this round and the target candidate question description prompt words into a new data sample and add it into the database.

[0146] The processes of steps S313 and S315 are similar to the process of step S206 , and the similarities can be referred to the above description.

[0147] It should be understood that the relevant technical explanations and descriptions in this embodiment can refer to Figure 5 The embodiments are described in detail here.

[0148] The method for generating configuration policy code provided by the present application can improve the problem description prompt words of the task during the process of generating the configuration policy code, and generate code after the problem description prompt words are improved, and the code is screened, and then based on the screened code and the improved prompt words, a data sample is generated again, and the next round of iteration is performed until the optimal configuration policy code is obtained. In this way, by improving the problem description to adaptively fit the dynamic constraints under the changing scenarios and adaptively generating code, the adaptability of code generation can be improved. Furthermore, through nested iteration (regenerating data samples for the next round of iteration), it can continuously adapt to task requirements and improve quality. For example, when the network mode persists and expansion and contraction occur, the appropriate policy generation algorithm (configuration policy code) can be adjusted in time to generate resource control strategies for each network mode on the node, which is conducive to improving network performance and resource utilization efficiency.

[0149] In some embodiments, the code generation method can be implemented by building a target model. The target model can include a task description evolution (Metamorphosis On Prompts, MOP) module and an algorithm evolution (Metamorphosis On Algorithms, MOA) module. Specifically, Figure 6 As shown, the MOP module may be responsible for executing steps S301 to S305 , and the MOA module may be responsible for executing steps S308 to S313 .

[0150] Among them, loops A and B are the iterative processes within the module of the same iteration round.

[0151] For example, for loop A in the MOP, a problem description can be generated each time (S306), and then some code can be generated on a small scale based on this description (S302, S303); the condition for judging whether to jump out of the module is whether the number of problem descriptions generated by the MOP module in this round has reached the upper limit. For example, it can be set to 12 descriptions (the first preset condition), and code is generated 10 times for each description. For loop B in the MOA module, a batch of code (such as 30) can be generated each time, and the module will be jumped out when the upper limit of the number of code generation reaches 300 times (the second preset condition).

[0152] The C loop is the primary embodiment of nested optimization and the core of the interaction logic between the MOP and MOA modules. For the entire target model, an iteration cycle begins with the current iteration, then enters the MOP module and generates problem description prompts through its internal A iteration, then enters the MOA module and uses the target candidate problem description prompts generated by the MOP module to perform large-scale B code iterations to identify the optimal code for that description, then enters the next iteration, then re-enters the MOP module. This cycle repeats until the number of iterations reaches the preset threshold, resulting in the optimal configuration strategy code.

[0153] The MOP module simulates the first stage of biological metamorphosis, optimizing problem description prompts to generate multiple variants to improve the guidance of solution generation. The MOA module simulates the second stage of biological metamorphosis and is responsible for optimizing the algorithm itself. Specifically, based on the target candidate problem description prompts provided in the first stage, it generates multiple intermediate candidate codes through a large language model and performs crossover and mutation operations. The nested interaction between the two simulates the third stage, driving the co-evolution of the task and the algorithm to improve the effectiveness of the final solution.

[0154] Understandably, this nested optimization process effectively alleviates the pressure of the related technologies that rely on manual and fixed rules to control quality, allowing the solution ideas (problem descriptions) to adaptively fit the dynamic constraints in changing scenarios, and thus optimize the task description while generating code. Specifically, the task description evolution process (the processing process of the MOP module) in this application corresponds to the growth and accumulation stage, which iteratively optimizes the task description to guide the search more effectively; the algorithm evolution process (the processing process of the MOA module) corresponds to the structural deformation stage, that is, iteratively optimizes the generated code to continuously adapt to the task requirements and improve its quality.

[0155] It should be understood that for this target model, both the MOP module and the MOA module are built on a heterogeneous initial population (database). The MOP module achieves an organic unity of human experience and generative diversity by integrating manually designed templates with prompt sequences generated autonomously by the model. The MOA module generates an initial algorithm library based on the prompt set output by the MOP module and uses a generator driven by a large language model to lead the search process. In the reflection loop, the large language model with reflection capabilities will analyze the historical evolution trajectory, extract key experiences, and generate strategic improvement suggestions to ensure that the optimization path continues to converge to the historical performance boundary.

[0156] After entering the evolution stage, the MOP module and the MOA module perform differentiated genetic operations: the MOP module focuses on structural recombination and semantic optimization of the prompt template, and the MOA module generates a variant library based on the existing algorithm skeleton. In periodic iterations, the performance of the candidate scheme is verified, and the survival of the fittest is completed in combination with the code predictor mechanism - high-quality schemes with development potential are retained, and the evolution of inefficient candidates is terminated.

[0157] The collaborative optimization system realizes bidirectional evolution through multiple iterations: after generating the original prompt library and algorithm population, the MOP module continuously refines the task description paradigm through historical feedback, improving its semantic expression accuracy and robustness; the MOA module uses the optimized prompt to drive algorithm genetic operations, simulating the natural selection mechanism to complete population update. The two modules form a closed loop feedback in the evolution period, driving the prompt representation and algorithm implementation to approach the global optimal solution.

[0158] For a multi-modal network, after obtaining the configuration policy code, the configuration policy code can be deployed on the network controller to obtain real-time state data in the network, such as the use of computing / storage / forwarding resources of each network mode on each node, and the current service input and quality of service. Through model prediction and analysis, it can generate resource configuration policies that adapt to the current constraint environment in a short time, and can trigger adjustments to the survival state of network modal through prompt iteration optimization over a long timeline. Ultimately, the resource configuration policy of each mode on the node will be calculated by the configuration policy code evolved by the model iteration. In this way, it can adapt to the changes in dynamic demand and constraint environment in a timely and adaptive manner, ensure resource utilization and service quality in the network, and improve system stability and efficiency.

[0159] In some embodiments, to improve the prediction performance of the code predictor and adapt the generated configuration policy code to the network environment, for the code predictor, an embedding space-based performance prediction method can be used to realize performance estimation based on execution trajectory simulation by constructing joint vector representation of code-task data.

[0160] In specific implementation, a structured training corpus can be constructed first, and a paired comparison strategy is used to generate training samples. In the encoding stage, the candidate code fragments and task description information are respectively vectorized as follows:

[0161] E i = Enc(code i ), D j = Enc(data j )

[0162] Wherein, code i represents code, data jIndicates the network status input data corresponding to the task description information (actually also the network status information). It can be understood that the network status input data, as the input data of the code predictor, comes from the same network environment as the aforementioned network status information, so that the filtered code can be more adaptable to the current network environment.

[0163] The encoded features are concatenated and input into a multi-layer perceptron (MLP) to determine which one has better performance.

[0164] p ij =MLP(E i ||D i ||E j ||D j )

[0165] Among them, p ij It represents the probability that the optimality of code i is higher than the optimality of code j.

[0166] Specifically, according to the code i Embedding and data i The combination of code-data sample pairs represents the fitness of the code when running on the data. This method is also used when training the MLP model, using (E i ||D i ) and (E j ||D j ) are embedded in two sample pairs, the two are concatenated, and the model is trained based on the comparison results of their fitness. If the first pair is better, the label is 1, otherwise it is 0, thus constructing a binary classification task.

[0167] The optimization objective is the binary cross-entropy loss function. After prediction is complete, the top K candidates are selected for full validation. The validation results are used to further fine-tune the code predictor parameters to improve prediction accuracy. The final high-quality algorithm (code) is retained as the output.

[0168] The code generation method and configuration strategy generation method provided in the embodiments of the present application, on the one hand, adopt a task description evolution mechanism driven by reflection: using the MOP module, a reflective large language model is introduced to summarize and rewrite historical prompts (Prompt), so as to achieve continuous optimization of task descriptions, so as to solve the problems of fixed prompts, weak expression ability and poor guidance in traditional methods. Through large-scale model reflection and automatic evolution of descriptive analysis of current needs and constraints and optimization ideas generated by iterative algorithms, such as paying attention to computing / storage / forwarding resources on specific nodes, focusing on fairness of modal business services, paying attention to feedback information such as high-traffic services to be deployed near a node, etc., code generation is provided with guiding ideas such as balanced allocation of resources for each modality, increasing the allocation ratio of specific modal resources near a node, and continuously evolving and updating prompts to improve the adaptability to the dynamic constraints of multimodal networks.

[0169] Secondly, a code generation mechanism based on collaborative evolution is employed. Leveraging the MOA module, the system dynamically evolves generation strategies by introducing reflection, crossover, and mutation operations on candidate codes. This, combined with the optimization of prompt words, forms a nested collaborative feedback mechanism, enhancing the diversity and task adaptability of code generation and flexibly adjusting the strategy generation process under changing environments. Based on the constrained problem task descriptions provided by the MOP, the effectiveness of the strategy generation algorithm is continuously adjusted and optimized to maintain efficient system operation.

[0170] Third, a candidate screening mechanism assisted by a code predictor is adopted: by using the code predictor, by learning the embedded representation of candidate code and task data, the performance of the code can be quickly estimated and low-quality candidates can be eliminated in advance, thereby significantly reducing the number of actual execution evaluations, improving the overall computing efficiency of the system and reducing resource consumption.

[0171] Based on the above three aspects, this code generation method breaks through the limitations of related technologies in terms of guidance, efficiency and scalability, thereby adapting to the game competition of multiple network modes in the same infrastructure-constrained environment, and timely adjusting the appropriate strategy generation code when the network mode survives and expands and contracts, generating resource control strategies for each network mode on the node.

[0172] The following describes the performance of the configuration strategy code generated by the code generation method of the present application.

[0173] In terms of actual testing, in order to test the effect of the target model, the traffic control and resource scheduling problem common in the multi-modal network can be mapped to the classic combinatorial optimization problem. Among them, the Travel Salesman Problem (TSP) is a typical routing problem, whose purpose is to find the shortest loop that traverses all nodes only once, which has important significance in network traffic control, such as improving information flow efficiency, reducing network congestion, and enhancing system stability. The Multiple Knapsack Problem (MKP) is a packing problem, whose core is to select some items to maximize the total value without exceeding the capacity of the knapsack, which has a wide range of applications in network resource allocation, such as adjusting resource allocation to maximize service volume under limited resources, improving resource utilization and service quality in the network, and reducing packet loss rate.

[0174] In terms of performance, as shown in Table 1, the present application presents significant technical leadership in both TSP and MKP standard combinatorial optimization problems. Specifically, in the TSP200 benchmark test, the present application controls the solution error to the order of 0.19%, which further breaks through compared with the current optimal technology ReEvo scheme (0.23%) and EoH scheme (0.36%). When the test scale is expanded to TSP500 and TSP1000, the present application achieves error indicators of 0.77% and 1.31% respectively, continuously maintaining the optimal algorithm performance. In the MKP300 and MKP500 test scenarios, the present application obtains total revenue values of 61.13 and 101.56 respectively, surpassing the corresponding 60.27 and 99.78 of the existing APR algorithm, fully verifying the stable generalization performance and optimal solution generation efficiency of the present application across scenarios and scales.

[0175] Table 1

[0176]

[0177] Among them, GLS stands for Guided Local Search; KGLS stands for Knowledge Guided Local Search; EBGLS stands for Elite Biased Guided Local Search; EEPE stands for Exploring the Effectiveness of Prompt Engineering; CoAPT stands for Context Attribute words for Prompt Tuning; APR stands for Adaptive Prompt Routing; EoH stands for Evolution of Heuristics; ReEvo stands for Reflective Evolution; VNS stands for Variable Neighborhood Search; TS stands for Tabu Search; ACO stands for Ant Colony Optimization; and NeRM stands for the application of the code generation method of this application.

[0178] See also Figure 7a-7c As shown, Figure 7a It is a schematic diagram of the time consumption of TSP problem instances of different sizes. Figure 7b This is a comparison diagram of the standardized error based on the MKP100 instance. Figure 7c The figure below shows a comparison of the standardized error based on the TSP200 instance. TSP200 refers to a TSP instance with a problem size of 200. For example, each TSP instance contains 200 city nodes. The standardized error is the normalized error from the optimal solution.

[0179] In terms of computational efficiency, by building a code predictor-assisted evaluation mechanism, this application can effectively reduce the computational load brought by the complete execution of all candidate solutions, compressing the average solution time to 83 seconds, a significant improvement over the 98 seconds of the ReEvo solution and the 124 seconds of the EoH solution. As shown in the analysis, for example, Figure 7aThe figure shows the evolution time required for different methods to reach the same error value relative to the optimal solution as the problem scale increases. Here, the exponential growth of this application is significantly reduced compared to other methods. For example, when processing TSP problems of different scales, the computational time growth curve of this application shows an approximately linear characteristic, while the comparative technical solution shows an exponential growth trend, which fully verifies the efficient scalability of the architecture of this application. In addition, the integration of the code predictor enables the evaluation process of invalid candidate algorithms to be intelligently skipped, significantly improving the utilization efficiency of computing resources.

[0180] Combined with the above experimental results, this application comprehensively surpasses relevant technical solutions in the two core indicators of "solution accuracy" and "computational efficiency", and shows significant advantages in multiple dimensions such as algorithm accuracy, model generalization and system scalability. It can adapt to dynamic constraint scenarios and meet the needs of multimodal network scheduling strategies.

[0181] Those skilled in the art will understand that in the above-mentioned method of the specific implementation method, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0182] Based on the same technical concept, the embodiments of the present disclosure also provide an apparatus for configuring resources for a multimodal network corresponding to the method for configuring resources for a multimodal network. Since the principle of solving the problem by the apparatus in the embodiments of the present disclosure is similar to the method for configuring resources for a multimodal network in the embodiments of the present disclosure, the implementation of the apparatus can refer to the implementation of the method, and the repeated parts will not be repeated.

[0183] Reference Figure 8 FIG. 4 is a schematic diagram of an apparatus for configuring resources for a multimodal network according to an embodiment of the present disclosure. The code generating apparatus 400 includes:

[0184] An information acquisition module 401 is configured to acquire network status information of each network mode in the multimodal network, wherein the network status information includes network request information and network environment indicator information;

[0185] A resource configuration module 402 is configured to run a preset configuration policy code to generate a configuration policy for the multimodal network based on the network status information, and optimize the resource configuration of the multimodal network according to the configuration policy;

[0186] The configuration strategy code is generated by nested iterative optimization using a large language model based on a pre-built database including multiple data samples; the network status information and the network status input data of the code predictor used to screen the code during the configuration strategy code generation process come from the same network environment.

[0187] In an alternative embodiment, see Figure 9 As shown, the device also includes:

[0188] A data acquisition module 403 is configured to acquire at least one data sample from the database, each data sample including a problem description prompt word and a corresponding reference code example;

[0189] a prompt word processing module 404 for processing the problem description prompt words in the at least one data sample using a large language model to obtain candidate problem description prompt words, and generating a first number of initial candidate codes based on the candidate problem description prompt words and the reference code examples;

[0190] A first code screening module 405 is configured to screen the initial candidate codes using a code predictor and obtain a target candidate data sample if the result of this iteration round satisfies a first preset condition; the target candidate data sample includes a target candidate question description prompt word and a corresponding target candidate reference code;

[0191] an intermediate code generating module 406 , configured to generate a second number of intermediate candidate codes based on the target candidate data samples using the large language model;

[0192] A second code screening module 407 is configured to screen the intermediate candidate codes using the code predictor and obtain the optimal code for this round if the result of this round of iteration meets a second preset condition;

[0193] The loop iteration module 408 is used to add a new data sample composed of the optimal code of this round and the target candidate question description prompt word into the database in response to the iteration round not reaching the preset round threshold, and repeat the above steps until the iteration round reaches the preset round threshold to obtain the optimal configuration strategy code.

[0194] In an optional embodiment, the second number is greater than the first number.

[0195] In an optional implementation, the prompt word processing module 404 is specifically configured to:

[0196] The large language model is used to reflect on the problem description ideas according to positive and negative examples, and the problem description prompt words are processed to obtain the candidate problem description prompt words.

[0197] In an optional implementation, the first code screening module 405 is further configured to:

[0198] In a case where the iteration result does not satisfy the first preset condition, a new question description prompt word is generated, and the step of processing the question description prompt word in the at least one data sample by using the large language model to obtain a candidate question description prompt word is returned to.

[0199] In an optional implementation, the first code screening module 405 is further configured to:

[0200] In a case where the number of generated question description prompt words reaches a first preset number threshold, it is determined that the iteration result satisfies the first preset condition.

[0201] In an optional implementation, the intermediate code generation module 406 is specifically configured to:

[0202] The large language model is used to construct positive and negative example candidate codes according to the target candidate reference code, and second numbers of intermediate candidate codes are generated according to the target candidate question description prompt word and the positive and negative example candidate codes.

[0203] In an optional implementation, the second code screening module 407 is further configured to:

[0204] In a case where the iteration result does not satisfy the second preset condition, at least one target candidate reference code is generated, and the step of constructing positive and negative example candidate codes according to the target candidate reference code by using the large language model is returned to.

[0205] In an optional implementation, the second code screening module 407 is further configured to:

[0206] In a case where the number of generated target candidate reference codes reaches a second preset number threshold, it is determined that the iteration result satisfies the second preset condition.

[0207] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. According to actual needs, part or all of the modules can be selected to achieve the purpose of the present application scheme. Those skilled in the art can understand and implement without creative labor.

[0208] Based on the same technical concept, the present disclosure also provides an electronic device. Refer to Figure 10As shown, a structural schematic diagram of the electronic device 500 provided by the embodiments of the present disclosure is shown, including a processor 501, a memory 502, and a bus 503. The memory 502 is configured to store execution instructions, including an internal memory 5021 and an external memory 5022. The internal memory 5021 is also referred to as an internal storage, and is configured to temporarily store operation data in the processor 501 and exchange data with the external memory 5022 such as a hard disk. The processor 501 exchanges data with the external memory 5022 through the internal memory 5021.

[0209] In the embodiments of the present application, the memory 502 is specifically configured to store application program codes for executing the schemes of the present application, and is controlled to execute by the processor 501. That is, when the electronic device 500 is running, the processor 501 and the memory 502 communicate through the bus 503, so that the processor 501 executes the application program codes stored in the memory 502, and further executes the methods described in any of the preceding embodiments.

[0210] The memory 502 can be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), and the like.

[0211] The processor 501 can be an integrated circuit chip with a processing capability. The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP), and the like; or can be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The general processor can be a microprocessor or the processor can also be any conventional processor.

[0212] It should be understood that the structures illustrated in the embodiments of the present application do not constitute a specific limitation on the electronic device 500. In other embodiments of the present application, the electronic device 500 may include more or fewer components than shown, or may combine or separate certain components, or arrange the components differently. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0213] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program executes the steps of the method for configuring resources for a multimodal network described in the above-described method embodiment. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0214] The present disclosure also provides a computer program product that carries program code. The program code includes instructions that can be used to execute the steps of the method for configuring resources for a multimodal network in the above-mentioned method embodiment. For details, please refer to the above-mentioned method embodiment and will not be repeated here.

[0215] The computer program product may be implemented in hardware, software, or a combination thereof. In one embodiment, the computer program product is implemented as a computer storage medium. In another embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0216] In addition, embodiments of the subject matter and functional operations described in this specification may be implemented in the following: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or a combination of one or more thereof. Embodiments of the subject matter described in this specification may be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier to be executed by a data processing device or to control the operation of the data processing device. Alternatively or additionally, the program instructions may be encoded on an artificially generated propagation signal, such as a machine-generated electrical, optical, or electromagnetic signal, which is generated to encode information and transmit it to a suitable receiver device for execution by the data processing device. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more thereof.

[0217] The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform the corresponding functions by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus can be implemented as, special purpose logic circuitry, such as an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).

[0218] Computers suitable for executing computer programs include, for example, general-purpose and / or special-purpose microprocessors, or any other type of central processing unit. Typically, the central processing unit will receive instructions and data from a read-only memory and / or random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include one or more mass storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, or the computer will be operably coupled to such mass storage devices to receive data from them or to transmit data to them, or both. However, a computer does not necessarily have such devices. In addition, a computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0219] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and storage devices, including, for example, semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD ROM and DVD-ROM disks. The processor and memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0220] Although this specification includes many specific implementation details, these should not be interpreted as limiting the scope of any invention or the scope of protection claimed, but are mainly used to describe the features of specific embodiments of specific inventions. Certain features described in multiple embodiments within this specification may also be implemented in combination in a single embodiment. On the other hand, the various features described in a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination. In addition, although features may work in certain combinations as described above and even initially claimed as such, one or more features from the claimed combination may be removed from the combination in some cases, and the claimed combination may point to a sub-combination or a variation of the sub-combination.

[0221] Similarly, although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that these operations be performed in the particular order shown or performed sequentially, or that all illustrated operations be performed to achieve the desired results. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

[0222] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the particular order shown or sequential sequence to achieve the desired results. In some implementations, multitasking and parallel processing may be advantageous.

[0223] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for configuring resources in a multimodal network, characterized in that: A network control agent configured to configure resources in the multimodal network, the method comprising: Acquiring network status information of each network modality in the multimodal network, the network status information including network request information and network environment indicator information; Based on the network status information, running a preset configuration policy code to generate a configuration policy for the multimodal network, and optimizing resource configuration of the multimodal network according to the configuration policy; The configuration strategy code is generated by nested iterative optimization using a large language model based on a pre-built database including multiple data samples; the network status information and the network status input data of the code predictor used to screen the code during the configuration strategy code generation process come from the same network environment.

2. The method according to claim 1, characterized in that The configuration policy code is generated by the following steps: Acquire at least one data sample from the database, each data sample including a problem description prompt word and a corresponding reference code example; Processing the problem description prompt words in the at least one data sample using a large language model to obtain candidate problem description prompt words, and generating a first number of initial candidate codes based on the candidate problem description prompt words and the reference code examples; Using a code predictor to screen the initial candidate codes, and obtaining a target candidate data sample when the result of this round of iteration meets a first preset condition; The target candidate data sample includes a target candidate question description prompt word and a corresponding target candidate reference code; generating a second number of intermediate candidate codes based on the target candidate data sample using the large language model; Using the code predictor to screen the intermediate candidate codes, and obtaining the optimal code for this round if the result of this round of iteration meets the second preset condition; In response to the iterative round not reaching the preset round threshold, the optimal code of this round and the target candidate question description prompt word form a new data sample and add it to the database, repeating the above steps until the iterative round reaches the preset round threshold to obtain the optimal configuration strategy code.

3. The method according to claim 2, characterized in that The second number is greater than the first number.

4. The method according to claim 2, characterized in that The step of processing the question description prompt words in the at least one data sample using the large language model to obtain candidate question description prompt words includes: The large language model is used to reflect on the problem description ideas according to positive and negative examples, and the problem description prompt words are processed to obtain the candidate problem description prompt words.

5. The method according to claim 2, characterized in that The method further comprises: If the iteration result does not meet the first preset condition, a new question description prompt word is generated, and the process returns to the step of using the large language model to process the question description prompt words in the at least one data sample to obtain candidate question description prompt words.

6. The method according to claim 5, characterized in that When the number of generated question description prompt words reaches a first preset number threshold, it is determined that the iteration result meets the first preset condition.

7. The method according to claim 2, characterized in that The generating a second number of intermediate candidate codes according to the target candidate data sample using the large language model includes: The large language model is used to construct positive and negative example candidate codes according to the target candidate reference codes, and a second number of intermediate candidate codes are generated according to the target candidate question description prompt words and the positive and negative example candidate codes.

8. The method according to claim 7, characterized in that The method further comprises: If the iteration result does not meet the second preset condition, at least one target candidate reference code is generated, and the process returns to the step of constructing positive and negative example candidate codes based on the target candidate reference code using the large language model.

9. The method according to claim 8, characterized in that The method further comprises: When the number of generated target candidate reference codes reaches a second preset number threshold, it is determined that the iteration result meets the second preset condition.

10. A device for configuring resources in a multimodal network, characterized in that: include: An information acquisition module, configured to acquire network status information of each network mode in the multimodal network, wherein the network status information includes network request information and network environment indicator information; A resource configuration module, configured to run a preset configuration policy code to generate a configuration policy for the multimodal network based on the network status information, and optimize the resource configuration of the multimodal network according to the configuration policy; The configuration strategy code is generated by nested iterative optimization using a large language model based on a pre-built database including multiple data samples; the network status information and the network status input data of the code predictor used to screen the code during the configuration strategy code generation process come from the same network environment.

11. An electronic device, characterized in that: include: A processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the method for configuring resources for a multimodal network according to any one of claims 1 to 9 is performed.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for configuring resources for a multimodal network according to any one of claims 1 to 9 is executed.

Images