Model service providing method and device, electronic equipment and storage medium

Abstract

The invention provides a model service providing method and device, electronic equipment and a storage medium. The invention relates to the technical field of artificial intelligence, in particular to the technical fields of large models, big data, intelligent search, data processing and the like, and can be used for application scenes such as generative search, document intelligent editing, intelligent assistants, virtual assistants, intelligent e-commerce and the like. According to the specific implementation scheme, a user demand is determined according to a received search request; determining a target model and a search path for the search request from the model architecture system in combination with user requirements; and controlling the target model to execute the search path to obtain a search result output by the target model. According to the disclosure of the scheme, the search efficiency and the result quality can be improved by intelligently matching the user requirements with the appropriate models and data resources.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to technical fields such as big models, big data, intelligent search and data processing. It can be used in application scenarios such as generative search, intelligent document editing, intelligent assistants, virtual assistants, and intelligent e-commerce, and specifically relates to a model service providing method, device, electronic device and storage medium. Background Art

[0002] With the development of artificial intelligence, some large language models have emerged. Although these large language models have been applied in many fields, how to improve their search efficiency and search quality is still an urgent problem to be solved. Summary of the invention

[0003] The present invention provides a model service providing method, device, electronic device and storage medium.

[0004] According to one aspect of the present disclosure, a model service providing method is provided, which is applied to a preset model architecture system, and the method includes:

[0005] Determining user needs based on received search requests;

[0006] Based on user needs, determine the target model and search path for the search request from the model architecture system;

[0007] Control the target model to execute the search path to obtain the search results output by the target model.

[0008] According to another aspect of the present disclosure, a model service providing device is provided, which is applied to a preset model architecture system, and the device includes:

[0009] A demand determination module, used to determine user demand based on the received search request;

[0010] A path selection module, used to determine a target model and a search path for a search request from a model architecture system in combination with user requirements;

[0011] The execution control module is used to control the target model to execute the search path to obtain the search results output by the target model.

[0012] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0013] at least one processor; and

[0014] a memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute any method in the embodiments of the present disclosure.

[0016] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any method according to the embodiments of the present disclosure.

[0017] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements any method according to the embodiments of the present disclosure when executed by a processor.

[0018] By adopting the solution disclosed in the present invention, search efficiency and result quality can be improved by intelligently matching user needs with appropriate models and data resources.

[0019] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0021] Figure 1 is a flowchart of a method for providing a model service according to an embodiment of the present disclosure;

[0022] Figure 2 is a schematic diagram of a model architecture system according to an embodiment of the present disclosure;

[0023] Figure 3 is a workflow diagram of a model service providing method according to an embodiment of the present disclosure;

[0024] Figure 4 is a workflow diagram of a model service providing method according to another embodiment of the present disclosure;

[0025] Figure 5 is a workflow diagram of a model service providing method according to another embodiment of the present disclosure;

[0026] Figure 6 is a workflow diagram of a model service providing method according to another embodiment of the present disclosure;

[0027] Figure 7 is a workflow diagram of a model service providing method according to another embodiment of the present disclosure;

[0028] Figure 8is a schematic diagram of a processing flow after receiving a user request according to an embodiment of the present disclosure;

[0029] Fig. 9 is a schematic diagram of the composition structure of the first data supply engine or the second data supply engine according to an embodiment of the present disclosure;

[0030] Fig.10 is a structural diagram of a model service providing device according to an embodiment of the present disclosure;

[0031] Fig.11 is a scenario diagram of a method for providing a model service according to an embodiment of the present disclosure;

[0032] Fig.12 It is a structural diagram of an electronic device used to implement the model service providing method of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0033] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0034] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there may be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The term "at least one" in this article means any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C. The terms "first" and "second" in this article refer to multiple similar technical terms and distinguish them. They do not mean to limit the order or to limit them to only two. For example, the first feature and the second feature refer to two types / two features. The first feature can be one or more, and the second feature can also be one or more.

[0035] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0036] Before introducing the technical solutions of the embodiments of the present disclosure, the technical terms that may be used in the present disclosure are further explained:

[0037] General large model: refers to a basic model that serves multiple fields or tasks and is a general model used to process a wide range of data types. The general large model has been fully trained and has high adaptability and generalization capabilities. It is often used as the basis for further training of models in other fields (such as vertical large models).

[0038] Vertical big model: refers to a big model that has been specially trained in a specific field, which is fine-tuned based on the general big model to meet the needs of specific businesses or fields.

[0039] Vertical small model: refers to a small-scale model designed for a specific field, focusing on efficiently processing data tasks in the vertical field.

[0040] Model architecture system: refers to a system architecture that supports the development, update and deployment of different types of models. This system includes general large models, vertical large models and vertical small models, and is combined with a data supply engine to achieve flexible and fast model updates and deployment by decoupling models from data.

[0041] In related technologies, general large models generally have the following problems: high training cost, requiring a lot of computing resources and time; long training cycle, difficult to adapt to rapidly changing environments, affecting real-time performance; difficult to update data, new data requires retraining the model, high resource consumption and inability to respond in real time; lack of data in specific fields, resulting in inaccurate model prediction results. Therefore, how to reduce computing resource consumption through low-cost real-time updates and improve the real-time performance, generalization ability and stability of the model in different scenarios is one of the key issues in this technical field.

[0042] In order to at least partially solve one or more of the above-mentioned problems and other potential problems, the present disclosure proposes a model service providing method, which adopts an innovative model architecture system design, significantly reduces training costs and cycles, realizes independent updating of data and models, improves real-time response and prediction accuracy, expands application scenarios, enhances system stability and convenience, and solves the problems of high cost, long cycle, difficult updating and inaccurate prediction in the prior art.

[0043] The present disclosure provides a model service providing method. Figure 1It is a flow chart of a model service providing method according to an embodiment of the present disclosure, and the model service providing method can be applied to a model service providing device. The model service providing device is located in an electronic device. The electronic device includes but is not limited to a fixed device and / or a mobile device. For example, a fixed device includes but is not limited to a server, and the server can be a cloud server or an ordinary server. For example, a mobile device includes but is not limited to: a mobile phone, a tablet computer, and a vehicle-mounted terminal. In some possible implementations, the model service providing method can also be implemented by a processor calling a computer-readable instruction stored in a memory. Figure 1 As shown, the model service providing method includes:

[0044] S101, determining user needs according to the received search request;

[0045] S102, determining a target model and a search path for the search request from the model architecture system in combination with user needs;

[0046] S103: Control the target model to execute the search path to obtain the search results output by the target model.

[0047] In the disclosed embodiments, a search request is an instruction input by a user into a search engine, application or system to express their specific needs for information, goods, services, etc. that they want to find, understand or obtain. Here, the instruction can be text, voice or other forms of data.

[0048] In the disclosed embodiment, user demand is the real purpose or intention behind the user's search, which is not limited to the literal meaning of the search request, but may also include the deep needs that the user has not explicitly expressed but can be inferred based on the context, historical behavior, etc.

[0049] In the disclosed embodiment, the target model refers to a model selected from the general large model, vertical large model, vertical small model and other models included in the model architecture system for executing the search request.

[0050] In the disclosed embodiment, the search path refers to a series of steps or processes selected and executed in the model architecture system to meet user needs, including the entire process from determining an applicable target model to ultimately generating and returning results.

[0051] In the disclosed embodiment, search results refer to the final results generated by executing the search path and calling the target model, including but not limited to text information, data reports, predictive analysis, chart visualization, recommendations, classification results, recognition output, etc.

[0052] In some embodiments, determining user needs based on received search requests includes: receiving and parsing user search requests, extracting keywords, parameters, and semantic information therein; identifying the user's specific intentions through natural language processing (NLP) technology, and understanding the relevance of needs in context; classifying and labeling needs to clarify the type and target of user requests; and finally generating a structured demand representation as a basis for selecting a suitable target model and planning a search path in subsequent steps.

[0053] In some embodiments, determining a target model and a search path for a search request includes: first, based on the parsed user needs and intentions, the model architecture system selects the target model that best handles the need from the available model library. This process may involve comprehensive consideration of factors such as the model type, performance, and applicable scenarios. Next, the model architecture system plans the optimal search path based on the user needs and the characteristics of the target model. The search path includes determining the data source, setting search parameters, selecting appropriate algorithm steps, and optimizing the execution order to ensure that accurate search results can be obtained efficiently. Finally, by executing the path, the target model generates and outputs results that meet the user's needs.

[0054] In some embodiments, the search results of the target model output are obtained, including: generating preliminary results based on the model, the system may filter, sort, denoise or optimize these results to ensure that the output results meet the needs and expectations of the user, and the processed search results will be formatted into a form suitable for display or further use and returned to the user. These results can be in various forms such as text, data reports, charts, suggestions, etc.

[0055] For example, suppose the user enters a search request of "recommend a science book suitable for children". After receiving the request, the system parses the keywords "recommend", "suitable for children", and "science book", and identifies the user's intention, namely "content recommendation" through natural language processing technology. Combined with the context, the system classifies the demand as "personalized recommendation" and marks it as "children" and "science book". Then, the system selects a suitable target model from the model architecture, such as a large vertical model "personalized recommendation model" or a small vertical model "children's book recommendation model". Next, the system plans the search path, first determining the data source, such as an existing children's book database or the bibliographic data of an online bookstore. Then, the system sets the search parameters, including the appropriate age range and scientific topics, and selects the appropriate recommendation algorithm, such as collaborative filtering or content-based recommendation. The system executes according to the planned path, and the target model selects and recommends several science books suitable for children from the data source according to the set parameters and the optimized process. Finally, the search results generated by the system may include a list of recommended books, a brief introduction to each book, the appropriate age group, and purchase links, and return these results to the user. For example, recommended books may include "A Brief History of the Human Body (Colored Edition for Children)", which is suitable for children aged 6-9 and introduces scientific knowledge about the human body; "Children's Encyclopedia of Space Exploration", which is suitable for children aged 5-8 and introduces the basic knowledge and exploration process of space.

[0056] The technical solution of the disclosed embodiment can accurately identify user needs, and call different target models based on the model architecture system, optimize the search path, achieve efficient and accurate personalized result generation, improve system response speed and result relevance, and significantly enhance user experience.

[0057] In some embodiments, Figure 2 As shown, the model architecture system may include: a first type of general large model, a second type of general large model, M vertical large models, N vertical small models, a first data supply engine and a second data supply engine. The first type of general large model and the second type of general large model are respectively connected to the first data supply engine; the vertical large model and the vertical small model are respectively connected to the second data supply engine. The order of magnitude of the first type of general large model is greater than that of the second type of general large model. M is an integer not less than 1, and N is an integer not less than 1.

[0058] Here, the first type of general large model refers to a model used to provide general data services. It has a large scale and data processing capability, usually a trillion-parameter scale, can handle data and tasks in a wide range of fields, and has a high generalization ability.

[0059] Here, the second type of general large model refers to the basic model split from the first type of general large model. This model is not trained in combination with any vertical business data. It is usually a basic model with a parameter scale of tens of billions. It is usually not used alone, but serves as the basis for further training or tuning.

[0060] Here, the vertical big model refers to a model that is trained based on the second type of general big model combined with data from a specific vertical field, and is usually used to handle specialized tasks in that vertical field. This model is usually a model with tens of millions of parameters, which is more adaptable, can more accurately reflect the needs of a specific field, and provide more accurate services and predictions.

[0061] Here, the small vertical model refers to a small model for the target vertical business in a specific vertical field, which is trained based on the second type of general large model, and is usually a model with a parameter scale of thousands.

[0062] Here, the first data supply engine refers to an engine used to provide data support for the first type of general large model and the second type of general large model. The engine is responsible for managing and supplying large-scale, highly versatile data sets to ensure that these models can handle tasks in multiple fields.

[0063] Here, the second data supply engine refers to the engine used to provide data support for vertical large models and vertical small models. This engine mainly manages and supplies specialized data sets related to specific vertical fields to ensure that these models can obtain highly targeted and high-quality data during training and reasoning.

[0064] It should be noted that the values ​​of M and N can be adjusted according to the requirements of specific application scenarios to adapt to tasks of different scales and complexities.

[0065] In this way, through this architectural design, the model system can flexibly respond to different types of tasks. It has the ability to process data in a wide range of fields, and can provide accurate and efficient services in specific vertical fields.

[0066] In some embodiments, the model architecture system may include:

[0067] The first type of general large model is a large model used to provide general data services. The order of magnitude of the first type of general large model is a first value, and the update period is a first duration.

[0068] The second type of general large model is a large model used to provide general data services. The order of magnitude of the second type of general large model is a second value, the update period is a second duration, the second value is smaller than the first value, and the second duration is smaller than the first duration.

[0069] The vertical big model is a big model for providing vertical business data services; the order of magnitude of the vertical big model is the third value, the update period is the third duration, and the third value is less than the second value;

[0070] The vertical small model is a small model used to provide vertical business data services; the order of magnitude of the vertical small model is the fourth value, the update period is the fourth duration, the fourth value is smaller than the third value, and the fourth duration is smaller than the third duration.

[0071] In the embodiments of the present disclosure, vertical business data services refer to specialized data processing and services for specific vertical industries or fields. Compared with general data services, vertical business data services focus more on specific fields, such as medical, financial, education, retail and other industries, and provide more in-depth and customized solutions. For example, in the medical field, vertical business data services may include diagnosis and treatment recommendations for specific diseases such as heart disease and diabetes, while in the financial field, they may involve specialized services such as stock analysis and risk assessment.

[0072] In the disclosed embodiments, order of magnitude refers to the parameter scale or complexity of the model. It is usually used to indicate the number of parameters contained in the model (such as thousands, millions, hundreds of millions, trillions, etc.). The larger the order of magnitude, the stronger the computing power and data processing capability of the model is usually, but it also means that more computing resources and time are required to train and run.

[0073] In the disclosed embodiments, the update cycle refers to the frequency or interval of model updates, which indicates the time from the start of model training or deployment to the next update.

[0074] In some embodiments, the first type of general large model has parameters in the trillion level, and the update cycle is on the monthly level; the second type of general large model has parameters in the tens of billions level, and the update cycle is on the daily level; the vertical large model has parameters in the tens of millions level, and the update cycle is on the daily level; the vertical small model has parameters in the thousands level, and the update cycle is on the hourly level.

[0075] For example, assuming that the first type of general large model of the model architecture system includes Wenxin Yiyan 4.0 (Ernie4.0), the order of parameters is at the trillion level, and the update cycle is at the monthly level; the second type of general large model split from the first type of general large model has a parameter order of tens of billions, and an update cycle is at the daily level; the medical vertical large model and the financial vertical large model trained on the basis of the second type of general large model have parameters at the tens of millions level, and an update cycle is at the daily level; the medical vertical small model and the financial vertical small model trained on the basis of the medical vertical large model and the financial vertical large model have parameters at the thousands level, and an update cycle is at the hourly level.

[0076] In this way, through the hierarchical model architecture, efficient data processing capabilities from general to vertical fields are provided. The first type of general large model ensures strong computing power and stability in a wide range of fields, the second type of general large model improves flexible response to data changes, and the vertical large model and small model focus on precise processing and rapid updating in specific fields, greatly improving the overall efficiency and adaptability of the system.

[0077] In some embodiments, the model architecture system may further include:

[0078] The first version of the vertical big model for the target vertical business is a big model for the target vertical business obtained by training based on the second type of general big model;

[0079] The P-th version of the vertical big model for the target vertical business is a big model trained based on the P-1-th version of the vertical big model for the target vertical business; wherein P is an integer not less than 2.

[0080] In some embodiments, the generation of the first version of the vertical category big model includes: the system first uses the second type of general big model as the basis, and uses the big model supervised fine-tuning (SFT) technology to train and tune the model in combination with the dedicated data set of the target vertical category business. Through SFT, the model maintains the powerful generalization ability of the general model while effectively optimizing for specific fields, thereby generating the first version of the vertical category big model.

[0081] In some embodiments, the generation of the Pth version of the vertical category big model includes: in the iteration of subsequent versions, the Pth version of the vertical category big model is further trained by the SFT technology based on the previous version of the vertical category big model (i.e., the P-1th version). Each iteration combines new business data and requirements, so that the performance of the model in a specific field is continuously improved.

[0082] In some embodiments, the vertical model of version P is obtained by training based on the second type of general large model of version Q, where Q is an integer not less than 2. The second type of general large model of version Q is selected as the basic model. Since the model has been trained with large-scale and diverse data sets and has good generalization ability, on the basis of this model, by training and fine-tuning data for specific fields, the vertical model of version P can be better adapted to industry needs, thereby improving the efficiency and performance of the vertical model in practical applications. For example, a lightweight large language model can be used for rapid joint debugging, and the training is usually completed within one day. The amount of data comes from the actual business data in the industry, and the first version of the medical vertical model is generated. The model can initially provide accurate diagnostic suggestions, and the training cycle is short, so it can be put into use quickly. With the continuous accumulation of new cases and research, the Pth version of the medical vertical model is further trained and optimized through SFT technology based on the previous version of the model, so that the accuracy and response speed of diagnosis continue to improve, ensuring that the system can adapt to the latest medical development trends in real time.

[0083] In this way, through SFT technology, it is possible to quickly generate large vertical models for specific vertical businesses from the second type of general models, significantly reducing training time and resource consumption. The initial version of the model can be completed and put into use in a short time to meet real-time needs. With the introduction of new data, the model continuously improves its accuracy and response speed through iterative optimization, ensuring that the system always maintains efficient and accurate service capabilities in a rapidly changing business environment.

[0084] In some embodiments, the model architecture system may further include:

[0085] The first version of the small vertical model for the target vertical business is a small model for the target vertical business obtained by training the second type of general large model;

[0086] The Kth version of the vertical small model for the target vertical business is a small model trained based on the K-1th version of the vertical small model for the target vertical business; wherein K is an integer not less than 2.

[0087] In some embodiments, the generation of the first version of the vertical category small model includes: the system first uses the second type of general large model as the basic model, combines it with the dedicated data set of the target vertical category business for training, and generates a small model for the vertical category business. These small models are usually built based on common vectorization methods, and their vector dimensions are generally less than 10,000, and the training cycle is completed in about hours. Here, the vectorization method refers to the process of converting data, usually unstructured data such as text, images, or audio, into vector form, so that it can be mathematically calculated and analyzed in a high-dimensional space.

[0088] In some embodiments, the generation of the Kth version of the vertical category model includes: in the subsequent iteration process, the Kth version of the vertical category model is further trained based on the previous version of the vertical category model (i.e., the K-1th version). Through this iterative training method, each time the model is updated, the latest data and requirements can be combined to continuously optimize the model so that it can better adapt to the new environment and data input.

[0089] In some embodiments, the Kth version of the vertical category small model is obtained by training based on the Sth version of the vertical category large model, where S is an integer not less than 2. The Sth version of the vertical category large model is selected as the basic model. On the basis that the Sth version of the vertical category large model has a good ability to understand the vertical category business field, the Sth version of the vertical category large model is used as the teacher model to compress its knowledge into the Kth version of the vertical category small model, so that the small model not only has the vertical category business field understanding ability of the large model, but also has a faster reasoning speed, providing users with fast and personalized recommendation services.

[0090] For example, assuming that the target vertical business is "intelligent financial risk control", in order to build a model suitable for this field, we first start from the second type of general large model and gradually iterate the vertical small model for intelligent financial risk control. Based on the second type of general large model, fine-tuning is performed using data from the field of intelligent financial risk control so that the model can learn the specific features and patterns of this field. By adjusting the model architecture (such as reducing the number of layers, adjusting parameters, etc.) and training strategies (such as learning rate, batch size, etc.), the performance of the model on the intelligent financial risk control task is optimized, and the first version of the vertical small model for intelligent financial risk control is obtained. The model shows good performance on the intelligent financial risk control task, but there is still room for improvement. Use the K-1 version of the vertical small model as a pre-trained model, and use the newly collected data for incremental training or continue fine-tuning. During the training process, some new optimization algorithms or regularization techniques can be introduced to further improve the generalization ability and stability of the model. At the same time, the model architecture can also be fine-tuned to adapt to new data distribution and task requirements. Finally, we obtained the Kth version of the vertical small model for intelligent financial risk control. Compared with the K-1th version of the vertical small model, this model has further improved in performance and can better cope with the challenges in the field of intelligent financial risk control.

[0091] In this way, by utilizing the second type of general large model and vectorization method, small vertical models can be quickly generated and iteratively optimized. These small models are trained within hours to adapt to specific vertical business needs. With each iterative update, the model is continuously optimized in combination with the latest data, so that it remains efficient and accurate when processing unstructured data, thereby ensuring that the system can provide continuous and accurate services in a constantly changing environment. This approach improves the adaptability and responsiveness of the model and reduces the training time and consumption of computing resources.

[0092] In some embodiments, Figure 3 As shown, in combination with user needs, the target model and search path are determined for the search request from the model architecture system, which may include:

[0093] In the case where there is a small vertical model that can meet the user's needs in the model architecture system, the small vertical model that meets the user's needs is determined as the target model; and

[0094] The search path determined for the search request may include: using the vertical category model to access the second data supply engine, and generating search results based on the data returned by the second data supply engine.

[0095] In some embodiments, when the Artificial Intelligence Application Program Interface (AIAPI) server receives a user request, the system performs demand understanding through a demand understanding module and performs execution planning through an execution planning module. In this process, the system generates an execution plan, decides whether to call a general large model or a vertical model, and determines how the final answer is organized. If the system detects that there is a small vertical model in the model architecture that can meet the user's needs, it is determined as the target model, skips the call of the general large model, and directly requests the small vertical model. Subsequently, the small vertical model accesses the second data supply engine, specifically retrieves data in related fields, and uses the returned data to perform operations similar to the Approximate Nearest Neighbor (ANN) search to find relevant high-dimensional data and perform fine sorting. This ensures the high relevance and accuracy of the acquired data and generates search results that meet user needs. After processing the data, the small vertical model further optimizes and generates the final result, and the system polishes these results through the medical vertical model or other related models, and finally returns the results to the AIAPI server.

[0096] In the disclosed embodiment, the AI ​​API server provides AI-related services and allows the client to interact with the backend AI model and system through the API interface. This server receives requests from users, is responsible for distributing the requests to the appropriate AI model or service module, processes them, and returns the results to the user.

[0097] In the disclosed embodiment, the demand understanding module refers to an AI module for understanding user needs, which parses user requests, extracts key information, and determines user intentions through natural language processing and semantic analysis technology.

[0098] In the disclosed embodiment, the execution planning module refers to a component used to coordinate and manage the interactions between various modules and services in the system. In the AI ​​system, the execution planning module is responsible for execution planning, determining which models or services to call based on the demand analysis of the demand understanding module, and arranging execution tasks in sequence.

[0099] In the disclosed embodiments, ANN search is a method for finding the data point closest to a given query point in a high-dimensional space. Since the computational complexity of an exact search in a high-dimensional space is very high, ANN search can find neighboring points close to the query point in a shorter time by using an approximate algorithm, thereby greatly improving the search efficiency.

[0100] For example, suppose a user requests the AIAPI server to query the latest diagnostic solutions for a specific disease. The system understands the needs through the needs understanding module and finds that the user needs belong to a specific medical field. A suitable medical vertical model is detected in the model architecture system, so the system determines that the model is the target model and skips the call of the general large model. The vertical model accesses the second data supply engine, retrieves relevant medical data, and ensures that the returned data is highly relevant through ANN search and fine sorting. Finally, the vertical model returns the generated analysis results to the user through the AIAPI server, providing accurate and fast diagnostic suggestions.

[0101] In this way, the system can select the most appropriate vertical model according to specific needs when processing user requests, and use efficient data retrieval and processing paths to provide users with accurate and fast responses. This flexible and efficient processing method is particularly suitable for vertical fields that require high customization and specialization, such as medical and financial industries.

[0102] In some embodiments, Figure 4 As shown, in combination with user needs, the target model and search path are determined for the search request from the model architecture system, which may include:

[0103] In the case where there is a vertical large model that can meet the user's needs in the model architecture system, the vertical large model that meets the user's needs is determined as the target model; and

[0104] The search path determined for the search request may include: using the vertical category model to access the second data supply engine, and generating search results based on the data returned by the second data supply engine.

[0105] In some embodiments, when the AIAPI server receives a user request, the system understands the demand and plans an execution plan through a demand understanding module and an execution planning module. In this process, the system generates an execution plan to determine whether it is necessary to call a general large model or a vertical large model to organize the final answer. If a vertical large model that can meet user needs is detected in the model architecture system, the system will determine it as the target model, skip the call of the general large model, and directly use the vertical large model to process the request. The vertical large model will retrieve data in a specific vertical field by accessing the second data supply engine. After receiving the data returned by the second data supply engine, the system uses this data to generate search results. This process usually involves efficient data retrieval and accurate result generation to ensure that the answer is highly relevant to user needs. Finally, the vertical large model returns the generated search results to the AIAPI server, and the system will integrate and finally regularize these results to ensure their accuracy and consistency, and then return the results to the user.

[0106] For example, suppose the system receives a user request to analyze the latest financial status of a company. The system understands the demand through the demand understanding module and the execution planning module, and determines that a specific financial vertical model is needed to process the request. The system detects that there is a suitable financial vertical model in the model architecture, so it is determined as the target model and the call of the general model is skipped. The financial vertical model retrieves financial data related to the company by accessing the second data supply engine. The system uses this data to generate accurate financial analysis reports to ensure that the results are highly relevant to user needs. Finally, the financial vertical model returns the generated analysis results to the AIAPI server. After the system integrates and organizes these results, it returns the final financial report to the user.

[0107] In this way, the system can flexibly select the most suitable vertical model according to specific user needs, and use efficient data supply paths to quickly generate and return highly relevant search results. This method is particularly suitable for processing search requests that require professional domain knowledge and in-depth analysis, ensuring that users receive accurate responses.

[0108] In some embodiments, Figure 5 As shown, in combination with user needs, the target model and search path are determined for the search request from the model architecture system, which may include:

[0109] In the case where there are vertical large models and vertical small models that can meet the user's needs in the model architecture system, the vertical large models and vertical small models that meet the user's needs are determined as target models; and

[0110] The search path determined for the search request may include: using the vertical category small model to access the second data supply engine, and sending the data returned by the second data supply engine to the vertical category large model, so that the vertical category large model generates search results.

[0111] In some embodiments, when the AIAPI server receives a user request, the system performs demand understanding and execution planning through a demand understanding module and an execution planning module. During this process, the system generates an execution plan and determines whether it is necessary to call a large vertical model and a small vertical model to meet user needs. If the system detects that there are large vertical models and small vertical models in the model architecture that can meet the needs, these two models are determined as target models. First, the system uses the small vertical model to access the second data supply engine to retrieve relevant data in a specific field. After obtaining the data, the data will be sent to the large vertical model for further processing and analysis to generate search results that meet user needs. Finally, the search results generated by the large vertical model will be returned to the AIAPI server, and the system will integrate and organize these results to ensure their accuracy and consistency, and then return the final results to the user.

[0112] For example, suppose a user requests a recommendation for the latest science and technology books. After the AIAPI server receives the request, the system understands the demand through the demand understanding module and the execution planning module, and finds that there are large and small vertical models of science and technology that meet the demand in the model architecture. The system first uses the small vertical model to access the second data supply engine to retrieve data related to science and technology books. After obtaining the data, the data is sent to the large vertical model for further analysis and optimization. Finally, the recommendation results generated by the large vertical model are returned to the AIAPI server, and the system integrates the results and provides them to the user.

[0113] In this way, by combining the vertical category large model and the vertical category small model, the system can make full use of their respective advantages. The vertical category small model first obtains accurate data, and then the vertical category large model performs in-depth analysis and optimization to generate high-quality search results. This method is particularly suitable for complex queries, ensuring that the system can quickly and accurately meet user needs.

[0114] In some embodiments, Figure 6 As shown, in combination with user needs, the target model and search path are determined for the search request from the model architecture system, which may include:

[0115] When there is no vertical large model or vertical small model that can meet the user's needs in the model architecture system, the second type of general large model is used as the target model; and

[0116] The search path determined for the search request may include: accessing the first data supply engine using the second type of general large model, and generating search results based on the data returned by the first data supply engine.

[0117] In some embodiments, when the AIAPI server receives a user request, the system understands the demand and performs execution planning through a demand understanding module and an execution planning module. When generating an execution plan, the system checks whether there is a vertical large model or a vertical small model that meets the user's needs in the model architecture system. If no vertical model that meets the requirements is found, the system determines the second type of general large model as the target model and uses the model to process the user request. The second type of general large model retrieves relevant general data by accessing the first data supply engine, and generates search results that meet user requirements based on this data. Finally, the generated search results are returned to the AIAPI server, and the system integrates and corrects the results to ensure their accuracy and consistency, and then returns the results to the user.

[0118] For example, suppose a user requests a recommendation for a destination suitable for a family trip. The system understands the user's needs and plans an execution plan through the demand understanding module and the execution planning module. When checking the model architecture system, it is found that there is no vertical large model or vertical small model that meets the demand. Therefore, the system chooses to use the second type of general large model to process the request. The second type of general large model retrieves general travel data related to family trips by accessing the first data supply engine, and generates recommendation results based on this data.

[0119] In this way, when the system cannot find a vertical model or a vertical model that meets the user's needs, the system will default to the second type of general large model to process the request. This method ensures that even in the absence of specialized models, the system can still effectively respond to user needs and provide reliable search results.

[0120] In some embodiments, Figure 7 As shown, combined with user needs, the target model and search path are determined for the search request from the model architecture system, which also includes:

[0121] When there is no vertical large model or vertical small model that can meet the user's needs in the model architecture system, the first type of general large model is used as the target model; and

[0122] The search path determined for the search request may include: accessing a first data supply engine using a first type of general large model, and generating search results based on data returned by the first data supply engine.

[0123] In some embodiments, when the AIAPI server receives a user request, the system understands the demand and generates an execution plan through the demand understanding module and the execution planning module. During this process, the system checks whether there are large vertical models or small vertical models in the model architecture system that can meet the user's needs. If no vertical model that meets the requirements is found, the system determines the first type of general large model as the target model and uses the model to process the request. The first type of general large model retrieves relevant general data by accessing the first data supply engine, and generates search results that meet user needs based on this data. The processed data is generated by the first type of general large model to generate the final search results and returned to the AIAPI server. The system then integrates and corrects the results to ensure their accuracy and consistency, and finally returns the results to the user.

[0124] For example, suppose a user request is received to obtain the latest international news. The system understands the demand through the demand understanding module and the execution planning module, and generates an execution plan. After checking the model architecture, it is found that there are no large vertical models or small vertical models suitable for the field of international news. Therefore, the system selects the first type of general large model as the target model. The first type of general large model retrieves relevant general news data by accessing the first data supply engine, and generates search results that meet user needs based on this data. Finally, the system integrates these search results and returns them to the user to ensure that the user can obtain accurate international news updates.

[0125] In this way, when the system cannot find a vertical model that meets the user's needs in the model architecture, it will call the first type of general large model by default. In this way, even if there is no specialized model, the system can still rely on the general large model and the first data supply engine to process the request and provide reliable search results.

[0126] Figure 8 is a schematic diagram of a processing flow after receiving a user request according to an embodiment of the present disclosure, such as Figure 8 As shown, after receiving the user request, the AIAPI server understands the demand through the demand understanding module, extracts the original information (including user demand), generates an execution plan through the execution planning module, and caches data through the storage module. The execution plan process usually includes three key steps: whether to request a general large model, whether to request a model of a certain vertical category, and the final answer organization. It should be noted that through demand identification and judgment, several steps can be skipped directly:

[0127] 1. Request a general large model access service. The large model access service internally decides how to perform background access.

[0128] 2. The big model service will first request the big model data supply engine to query the background data set. This query process is similar to the ANN search of the search request;

[0129] 3. After obtaining the search results returned by the search engine, first sort the data through fine sorting to select the top R results, then return the result with context request to the general large model for further processing, and finally return it to the upstream AIAPI server through the general large model access service;

[0130] 4. Assume that the demand type is identified as medical through demand understanding. The orchestrator finds that the backend itself contains a fine-tuning model of the medical type. Therefore, the backend directly distributes the request to the vertical backend service and addresses the fine-tuning model of the medical type.

[0131] 5. Search the medical database through ANN, and obtain the positive summary for data refinement. The NN index library created by the fine-tuning small model of the data is used in the process of searching the data.

[0132] 6. After the data is acquired, it is passed to the medical data fine-tuning model results for data supplementation and overall database construction results polishing, and finally returned to the AIAPI server through the vertical large model access server;

[0133] 7. At this point, all results are acquired, and the final data is normalized before being returned to the user.

[0134] This model architecture system can support diverse search needs by combining the advantages of general large models and vertical models. Whether it is extensive knowledge exploration or in-depth professional research, users can find search results that meet their needs in this system. Through the target model and path selection mechanism, the most appropriate model (whether it is a general large model or a vertical model) and search path can be intelligently selected based on user needs, avoiding unnecessary calculations and resource waste, thereby improving search efficiency.

[0135] In addition, the design of the model architecture system takes into account scalability and maintainability. With the continuous advancement of technology and the continuous changes in user needs, new models can be easily added, the model architecture can be adjusted, or the data supply engine can be updated to adapt to new application scenarios and needs. This model service provision method and its model architecture system effectively improve the search efficiency and result quality through the characteristics of intelligent matching of user needs and model resources, flexible data supply mechanism, and support for diverse search needs, which helps to provide users with a more efficient and accurate search experience.

[0136] Fig. 9 is a schematic diagram of the composition structure of the first data supply engine or the second data supply engine according to an embodiment of the present disclosure, such as Fig. 9As shown, the first data supply engine and the second data supply engine are both divided into three main modules: data reverse sorting, data summary and data forward sorting.

[0137] Here, data inversion is done by building an inverted index to organize the data index by keywords or specific attributes. The inverted index allows the system to quickly locate all records containing a keyword, which is very efficient in large-scale data retrieval. For example, in medical data processing, the inverted index can quickly find all medical records containing a symptom, diagnosis, or drug, thereby speeding up the search and analysis process.

[0138] Here, data summaries simplify and refine large amounts of data to generate concise summary information. These summaries usually retain the core information of the data, but reduce unnecessary redundancy, making it easier to browse and process quickly. For example, in medical scenarios, data summaries can provide a concise overview of patient medical records or a summary of research reports to help medical staff quickly understand the key content.

[0139] Here, the data is used to construct a positive index to organize the data in the original order or a specific order. The positive index ensures that the data can be fully accessed and analyzed in the original order. For example, in medical data processing, the positive index can be used to process continuous time series data, such as patient medical history, drug use records, etc., to ensure that the data is correctly presented in chronological order or record order.

[0140] In this way, by combining the three indexing methods of reverse, summary and forward, efficient retrieval, accurate positioning and sequential maintenance of data are ensured. It can quickly find relevant information in massive data and display it in a concise and orderly manner, greatly improving the system's processing efficiency and data utilization.

[0141] It should be understood that Figure 8 , Fig. 9 The schematic diagram shown is only exemplary and not restrictive, and it is scalable, and those skilled in the art can Figure 8 , Fig. 9 Various obvious changes and / or substitutions can be made to the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0142] The present disclosure provides a model service providing device, such as Fig.10 As shown, the model service providing device may include: a demand determination module 1001, used to determine user demand based on a received search request; a path selection module 1002, used to determine a target model and a search path for the search request from a model architecture system in combination with user demand; and an execution control module 1003, used to control the target model to execute the search path to obtain the search results output by the target model.

[0143] In some embodiments, the model architecture system includes: a first type of general large model, a second type of general large model, M vertical large models, N vertical small models, a first data supply engine and a second data supply engine. The first type of general large model and the second type of general large model are respectively connected to the first data supply engine; the vertical large model and the vertical small model are respectively connected to the second data supply engine. The order of magnitude of the first type of general large model is greater than that of the second type of general large model. M is an integer not less than 1, and N is an integer not less than 1.

[0144] In some embodiments, the first type of general big model is a big model for providing general data services, the order of magnitude of the first type of general big model is a first value, and the update period is a first duration; the second type of general big model is a big model for providing general data services, the order of magnitude of the second type of general big model is a second value, and the update period is a second duration, the second value is smaller than the first value, and the second duration is smaller than the first duration; the vertical big model is a big model for providing vertical business data services; the order of magnitude of the vertical big model is a third value, and the update period is a third duration, and the third value is smaller than the second value; the vertical small model is a small model for providing vertical business data services; the order of magnitude of the vertical small model is a fourth value, and the update period is a fourth duration, the fourth value is smaller than the third value, and the fourth duration is smaller than the third duration.

[0145] In some implementations, the path selection module 1002 is used to:

[0146] In the case where there is a small vertical model that can meet the user's needs in the model architecture system, the small vertical model that meets the user's needs is determined as the target model; and

[0147] The search path determined for the search request includes: using the vertical category model to access the second data supply engine, and generating search results based on the data returned by the second data supply engine.

[0148] In some implementations, the path selection module 1002 is used to:

[0149] In the case where there is a vertical large model that can meet the user's needs in the model architecture system, the vertical large model that meets the user's needs is determined as the target model; and

[0150] The search path determined for the search request includes: using the vertical category model to access the second data supply engine, and generating search results based on the data returned by the second data supply engine.

[0151] In some implementations, the path selection module 1002 is used to:

[0152] In the case where there are vertical large models and vertical small models that can meet the user's needs in the model architecture system, the vertical large models and vertical small models that meet the user's needs are determined as target models; and

[0153] The search path determined for the search request includes: using the vertical category small model to access the second data supply engine, and sending the data returned by the second data supply engine to the vertical category large model, so that the vertical category large model generates search results.

[0154] In some implementations, the path selection module 1002 is used to:

[0155] When there is no vertical large model or vertical small model that can meet the user's needs in the model architecture system, the second largest model is used as the target model; and

[0156] The search path determined for the search request includes: accessing the first data supply engine using the second type of general large model, and generating search results based on the data returned by the first data supply engine.

[0157] In some implementations, the path selection module 1002 is used to:

[0158] In the case that there is no vertical large model or vertical small model that can meet the user's needs in the model architecture system, the first large model is used as the target model; and

[0159] The search path determined for the search request includes: accessing a first data supply engine using a first type of general large model, and generating search results based on data returned by the first data supply engine.

[0160] The present disclosure provides a scenario diagram of a method for providing a model service, such as Fig.11 shown.

[0161] As mentioned above, the model service providing method provided by the embodiment of the present disclosure is applied to electronic devices. The electronic devices are intended to represent various forms of digital computers, such as

[0162] For example, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices.

[0163] Specifically, the electronic device can perform the following operations:

[0164] Determining user needs based on received search requests;

[0165] Based on user needs, determine the target model and search path for the search request from the model architecture system;

[0166] Control the target model to execute the search path to obtain the search results output by the target model.

[0167] It should be understood that Fig.11 The scene diagram shown is only illustrative and not restrictive. Those skilled in the art can Fig.11 Various obvious changes and / or substitutions can be made to the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0168] In the technical solution disclosed herein, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0169] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0170] Fig.12 A schematic block diagram of an example electronic device 1200 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0171] like Fig.12 As shown, the device 1200 includes a computing unit 1201, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1202 or a computer program loaded from a storage unit 1208 to a random access memory (RAM) 1203. In the RAM 1203, various programs and data required for the operation of the device 1200 can also be stored. The computing unit 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0172] A number of components in the device 1200 are connected to the I / O interface 1205, including: an input unit 1206, such as a keyboard, a mouse, etc.; an output unit 1207, such as various types of displays, speakers, etc.; a storage unit 1208, such as a disk, an optical disk, etc.; and a communication unit 1209, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1209 allows the device 1200 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0173] The computing unit 1201 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, a CPU, a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSP), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1201 performs the various methods and processes described above, such as the target model service providing method. For example, in some embodiments, the target model service providing method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1208. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1200 via the ROM 1202 and / or the communication unit 1209. When the computer program is loaded into the RAM 1203 and executed by the computing unit 1201, one or more steps of the model service providing method described above may be performed. Alternatively, in other embodiments, the computing unit 1201 may be configured to execute the model service providing method in any other appropriate manner (eg, by means of firmware).

[0174] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0175] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0176] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0177] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube (CRT) or a liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0178] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

[0179] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0180] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0181] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A model service providing method, applied to a preset model architecture system, wherein: The method comprises: Determining user needs based on received search requests; In combination with the user demand, determining a target model and a search path for the search request from the model architecture system; The target model is controlled to execute the search path to obtain the search result output by the target model.

2. The method according to claim 1, wherein: The model architecture system includes: a first type of general large model, a second type of general large model, M vertical large models, N vertical small models, a first data supply engine and a second data supply engine. The first type of general large model and the second type of general large model are respectively connected to the first data supply engine; the vertical large model and the vertical small model are respectively connected to the second data supply engine. The order of magnitude of the first type of general large model is greater than that of the second type of general large model. M is an integer not less than 1, and N is an integer not less than 1.

3. The method according to claim 2, wherein: The first type of general large model is a large model used to provide general data services, the order of magnitude of the first type of general large model is a first value, and the update period is a first duration; The second type of general large model is a large model used to provide general data services, the order of magnitude of the second type of general large model is a second value, the update period is a second duration, the second value is less than the first value, and the second duration is less than the first duration; The vertical category big model is a big model for providing vertical category business data services; the order of magnitude of the vertical category big model is a third value, the update period is a third duration, and the third value is less than the second value; Among them, the vertical small model is a small model used to provide vertical business data services; the order of magnitude of the vertical small model is a fourth value, the update period is a fourth time duration, the fourth value is smaller than the third value, and the fourth time duration is smaller than the third time duration.

4. The method according to claim 2, wherein: The first version of the vertical big model for the target vertical business is a big model for the target vertical business obtained by training based on the second type of general big model; The P-th version of the vertical big model for the target vertical business is a big model trained based on the P-1-th version of the vertical big model for the target vertical business; wherein P is an integer not less than 2.

5. The method according to claim 2, wherein: The first version of the small vertical model for the target vertical business is a small model for the target vertical business obtained by training based on the second type of general large model; The Kth version of the vertical small model for the target vertical business is a small model trained based on the K-1th version of the vertical small model for the target vertical business; wherein K is an integer not less than 2.

6. The method according to any one of claims 1 to 5, wherein: The step of determining a target model and a search path for the search request from the model architecture system in combination with the user demand includes: In the case where there is a small vertical model that can meet the user's needs in the model architecture system, determining the small vertical model that meets the user's needs as the target model; and The search path determined for the search request includes: using the vertical category model to access the second data supply engine, and generating the search results based on the data returned by the second data supply engine.

7. The method according to any one of claims 1 to 5, wherein: The step of determining a target model and a search path for the search request from the model architecture system in combination with the user demand includes: In the case where there is a vertical large model that can meet the user's needs in the model architecture system, determining the vertical large model that meets the user's needs as the target model; and The search path determined for the search request includes: using the vertical category model to access the second data supply engine, and generating the search results based on the data returned by the second data supply engine.

8. The method according to any one of claims 1 to 5, wherein: The step of determining a target model and a search path for the search request from the model architecture system in combination with the user demand includes: In the case where there are vertical large models and vertical small models that can meet the user's needs in the model architecture system, the vertical large models and vertical small models that meet the user's needs are determined as the target models; and The search path determined for the search request includes: using the vertical category small model to access the second data supply engine, and sending the data returned by the second data supply engine to the vertical category large model, so that the vertical category large model generates the search results.

9. The method according to any one of claims 1 to 5, wherein: The step of determining a target model and a search path for the search request from the model architecture system in combination with the user demand includes: In the case that there is no vertical large model or vertical small model in the model architecture system that can meet the user's needs, using the second type of general large model as the target model; and The search path determined for the search request includes: accessing the first data supply engine using the second type of general large model, and generating the search result based on the data returned by the first data supply engine.

10. The method according to any one of claims 1 to 5, wherein: The step of determining a target model and a search path for the search request from the model architecture system in combination with the user demand includes: In the case that there is no vertical large model or vertical small model in the model architecture system that can meet the user's needs, using the first type of general large model as the target model; and The search path determined for the search request includes: accessing the first data supply engine using the first type of general large model, and generating the search result based on the data returned by the first data supply engine.

11. A model service providing device, applied to a preset model architecture system, wherein: The device comprises: A demand determination module, used to determine user demand based on the received search request; A path selection module, used to determine a target model and a search path for the search request from the model architecture system in combination with the user requirements; An execution control module is used to control the target model to execute the search path to obtain the search result output by the target model.

12. The device according to claim 11, wherein: The model architecture system includes: a first type of general large model, a second type of general large model, M vertical large models, N vertical small models, a first data supply engine and a second data supply engine. The first type of general large model and the second type of general large model are respectively connected to the first data supply engine; the vertical large model and the vertical small model are respectively connected to the second data supply engine. The order of magnitude of the first type of general large model is greater than that of the second type of general large model. M is an integer not less than 1, and N is an integer not less than 1.

13. The device according to claim 12, wherein: The first type of general large model is a large model used to provide general data services, the order of magnitude of the first type of general large model is a first value, and the update period is a first duration; The second type of general large model is a large model used to provide general data services, the order of magnitude of the second type of general large model is a second value, the update period is a second duration, the second value is less than the first value, and the second duration is less than the first duration; The vertical category big model is a big model for providing vertical category business data services; the order of magnitude of the vertical category big model is a third value, the update period is a third duration, and the third value is less than the second value; Among them, the vertical small model is a small model used to provide vertical business data services; the order of magnitude of the vertical small model is a fourth value, the update period is a fourth time duration, the fourth value is smaller than the third value, and the fourth time duration is smaller than the third time duration.

14. The device according to any one of claims 12 to 13, wherein: The path selection module is used to: In the case where there is a small vertical model that can meet the user's needs in the model architecture system, determining the small vertical model that meets the user's needs as the target model; as well as The search path determined for the search request includes: using the vertical category model to access the second data supply engine, and generating the search results based on the data returned by the second data supply engine.

15. The device according to any one of claims 12 to 13, wherein: The path selection module is used to: In the case where there is a vertical large model that can meet the user's needs in the model architecture system, determining the vertical large model that meets the user's needs as the target model; as well as The search path determined for the search request includes: using the vertical category model to access the second data supply engine, and generating the search results based on the data returned by the second data supply engine.

16. The device according to any one of claims 12 to 13, wherein: The path selection module is used to: In the case where there are vertical large models and vertical small models that can meet the user's needs in the model architecture system, the vertical large models and vertical small models that meet the user's needs are determined as the target models; as well as The search path determined for the search request includes: using the vertical category small model to access the second data supply engine, and sending the data returned by the second data supply engine to the vertical category large model, so that the vertical category large model generates the search results.

17. The device according to any one of claims 12 to 13, wherein: The path selection module is used to: In the case that there is no vertical large model or vertical small model that can meet the user's needs in the model architecture system, the second large model is used as the target model; as well as The search path determined for the search request includes: using the second large model to access the first data supply engine, and generating the search result based on the data returned by the first data supply engine.

18. The device according to any one of claims 12 to 13, wherein: The path selection module is used to: In the case that there is no vertical large model or vertical small model that can meet the user's needs in the model architecture system, taking the first large model as the target model; as well as The search path determined for the search request includes: accessing the first data supply engine using the first large model, and generating the search result based on the data returned by the first data supply engine.

19. An electronic device comprising: at least one processor; as well as a memory communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 10.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are for causing a computer to perform a method according to any one of claims 1-10.

21. A computer program product comprising a computer program stored on a storage medium, the computer program implementing the method according to any one of claims 1 to 10 when executed by a processor.

Images