Artificial intelligence large model cooperation method and device, terminal and medium
By acquiring task requirements, determining task complexity, and dynamically allocating tasks to heterogeneous models, and using a consensus algorithm to handle task feedback, the problems of inflexible task allocation and response delay in existing technologies are solved, and efficient task allocation for heterogeneous model collaboration is achieved.
Patent Information
- Application Number
- CN202511515798.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-22
- Publication Date
- 2026-02-17
AI Technical Summary
Existing large-scale model collaboration methods cannot select the optimal task allocation based on the heterogeneous models, resulting in poor flexibility in task allocation and a tendency for response delays in heterogeneous models.
By obtaining task requirements, the complexity of chapter tasks is determined, and tasks are dynamically allocated to heterogeneous models based on task complexity, edge computing, and the processing complexity of heterogeneous models. Consistency algorithms are used to handle task feedback from heterogeneous models, thereby optimizing the task allocation process.
It improves the flexibility of task allocation, reduces the response latency of heterogeneous models, and enables dynamic optimization selection for heterogeneous model collaboration.
Smart Images
Figure CN121541987A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, specifically to a method, apparatus, terminal, and medium for collaborative large-scale artificial intelligence models. Background Technology
[0002] With the rapid development of artificial intelligence technology, large-model collaboration methods are increasingly widely used in various fields. Traditional large-model collaboration methods mainly rely on a single model for processing and use fixed rules for task allocation during collaboration. This makes it difficult to handle the dynamic optimization selection of multiple complex heterogeneous models and the response latency during model collaboration. In recent years, multi-model collaboration methods have gradually gained attention, improving the overall performance and adaptability of the system by integrating the advantages of multiple specialized models.
[0003] However, existing large-scale model collaboration methods cannot select the optimal task allocation based on the heterogeneous model situation, resulting in poor flexibility in task allocation and a tendency for heterogeneous models to experience response delays. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention proposes a collaborative method for large artificial intelligence models, aiming to achieve dynamic optimization when assigning tasks to heterogeneous models and improve the flexibility of task allocation.
[0005] In a first aspect, embodiments of this application provide a collaborative method for large artificial intelligence models, based on a large model and several heterogeneous models, including:
[0006] Obtain the task requirements, which consist of several chapter tasks;
[0007] The task complexity of each chapter task is determined, and the chapter tasks are dynamically allocated to several heterogeneous models based on the task complexity, edge computing algorithm, and processing complexity of the heterogeneous model.
[0008] The collaborative results are obtained by processing the task feedback returned by several heterogeneous models using a consensus algorithm.
[0009] Optionally, determining the task complexity of each chapter task and dynamically allocating the chapter tasks to several heterogeneous models based on the task complexity, edge computing algorithms, and the processing complexity of the heterogeneous models includes:
[0010] The chapter language information of each chapter in the task requirements is extracted by a large language model. The chapter language information includes the language keywords corresponding to each chapter, the semantic relevance between language keywords, and the chapter relevance between each chapter.
[0011] The task complexity of each chapter task is determined based on the language keywords, the semantic relevance, and the chapter relevance.
[0012] Obtain heterogeneous model information, including the processing complexity and processing status of the heterogeneous model;
[0013] Based on the task complexity, the processing complexity, and the processing status, determine the matching data for processing each chapter's tasks;
[0014] The chapter tasks are dynamically assigned to several heterogeneous models using edge algorithms and the matching data.
[0015] Optionally, the step of obtaining heterogeneous model information, wherein the heterogeneous model information includes the processing complexity and processing status of the heterogeneous model, includes:
[0016] Obtain the theoretical computational power, algorithm complexity, number of current tasks, model computation latency, and model energy consumption of the heterogeneous model;
[0017] Based on the theoretical computing power, the algorithm complexity, and the current number of tasks, the processing complexity of the heterogeneous model is obtained.
[0018] The processing status of the heterogeneous model is obtained based on the current number of tasks, the model calculation delay time, and the model energy consumption.
[0019] Optionally, the step of dynamically allocating chapter tasks to several heterogeneous models using edge algorithms and the matching data includes:
[0020] Initialize the model load;
[0021] Optimized matching data is obtained by optimizing the matching data using the edge distance factor in the edge algorithm;
[0022] Assign the chapter corresponding to the largest optimized matching data to the corresponding model;
[0023] Update the model load and update the optimization matching data, and assign the chapter corresponding to the current largest optimization matching data to the corresponding model;
[0024] Repeat the above steps until all chapters in the task requirements have been assigned.
[0025] Optionally, determining the task complexity of each chapter task and dynamically allocating the chapter tasks to several heterogeneous models based on the task complexity, edge computing algorithm, and the processing complexity of the heterogeneous models includes:
[0026] Send verification commands to each heterogeneous model;
[0027] Receive the verification results returned by each heterogeneous model according to the verification command;
[0028] Analyze the verification results to determine information discrepancies;
[0029] Adjust the training set and training parameters of the heterogeneous model corresponding to the information differences based on the information differences.
[0030] Optionally, the step of processing the task feedback returned by several heterogeneous models using a consensus algorithm to obtain the collaborative result includes:
[0031] Text embedding is performed on the task feedback returned by all the heterogeneous models to obtain embedded task feedback;
[0032] The embedded text in the feedback of the embedding task is subjected to vector clustering using a clustering algorithm, resulting in multiple cluster centers;
[0033] The task feedback is optimized based on multiple cluster centers to obtain collaborative results.
[0034] Optionally, optimizing the task feedback based on multiple cluster centers to obtain the collaborative result includes:
[0035] The data in the task feedback that is far from the center of the cluster is deleted to obtain the collaboration result.
[0036] Secondly, embodiments of this application provide an artificial intelligence large model collaboration device, comprising:
[0037] The requirement acquisition module is used to acquire task requirements, which consist of several chapter tasks.
[0038] The task allocation module is used to determine the task complexity of each chapter task, and dynamically allocate the chapter tasks to several heterogeneous models according to the task complexity, edge computing algorithm and the processing complexity of the heterogeneous model.
[0039] The result determination module is used to process the task feedback returned by several heterogeneous models through a consensus algorithm to obtain the collaborative result.
[0040] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the artificial intelligence large model collaboration method as described in any one of the first aspects above.
[0041] Fifthly, embodiments of this application provide a computer program product that, when run on a terminal device, causes the terminal device to execute the artificial intelligence large model collaboration method described in any one of the first aspects.
[0042] In this embodiment, task requirements are obtained, consisting of several chapter tasks; the task complexity of each chapter task is determined, and the chapter tasks are dynamically allocated to several heterogeneous models based on the task complexity, edge computing algorithm, and the processing complexity of the heterogeneous models; the task feedback returned by the several heterogeneous models is processed through a consensus algorithm to obtain a collaborative result. This improves the dynamic optimization selection when allocating tasks to heterogeneous models and enhances the flexibility of task allocation. Attached Figure Description
[0043] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.
[0044] Figure 1 This is a flowchart illustrating an embodiment of the artificial intelligence large model collaboration method provided in this application;
[0045] Figure 2 This is a flowchart illustrating the second embodiment of the AI large model collaboration method provided in this application;
[0046] Figure 3 This is a flowchart illustrating the third embodiment of the AI large model collaboration method provided in this application;
[0047] Figure 4 This is a schematic diagram of the structure of the artificial intelligence large model collaboration device provided in the embodiments of this application;
[0048] Figure 5 This is a schematic diagram of the structure of the terminal device provided in the embodiments of this application. Detailed Implementation
[0049] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0050] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.
[0051] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0052] The execution entity of the AI large model collaboration method provided in this application embodiment can be an AI large model collaboration device, which obtains task requirements, and the task requirements consist of several chapter tasks.
[0053] The task complexity of each chapter task is determined, and the chapter tasks are dynamically allocated to several heterogeneous models based on the task complexity, edge computing algorithm, and processing complexity of the heterogeneous model.
[0054] The collaborative results are obtained by processing the task feedback returned by several heterogeneous models using a consensus algorithm.
[0055] Figure 1 The illustration shown is a schematic flowchart of a large-scale AI model collaboration method provided in an embodiment of this application. This is illustrative and not limiting; the method can be applied to the aforementioned large-scale AI model collaboration device, which can be used for large-scale AI models. Figure 1 As shown, the method may include:
[0056] S10, Obtain task requirements, which consist of several chapter tasks;
[0057] To achieve dynamic optimization when assigning tasks to heterogeneous models and improve the flexibility of task allocation, large AI models collaborate to obtain task requirements, which consist of several chapter tasks.
[0058] S20, determine the task complexity of each chapter task, and dynamically allocate the chapter tasks to several heterogeneous models according to the task complexity, edge computing algorithm and the processing complexity of the heterogeneous model;
[0059] After acquiring the task requirements, the AI large-scale model collaboration device determines the task complexity of each chapter task and dynamically allocates the chapter tasks to several heterogeneous models based on the task complexity, edge computing algorithms, and the processing complexity of the heterogeneous models. The heterogeneous models are pre-trained neural network models and may include at least two of the following architectures: CNN, Transformer, and RNN.
[0060] As one implementation method, determining the task complexity of each chapter task and dynamically allocating the chapter tasks to several heterogeneous models based on the task complexity, edge computing algorithm, and the processing complexity of the heterogeneous models may include: sending verification instructions to each heterogeneous model; receiving verification results returned by each heterogeneous model according to the verification instructions; analyzing the verification results to determine information differences; and adjusting the training set and training parameters of the heterogeneous models corresponding to the information differences based on the information differences.
[0061] As one implementation method, adjusting the training set and training parameters of the heterogeneous model corresponding to the information differences based on the information differences may include: locating the difference layer by comparing the KL divergence of the verification results, and adjusting the training parameters and training set by using adversarial examples to incrementally adjust the training parameters and training set.
[0062] By sending verification commands to heterogeneous models, receiving verification results, and analyzing the received verification results, it can be determined whether there are deviations or differences in the processing capabilities of each heterogeneous model. If some of the heterogeneous models have differences, the differences or deviations can be adjusted by adjusting the training set and training parameters of the heterogeneous models with differences or deviations.
[0063] Furthermore, referring to Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the artificial intelligence large-scale model collaboration method of the present invention. Based on the above... Figure 2 The illustrated embodiment determines the task complexity of each chapter task and dynamically allocates the chapter tasks to several heterogeneous models based on the task complexity, edge computing algorithm, and the processing complexity of the heterogeneous models. Specifically, this includes:
[0064] S21, extract chapter language information of each chapter in the task requirements through a large language model. The chapter language information includes language keywords corresponding to each chapter, semantic relevance between language keywords, and chapter relevance between each chapter.
[0065] After acquiring the task requirements, the AI large model collaborative device extracts the chapter language information of each chapter in the task requirements through a large language model. The chapter language information includes the language keywords corresponding to each chapter, the semantic relevance between the language keywords, and the chapter relevance between each chapter.
[0066] The large language model is used to extract the language keywords corresponding to each chapter in the task requirements, the semantic relevance between language keywords, and the chapter relevance between each chapter.
[0067] Among them, for each chapter Extract its linguistic keywords to obtain the first Collection of language keywords for each chapter And calculate the weight of each language keyword. .
[0068] This involves calculating the weight of each language keyword. Calculated using the term frequency-inverse document frequency (TF-IDF) method;
[0069] ;
[0070] in, For language keywords in chapters The frequency in; The inverse document frequency of language keywords throughout the book;
[0071] Then, the word vectors of language keywords are processed using cosine similarity to calculate the language keywords. and semantic relevance between ;
[0072] ;
[0073] in, Keywords Word vectors; Keywords Word vectors;
[0074] The Jaccard similarity of the language keyword set can be used to calculate the first... Chapter and Number Similarity between chapters ;
[0075] ;
[0076] in, For the first A collection of language keywords for each chapter; For the first A collection of language keywords for each chapter;
[0077] S22, determine the task complexity of each chapter task based on the language keywords, the semantic relevance, and the chapter relevance;
[0078] After obtaining the language keywords corresponding to each chapter, the semantic relevance between the language keywords, and the chapter relevance between each chapter, the AI large model collaborative device determines the task complexity of each chapter task based on the language keywords, the semantic relevance, and the chapter relevance.
[0079] Determining the task complexity of each chapter task based on the language keywords, semantic relevance, and chapter relevance may include:
[0080] ;
[0081] in, The first weighting coefficient; For the first Chapter task complexity; The weight of each language keyword; For the first The chapter's language keywords; For the first A collection of language keywords for each chapter; This is the second weighting coefficient; For the first Language keywords in the chapter's language keyword set; For the first Language keywords in the chapter's language keyword set; Language keywords and language keywords The semantic relevance between them; For the first Number of language keywords in the chapter; This is the third weighting coefficient; For the first Chapter and Number Similarity between chapters; The total number of chapters required for the task; among which, It can be 0.4; It can be 0.4; It can be 0.2. Wherein, It can also be 0.5; It can also be 0.3; It can also be 0.2. Wherein, It can also be 0.4; It can also be 0.3; It can also be 0.3.
[0082] If the first Chapter Dependency Chapters will be added. Weighting coefficients The value of the second weighting factor is reduced when the task requires a narrative chapter. The value is set to 0.3; when the task requirement is a technical chapter, the second weighting coefficient is increased. The value is reduced to 0.5.
[0083] S23, Obtain heterogeneous model information of the heterogeneous model, the heterogeneous model information including the processing complexity and processing status of the heterogeneous model;
[0084] After obtaining the task complexity of each chapter's task, the AI large model collaboration device acquires heterogeneous model information, which includes the processing complexity and processing status of the heterogeneous model.
[0085] As one implementation method, obtaining heterogeneous model information, including the processing complexity and processing status of the heterogeneous model, may include: obtaining the theoretical computing power, algorithm complexity, current number of tasks, model computation latency, and model energy consumption of the heterogeneous model; obtaining the processing complexity of the heterogeneous model based on the theoretical computing power, algorithm complexity, and current number of tasks; and obtaining the processing status of the heterogeneous model based on the current number of tasks, model computation latency, and model energy consumption. Wherein, model computation latency refers to the computation latency of the heterogeneous model; and model energy consumption refers to the energy consumption of the heterogeneous model.
[0086] Heterogeneous model Processing complexity The calculation formula is:
[0087] ;
[0088] in, Heterogeneous model The algorithm complexity is denoted by , where higher algorithm complexity allows for the processing of more complex tasks. Heterogeneous model The higher the theoretical computing power, the stronger the processing power. Heterogeneous model The number of current tasks; the larger the number of current tasks, the slower the processing speed. The first minimum value can be... ;
[0089] Heterogeneous model Processing status The calculation formula is:
[0090] ;
[0091] in, This is the fourth weighting coefficient; Heterogeneous model The model calculates the delay time; This is the fifth weighting coefficient; This is the sixth weighting coefficient; Heterogeneous model The model's energy consumption. The model's energy consumption... The power consumption of the GPU is collected in real time by a power monitoring chip and calculated in combination with the task scheduling frequency.
[0092] S24, Based on the task complexity, the processing complexity, and the processing status, determine the matching data for processing each chapter's task;
[0093] After obtaining the processing complexity and processing status of the heterogeneous models, the AI large-scale model collaboration device determines the matching data for processing each chapter's tasks based on the task complexity, processing complexity, and processing status. The matching data refers to the matching processing capability between the device and the heterogeneous models when processing each chapter's tasks.
[0094] Based on the task complexity, the processing complexity, and the processing status, the matching data for processing each chapter's task can be determined, which may include:
[0095] ;
[0096] in, For chapters With model Matching data between them; The second minimum value can be... Among them, the larger the ratio between task complexity and model processing capability, the better the model... With Chapter The better the matching degree between them; the better the model's processing state. The lower the value, the stronger the processing capability.
[0097] S25, dynamically allocate chapter tasks to several heterogeneous models using edge algorithms and the matching data;
[0098] After obtaining the matching data for processing each chapter's tasks, the AI large-scale model collaboration device dynamically allocates the chapter tasks to several heterogeneous models using edge computing algorithms and the matching data.
[0099] The step of dynamically allocating chapter tasks to several heterogeneous models using edge algorithms and matching data may include: initializing the model load; optimizing the matching data using the edge distance factor in the edge algorithm to obtain optimized matching data; then allocating the chapter corresponding to the largest optimized matching data to the corresponding model; updating the model load and updating the optimized matching data, and then allocating the chapter corresponding to the current largest optimized matching data to the corresponding model; iterating the above steps until all chapters in the task requirements have been allocated.
[0100] Edge matching algorithms can be used to optimize matching data, thereby reducing the response latency of heterogeneous models.
[0101] To reduce latency in the edge matching algorithm, optimization is performed using edge distance factor matching data to obtain optimized matching data. The calculation formula is:
[0102] ;
[0103] in, The edge distance factor is the transmission delay coefficient between edge nodes and center nodes in the network topology. It is obtained through real-time testing and has a value range of 0.1 to 0.5.
[0104] As another implementation, dynamically allocating chapter tasks to several heterogeneous models using edge algorithms and the matching data can include: preprocessing the chapter tasks by using edge algorithms, language keywords, and semantic relevance to perform data cleaning, filtering, and noise reduction, resulting in preprocessed task information; and then dynamically allocating the preprocessed task information to several heterogeneous models using edge algorithms and the matching data. By performing data cleaning, filtering, and noise reduction preprocessing on the chapter tasks, the accuracy of heterogeneous models in processing the chapter tasks can be improved, and more accurate task feedback can be obtained.
[0105] S30, the task feedback returned by several heterogeneous models is processed by the output consistency algorithm to obtain the collaboration result.
[0106] After dynamically assigning chapter tasks to heterogeneous models, the AI large model collaboration device processes the task feedback returned by several heterogeneous models through an output consistency algorithm to obtain the collaboration result.
[0107] Furthermore, referring to Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the artificial intelligence large-scale model collaboration method of the present invention. Based on the above... Figure 3The illustrated embodiment processes the task feedback returned by several heterogeneous models using a consensus algorithm to obtain a collaborative result, specifically including:
[0108] S31, perform text embedding on the task feedback returned by all the heterogeneous models to obtain embedded task feedback;
[0109] After dynamically assigning chapter tasks to heterogeneous models, the AI large model collaboration device embeds text into the task feedback returned by all the heterogeneous models to obtain embedded task feedback.
[0110] S32, the embedded text in the embedded task feedback is subjected to vector clustering using a clustering algorithm, resulting in multiple cluster centers;
[0111] After receiving the embedded task feedback, the AI large model collaborative device performs vector clustering on the embedded text in the embedded task feedback using a clustering algorithm, resulting in multiple cluster centers.
[0112] S33, optimize the task feedback based on multiple cluster centers to obtain collaborative results.
[0113] After obtaining multiple cluster centers, the AI large-scale model collaboration device optimizes the task feedback based on these cluster centers to obtain the collaboration result.
[0114] As another implementation, optimizing the task feedback based on multiple cluster centers to obtain a collaborative result includes deleting data in the task feedback that is far from the cluster centers to obtain a collaborative result.
[0115] As another implementation, the task feedback is optimized based on multiple cluster centers to obtain a collaborative result, including: calculating a confidence interval based on Mahalanobis distance, and performing weighted fusion on task feedback far from the cluster centers, with the weights inversely proportional to the cluster center density, to obtain the collaborative result.
[0116] After the large model receives the collaborative results returned by the heterogeneous models and undergoes consistency processing, it can directly output the collaborative results as the task requirements.
[0117] After receiving the collaborative results from the heterogeneous models and undergoing consistency processing, the large model can be further processed and analyzed to obtain the final task results required for the task. The processed and analyzed task results are then output.
[0118] In summary, by obtaining task requirements, which consist of several chapter tasks, determining the task complexity of each chapter task, and dynamically allocating the chapter tasks to several heterogeneous models based on the task complexity, edge computing algorithm, and the processing complexity of the heterogeneous models, and processing the task feedback returned by the heterogeneous models through a consensus algorithm to obtain the collaborative result, this approach improves the dynamic optimization selection when allocating tasks to heterogeneous models and enhances the flexibility of task allocation.
[0119] For those consistent with the above, please refer to Figure 4 , Figure 4 This application provides a schematic diagram of the structure of an artificial intelligence large-scale model collaboration device. For example... Figure 4 As shown, the device includes:
[0120] The requirement acquisition module 401 is used to acquire task requirements, which consist of several chapter tasks;
[0121] The task allocation module 402 is used to determine the task complexity of each chapter task, and dynamically allocate the chapter tasks to several heterogeneous models according to the task complexity, edge computing algorithm and the processing complexity of the heterogeneous model.
[0122] The result determination module 403 is used to process the task feedback returned by several heterogeneous models through a consensus algorithm to obtain the collaborative result.
[0123] This application also provides a terminal device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor. When the processor executes the computer program, it implements the steps in the embodiment of the artificial intelligence large model collaboration method.
[0124] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the artificial intelligence large model collaboration methods described in the above method embodiments.
[0125] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the large-scale artificial intelligence model collaboration methods described in the above method embodiments.
[0126] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include at least: any entity or device capable of carrying computer program code to a device / terminal equipment, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable storage media cannot be electrical carrier signals or telecommunication signals.
[0127] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0128] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0129] In the embodiments provided in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.
[0130] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
Claims
1. An artificial intelligence large model cooperation method based on a large model and a plurality of heterogeneous models, characterized in that, The method comprises the following steps: acquiring a task requirement composed of a plurality of chapter tasks; determining the task complexity of each chapter task, and dynamically allocating the chapter tasks to a plurality of heterogeneous models according to the task complexity, an edge algorithm, and the processing complexity of the heterogeneous models; processing the task feedback returned by the plurality of heterogeneous models through a consistency algorithm to obtain a collaboration result.
2. The artificial intelligence large model cooperation method according to claim 1, characterized in that, The determination of the task complexity of each chapter task and the dynamic allocation of the chapter tasks to the plurality of heterogeneous models according to the task complexity, the edge algorithm, and the processing complexity of the heterogeneous models comprise: extracting chapter language information of each chapter in the task requirement through a large language model, wherein the chapter language information comprises language keywords corresponding to each chapter, semantic correlation between the language keywords, and chapter correlation between each chapter; determining the task complexity of each chapter task according to the language keywords, the semantic correlation, and the chapter correlation; acquiring heterogeneous model information of the heterogeneous models, wherein the heterogeneous model information comprises the processing complexity and the processing state of the heterogeneous models; determining matching data for processing each chapter task according to the task complexity, the processing complexity, and the processing state; dynamically allocating the chapter tasks to the plurality of heterogeneous models through the edge algorithm and the matching data.
3. The artificial intelligence large model cooperation method according to claim 2, characterized in that, The acquisition of the heterogeneous model information of the heterogeneous models, wherein the heterogeneous model information comprises the processing complexity and the processing state of the heterogeneous models, comprises: acquiring the theoretical calculation capability, the algorithm complexity, the current task number, the model calculation delay time, and the model energy consumption of the heterogeneous models; determining the processing complexity of the heterogeneous models according to the theoretical calculation capability, the algorithm complexity, and the current task number; determining the processing state of the heterogeneous models according to the current task number, the model calculation delay time, and the model energy consumption.
4. The artificial intelligence large model cooperation method according to claim 3, characterized in that, The dynamic allocation of the chapter tasks to the plurality of heterogeneous models through the edge algorithm and the matching data comprises: initializing model load; optimizing the matching data through an edge distance factor in the edge algorithm to obtain optimized matching data; allocating the chapter corresponding to the largest optimized matching data to the corresponding model; updating the model load and the optimized matching data, and allocating the chapter corresponding to the current largest optimized matching data to the corresponding model; iterating the above steps until all chapters in the task requirement are allocated.
5. The artificial intelligence large model cooperation method according to any one of claims 1 to 4, characterized in that, Before the determination of the task complexity of each chapter task and the dynamic allocation of the chapter tasks to the plurality of heterogeneous models according to the task complexity, the edge algorithm, and the processing complexity of the heterogeneous models, the method comprises the following steps: sending a verification instruction to each heterogeneous model; receiving a verification result returned by each heterogeneous model according to the verification instruction; analyzing the verification result to determine information difference; adjusting the training set and the training parameter of the heterogeneous model corresponding to the information difference according to the information difference.
6. The artificial intelligence large model cooperation method according to claim 5, characterized in that, The processing of the task feedback returned by the plurality of heterogeneous models through the consistency algorithm to obtain the collaboration result comprises: Text embedding is performed on task feedback returned by all the isomorphic models to obtain embedded task feedback; Vector clustering is performed on embedded text in the embedded task feedback by a clustering algorithm to obtain a plurality of clustering cluster centers; Task feedback is optimized according to the plurality of clustering cluster centers to obtain a collaboration result.
7. The artificial intelligence large model cooperation method according to claim 5, characterized in that, The optimization of the task feedback according to the plurality of clustering cluster centers to obtain the collaboration result comprises: Data in the task feedback far from the clustering cluster center is deleted to obtain the collaboration result.
8. An artificial intelligence large model collaboration apparatus, characterized by, It comprises: A demand acquisition module is configured to acquire task demands, wherein the task demands are composed of a plurality of chapter tasks; A task allocation module is configured to determine the task complexity of each chapter task, and dynamically allocate chapter tasks to a plurality of isomorphic models according to the task complexity, an edge algorithm, and the processing complexity of the isomorphic models; A result determination module is configured to process task feedback returned by a plurality of isomorphic models by a consistency algorithm to obtain a collaboration result.
9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the artificial intelligence large model collaboration method of any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the artificial intelligence large model collaboration method of any one of claims 1 to 7.