Model validation implementation methods, devices, and servers

Abstract

This disclosure provides a model verification method, apparatus, and server, relating to artificial intelligence technology. The method includes: acquiring an application call request sent by a terminal, the application call request including a model package of a first model to be verified, and a first machine learning platform and a first framework corresponding to the first model; parsing the model package of the first model using a preset method to obtain the model structure corresponding to the first model; determining the target model specification corresponding to the first machine learning platform and the first framework according to a preset specification mapping table; verifying the model structure corresponding to the first model using the target model specification to obtain the verification result of the first model; and feeding back the verification result to the terminal. This solution can parse and process the model and its corresponding model specification, converting them into a simpler method than existing technologies before verification, making it easier to understand and maintain.

Description

Technical Field

[0001] This disclosure relates to artificial intelligence technology, and in particular to a method, apparatus and server for implementing model verification. Background Technology

[0002] With the rise of artificial intelligence, machine learning has swept across the technology industry in recent years. Machine learning platforms aim to provide one-stop services for machine learning practitioners, defining model specifications applicable to the platform's own machine learning implementation processes. These machine learning platforms are welcomed by many enterprises and the open-source community, and are updated and iterated frequently. A problem that arises is that the model specifications defined by different machine learning platforms often differ significantly, and even different versions of the same platform may be incompatible. When developing, applying, and managing models internally, enterprises need to uniformly verify models produced by developers with different technology stacks and from different machine learning platforms to ensure that model files, model data, and model engineering conform to the corresponding specifications.

[0003] In existing technologies, to support different versions of model specifications across various machine learning platforms, abstraction is often done at the code level, with additional code added to handle changes and extensions to the model specifications. In practice, this involves hard-coding and writing fixed rules to validate the model structure. A more elegant approach is to save the rules to a database and provide a visual interface for rule editing.

[0004] However, due to the extensive and complex engineering involved in machine learning, and the intricate model structures, the aforementioned verification methods are also quite complex and difficult to understand and maintain. Summary of the Invention

[0005] This disclosure provides a method, apparatus, and server for model verification, to address the problems that existing model verification methods are complex and difficult to understand and maintain.

[0006] According to the first aspect of this disclosure, a method for implementing model validation is provided, including:

[0007] Obtain the application call request sent by the terminal, wherein the application call request includes the model package of the first model to be verified, as well as the first machine learning platform and the first framework corresponding to the first model;

[0008] The model package of the first model is parsed using a preset method to obtain the model structure corresponding to the first model;

[0009] According to a preset specification mapping table, the target model specification corresponding to the first machine learning platform and the first framework is determined; wherein, the preset specification mapping table includes the mapping relationship between machine learning platforms, frameworks, and model specifications; the model specification is determined by parsing the original model specification;

[0010] Using the target model specification, the model structure corresponding to the first model is verified, and the verification result of the first model is obtained; the verification result is then fed back to the terminal.

[0011] According to a second aspect of this disclosure, a model verification implementation apparatus is provided, comprising:

[0012] The acquisition unit is used to acquire the application call request sent by the terminal, wherein the application call request includes the model package of the first model to be verified and the first machine learning platform and the first framework corresponding to the first model;

[0013] The model parsing unit is used to parse the model package of the first model using a preset method to obtain the model structure corresponding to the first model.

[0014] The specification mapping unit is used to determine the target model specification corresponding to the first machine learning platform and the first framework according to a preset specification mapping table; wherein, the preset specification mapping table includes the mapping relationship between machine learning platforms, frameworks, and model specifications; the model specification is determined by parsing the original model specification;

[0015] The model verification unit is used to verify the model structure corresponding to the first model using the target model specification, obtain the verification result of the first model, and feed back the verification result to the terminal.

[0016] According to a third aspect of this disclosure, a server is provided, including a memory and a processor; wherein,

[0017] The memory is used to store computer programs;

[0018] The processor is configured to read the computer program stored in the memory and execute the model verification implementation method as described in the first aspect according to the computer program in the memory.

[0019] According to a fourth aspect of this disclosure, a computer-readable storage medium is provided, wherein computer-executable instructions are stored therein, and when a processor executes the computer-executable instructions, the model verification implementation method as described in the first aspect is implemented.

[0020] According to a fifth aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the model verification method as described in the first aspect.

[0021] The model verification implementation method, apparatus, and server disclosed herein include: obtaining an application call request sent by a terminal, the application call request including a model package of a first model to be verified, and a first machine learning platform and a first framework corresponding to the first model; parsing the model package of the first model using a preset method to obtain the model structure corresponding to the first model; determining the target model specification corresponding to the first machine learning platform and the first framework according to a preset specification mapping table; wherein the preset specification mapping table includes the mapping relationship between the machine learning platform, the framework, and the model specification; the model specification is determined by parsing the original model specification; verifying the model structure corresponding to the first model using the target model specification, obtaining the verification result of the first model; and feeding back the verification result to the terminal. The model verification implementation method, apparatus, and server provided in this solution can parse the model and its corresponding model specification, converting them into a simpler method than existing technologies before verification, making them easier to understand and maintain. Furthermore, the use of a preset specification mapping table improves the universality of the model specification. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating an exemplary embodiment of the model verification method disclosed herein.

[0024] Figure 2 This is a flowchart illustrating a model validation implementation method as shown in another exemplary embodiment of this disclosure;

[0025] Figure 3 This is a schematic diagram illustrating the implementation process of model verification as an exemplary embodiment of the present disclosure;

[0026] Figure 4 A structural diagram illustrating a model verification implementation apparatus as shown in an exemplary embodiment of this disclosure;

[0027] Figure 5 This is a structural diagram of a server shown as an exemplary embodiment of the present disclosure. Detailed Implementation

[0028] With the rise of artificial intelligence, machine learning has swept across the tech industry in recent years. However, model training code is only a small part of the machine learning workflow. Other indispensable components include monitoring training tasks, log collection, hyperparameter selection and optimization, model deployment and integration, data cleaning, and feature extraction. Due to the extensive and complex engineering involved in machine learning, numerous company products and open-source tools help users better implement their machine learning initiatives. Currently, major tech companies promote machine learning platforms such as Kubeflow, MLflow, Argo, AirFlow, Seldon, Comet, FloydHub, RiseML, and SageMaker. These platforms aim to provide one-stop services for machine learning practitioners, defining model specifications suitable for implementing machine learning workflows within the platform itself. These platforms are popular with many enterprises and the open-source community, and are frequently updated. However, this also leads to significant differences in the model specifications defined by various machine learning platforms, and even incompatibility can exist between different versions of the same platform. When developing, applying, and managing models within an enterprise, it is necessary to conduct unified verification of models produced by R&D personnel with different technology stacks and different machine learning platforms to ensure that model files, model data, and model engineering comply with the relevant specifications.

[0029] In existing technologies, to support different versions of model specifications across various machine learning platforms, abstraction is often done at the code level, with additional code added to handle changes and extensions to the model specifications. In practice, this involves hard-coding and writing fixed rules to validate the model structure. A more elegant approach is to save the rules to a database and provide a visual interface for rule editing, but these are still fixed rules.

[0030] However, due to the extensive and complex engineering involved in machine learning, and the intricate structure of models, the aforementioned verification methods are correspondingly complex and difficult to understand and maintain. Furthermore, existing technologies cannot flexibly extend to models produced by developers using different machine learning platforms and technology stacks, often requiring additional code to extend the verification module.

[0031] To address the aforementioned technical issues, the solution provided in this disclosure parses the model and its corresponding specifications, converting them into a simpler format than existing technologies before verification, making them easier to understand and maintain. Furthermore, it employs a pre-defined specification mapping table, improving the universality of the model specifications.

[0032] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0033] The technical solutions of this disclosure and how they solve the aforementioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this disclosure will now be described with reference to the accompanying drawings.

[0034] Figure 1 This is a flowchart illustrating an exemplary embodiment of the model verification method of this disclosure.

[0035] like Figure 1 As shown, the model validation implementation method provided in this embodiment includes:

[0036] Step 101: Obtain the application call request sent by the terminal. The application call request includes the model package of the first model to be verified, as well as the first machine learning platform and the first framework corresponding to the first model.

[0037] The method provided in this disclosure can be executed by a server.

[0038] The server can be pre-configured with an Application Programming Interface (API) for model validation, which can be called by the terminal. The terminal can deploy a Software Development Kit (SDK) corresponding to the API for model validation. Specifically, the server can receive application call requests sent by the terminal and invoke the application for model validation.

[0039] Specifically, the application call request may also include the model package of the first model to be verified, as well as the first machine learning platform and the first framework corresponding to the first model.

[0040] Specifically, according to the specified model specifications, the model files, model data, model project, and configuration files corresponding to the first model can be packaged into a model package for the first model.

[0041] In this context, a model refers to a numerical method or system that uses statistical, economic, financial, or mathematical theories and techniques to transform input data into numerical predictive conclusions. It also includes methods or systems that output numerical results based on some or all qualitative inputs, or based on expert judgment. A model consists of three parts: an input section, which feeds data into the model; a processing section, which transforms the input into predictive conclusions; and a reporting section, which translates the predictive conclusions into usable business information.

[0042] Among them, model files refer to the serialization of the trained model in memory and its persistence to the hard disk in the form of files.

[0043] Model data typically includes the training dataset, test dataset, and other supplementary data required for the model.

[0044] Model engineering refers to the engineering files and configuration files required for subsequent operations such as model deployment and retraining.

[0045] Model validation refers to verifying whether a model conforms to the model specifications of the platform or operating environment before its use. Specifically, it verifies whether the first model conforms to the model specifications corresponding to the first machine learning platform and framework.

[0046] Step 102: Using a preset method, the model package of the first model is parsed to obtain the model structure corresponding to the first model.

[0047] Specifically, a pre-set method can be used to parse the model package of the first model, convert the contents of the model package into a more concise data format, and determine the converted data format as the model structure corresponding to the first model.

[0048] Step 103: Determine the target model specification corresponding to the first machine learning platform and the first framework according to the preset specification mapping table; wherein, the preset specification mapping table includes the mapping relationship between machine learning platform, framework and model specification; the model specification is determined by parsing the original model specification.

[0049] The preset specification mapping table is a pre-set specification mapping table based on actual conditions. This preset specification mapping table includes the mapping relationship between machine learning platforms, frameworks, and model specifications.

[0050] The model specification is determined by parsing the original model specification using a method corresponding to the parsed model package.

[0051] Specifically, based on the first machine learning platform and the first framework, a preset specification mapping table can be matched to obtain the model specification corresponding to the first machine learning platform and the first framework, and this model specification can be determined as the target model specification of the first model.

[0052] Step 104: Using the target model specification, verify the model structure corresponding to the first model, obtain the verification result of the first model, and feed the verification result back to the terminal.

[0053] Specifically, the target model specification can be used to validate the model structure corresponding to the first model, and the validation result of the first model can be obtained. The validation result lists the parts of the model structure that do not conform to the target model specification.

[0054] The verification result can then be fed back to the terminal so that the terminal can use the verification result to adjust the first model.

[0055] The model verification method disclosed herein includes: obtaining an application call request sent by a terminal, the application call request including the model package of a first model to be verified, as well as the first machine learning platform and the first framework corresponding to the first model; parsing the model package of the first model using a preset method to obtain the model structure corresponding to the first model; determining the target model specification corresponding to the first machine learning platform and the first framework according to a preset specification mapping table; wherein, the preset specification mapping table includes the mapping relationship between the machine learning platform, the framework, and the model specification; the model specification is determined by parsing the original model specification; verifying the model structure corresponding to the first model using the target model specification, obtaining the verification result of the first model; and feeding back the verification result to the terminal. The method adopted in this disclosure can parse the model and its corresponding model specification, converting them into a simpler method than existing technologies before verification, making it easier to understand and maintain. Furthermore, the use of a preset specification mapping table improves the universality of the model specification.

[0056] Figure 2 This is a flowchart illustrating a method for implementing model verification, as shown in another exemplary embodiment of this disclosure.

[0057] like Figure 2 As shown, the model validation implementation method provided in this embodiment includes:

[0058] Step 201: Obtain the application call request sent by the terminal. The application call request includes the model package of the first model to be verified, as well as the first machine learning platform and the first framework corresponding to the first model.

[0059] Specifically, the principle and implementation of step 201 are similar to those of step 101, and will not be repeated here.

[0060] Step 202: Convert the directory structure, file structure, file format, file content and attribute values ​​included in the model package of the first model into a data exchange format string, and determine the data exchange format string as the model structure corresponding to the first model.

[0061] Specifically, the directory structure, file structure, file format, file content, and attribute values ​​included in the model package of the first model can be converted into a JavaScript Object Notation (JSON) string, and this JSON string can be used as the model structure corresponding to the first model.

[0062] JSON is a lightweight data exchange format that uses a language-independent text format.

[0063] Step 203: Determine the target model specification corresponding to the first machine learning platform and the first framework according to the preset specification mapping table; wherein, the preset specification mapping table includes the mapping relationship between machine learning platform, framework and model specification; the model specification is determined by parsing the original model specification.

[0064] Specifically, based on the first machine learning platform and the first framework, a preset specification mapping table can be matched to obtain the model specification corresponding to the first machine learning platform and the first framework, and this model specification can be determined as the target model specification of the first model.

[0065] Specifically, the model specification is determined by parsing the original model specification using a method corresponding to the parsed model package.

[0066] For example, the model specification can be a JSON schema. A JSON schema is used to describe the JSON data format and defines a standard for JSON data constraints. Based on this convention, both parties exchanging data can understand the requirements and constraints of the JSON data, and can also use it to verify the information, ensuring the correctness of the data exchange.

[0067] In one possible implementation, a pre-defined specification mapping table is visualized.

[0068] In one possible implementation, a pre-defined specification mapping table is edited in response to a user's mapping editing command.

[0069] Specifically, the preset specification mapping table can be visualized. Furthermore, users can edit the preset specification mapping table visually.

[0070] In one feasible approach, a pre-defined version management table is used to manage the model specifications.

[0071] Specifically, a pre-set version management table can be used to manage model specifications. This pre-set version management table can be linked to a pre-set specification mapping table. The pre-set specification mapping table can retrieve the corresponding model specification from the pre-set version management table based on the model specification's name.

[0072] In one possible implementation, a pre-defined version management table is displayed visually.

[0073] In one possible implementation, the preset version management table is edited in response to the user's specified editing instructions.

[0074] Specifically, the preset version management table can be displayed visually. Furthermore, users can edit the preset version management table visually.

[0075] Specifically, this solution can provide version management and visual editing of the JSON Schema corresponding to different versions of the original model specifications for various machine learning platforms.

[0076] Step 204: Identify the configuration information in the model package of the first model, and update the target model specification according to the configuration information.

[0077] Specifically, when parsing the model package of the first model, the configuration information within the model package can be identified simultaneously. This configuration information can be configuration items in the configuration files included in the model package. For example, if the configuration file `my.conf` in the model package has a configuration item `env_file:abc.env`, then parsing the `my.conf` file and identifying the `env_file` configuration item can automatically add a rule to validate the `abc.env` file in the currently loaded JSON-Schema, without needing to predefine a fixed rule for validating the `abc.env` file, thus achieving dynamic model validation. Note that `abc.env` here is a user-defined file; the model specification is unaware of its existence beforehand. Documents can be predefined as needed, such as `abc.env`, `def.prop`, etc.

[0078] Specifically, based on this configuration information, automatic specification expansion can be performed, and the expanded model specification can be used to update the target model specification.

[0079] In one possible implementation, the extended specification corresponding to the configuration information is determined based on the configuration information and a preset specification extension mapping table; the preset specification extension mapping table includes the mapping relationship between the configuration information and the extended specification.

[0080] The configuration information may include the attributes, parameters, and runtime environment configuration of the first model.

[0081] The preset specification extension mapping table is a pre-configured specification extension mapping table based on actual conditions. This preset specification extension mapping table includes the mapping relationship between configuration information and extension specifications.

[0082] The extended specification can be determined by parsing the original model specification using a method corresponding to the parsing model package. Specifically, the extended specification can be in JSON format.

[0083] Specifically, based on the configuration information of the first identified model, a preset specification extension mapping table can be queried to obtain the extension specification corresponding to the configuration information.

[0084] Then, the extended specification is added to the target model specification to generate the updated target model specification.

[0085] Specifically, the extended specifications can be added to the target model specifications to generate an updated target model specifications.

[0086] Furthermore, pre-defined specification tables can be used to manage extended specifications. These pre-defined specification tables can be visualized, allowing users to edit them visually.

[0087] Furthermore, the preset canonical extension mapping table can be visualized. Users can also edit this preset canonical extension mapping table visually.

[0088] Specifically, by using a pre-defined specification extension mapping table, extended specifications can be automatically added based on the identified configuration information, thereby updating the target model specification. It can automatically adapt to the attributes, parameters, and runtime environment configurations of different models, allowing for flexible specification extension.

[0089] For example, model A is a Python model, and the runtime environment configuration may specify a Python version requirement; model B is a Java model, and the runtime environment may specify a Java version requirement. Another example is binary classification models and clustering models; these two types of models have different hyperparameters, which need to be automatically adjusted according to the configuration file.

[0090] Step 205: Using the updated target model specification, verify the model structure corresponding to the first model, obtain the verification result of the first model, and feed the verification result back to the terminal.

[0091] Specifically, the updated target model specification can be used to validate the model structure corresponding to the first model. Specifically, JSON Schema can be used to verify whether the JSON string representing the machine learning model structure conforms to the constraints of the JSON Schema defining the model specification. The validation result for the first model is then obtained. This result lists the parts of the model structure that do not conform to the target model specification.

[0092] The verification result can then be fed back to the terminal so that the terminal can use the verification result to adjust the first model.

[0093] Specifically, this solution uses JSON Schema to define model specifications, transforming complex model specifications into structured data that is easier for users to understand and maintain. A JSON Schema validator is used to verify whether the model structure (JSON string) conforms to the model specification (JSON Schema). Simultaneously, it supports parsing attribute values ​​from the model configuration file during the validation process, leveraging the characteristics of JSON Schema to automatically add rules and dynamically adjust the model specification for this validation. This transforms complex model validation into a simple, flexible, and dynamic JSON string validation.

[0094] Furthermore, through visualization, version management, and specification mapping, it achieves unified management and convenient maintenance of model specifications for different versions of different machine learning platforms.

[0095] Figure 3 This is a schematic diagram illustrating the implementation process of model verification as an exemplary embodiment of this disclosure.

[0096] like Figure 3 As shown, 1 indicates that users can input / edit model specifications (JSON Schema) through the visual interface. 2 indicates that users can input / edit the mapping relationship between model specifications and machine learning platforms and frameworks through the visual interface. 3 indicates that the obtained model can be input into the model parser through the API / SDK (i.e., a pre-set application interface), and the model parser will parse it to obtain the model structure (JSON string). 4 indicates that the API / SDK can pass the machine learning platform and framework information corresponding to the model to the model validator. 5 indicates that the model validator can obtain the model structure (JSON string) from the model parser. 6 indicates that the model validator can obtain the model specification (JSON Schema) from the version management and specification mapping module based on the machine learning platform and framework information. Subsequently, the model validator can use the model specification to validate the model and obtain the validation result. 7 indicates that the model validator can send the validation result to the user / caller.

[0097] Figure 3In this system, version management enables real-time editing and version control of model specifications defined by JSON Schema. Specification mapping establishes the mapping relationship between model specifications and various machine learning platforms and frameworks, allowing for dynamic user adjustments. Visualization provides visual support for version management and specification mapping, facilitating user operation and management. The model parser reads machine learning models, mapping their directory structure, file structure, file format, file content, and attribute values ​​into JSON strings. The model validator reads the corresponding model specification based on the machine learning platform and framework used by the model, validates the JSON string representing the model structure using JSON Schema, and supports parsing attribute values ​​from the model configuration file during validation, automatically adding rules and dynamically adjusting the JSON Schema for this validation using the characteristics of JSON Schema. The API / SDK provides the model validation capabilities of this invention to the application layer and external systems.

[0098] Figure 4 This is a structural diagram illustrating a model verification implementation apparatus as an exemplary embodiment of the present disclosure.

[0099] like Figure 4 As shown, the model verification implementation device 400 provided in this disclosure includes:

[0100] The acquisition unit 410 is used to acquire the application call request sent by the terminal. The application call request includes the model package of the first model to be verified, as well as the first machine learning platform and the first framework corresponding to the first model.

[0101] The model parsing unit 420 is used to parse the model package of the first model using a preset method to obtain the model structure corresponding to the first model.

[0102] The specification mapping unit 430 is used to determine the target model specification corresponding to the first machine learning platform and the first framework according to a preset specification mapping table; wherein, the preset specification mapping table includes the mapping relationship between the machine learning platform, the framework, and the model specification; the model specification is determined by parsing the original model specification;

[0103] The model verification unit 440 is used to verify the model structure corresponding to the first model using the target model specification, obtain the verification result of the first model, and feed the verification result back to the terminal.

[0104] The model verification unit 440 is specifically used to identify the configuration information in the model package of the first model and update the target model specification according to the configuration information.

[0105] The updated target model specification is used to validate the model structure corresponding to the first model.

[0106] The model verification unit 440 is specifically used to determine the extended specification corresponding to the configuration information based on the configuration information and the preset specification extension mapping table; the preset specification extension mapping table includes the mapping relationship between the configuration information and the extended specification.

[0107] Add the extended specification to the target model specification to generate the updated target model specification.

[0108] The model parsing unit 420 is specifically used to convert the directory structure, file structure, file format, file content and attribute values ​​included in the model package of the first model into a data exchange format string, and to determine the data exchange format string as the model structure corresponding to the first model.

[0109] The model verification implementation apparatus 400 provided in this disclosure further includes:

[0110] Visualization unit 450 is used to visualize and display a preset specification mapping table;

[0111] Responding to user-defined mapping edit commands, the system edits the preset standard mapping table.

[0112] Version management unit 460 is used to manage model specifications using a preset version management table.

[0113] The visualization unit 450 is also used to visualize and display a preset version management table;

[0114] Responding to the user's specified editing instructions, the preset version management table is edited.

[0115] Figure 5 This is a structural diagram of a server shown as an exemplary embodiment of the present disclosure.

[0116] like Figure 5 As shown, the server provided in this embodiment includes:

[0117] Memory 501;

[0118] Processor 502; and

[0119] Computer programs;

[0120] The computer program is stored in memory 501 and configured to be executed by processor 502 to implement any of the above-described model verification methods.

[0121] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement any of the model verification methods described above.

[0122] This embodiment also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described model verification methods.

[0123] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of implementing model checking, characterized by, The application comprises the following steps: An application calling request sent by a terminal is acquired, the application calling request comprising a model package of a first model to be verified and a first machine learning platform and a first framework corresponding to the first model; The model package of the first model is parsed in a preset manner to acquire a model structure corresponding to the first model; A target model specification corresponding to the first machine learning platform and the first framework is determined according to a preset specification mapping table, wherein the preset specification mapping table comprises a mapping relationship among machine learning platforms, frameworks and model specifications, and the model specification is determined by parsing an original model specification; The model structure corresponding to the first model is verified by using the target model specification to acquire a verification result of the first model, and the verification result is fed back to the terminal; The model structure corresponding to the first model is verified by using the target model specification, which comprises the following steps: Configuration information in the model package of the first model is identified, and the target model specification is updated according to the configuration information; The model structure corresponding to the first model is verified by using the updated target model specification; The target model specification is updated according to the configuration information, which comprises the following steps: An extended specification corresponding to the configuration information is determined according to the configuration information and a preset specification extension mapping table, wherein the preset specification extension mapping table comprises a mapping relationship between configuration information and extended specifications; The extended specification is added to the target model specification to generate an updated target model specification.

2. The method of claim 1, wherein, The model package of the first model is parsed in a preset manner to acquire the model structure corresponding to the first model, which comprises the following steps: Directory structure, file structure, file format, file content and attribute values included in the model package of the first model are converted into a data exchange format string, and the data exchange format string is determined as the model structure corresponding to the first model.

3. The method according to any of claims 1-2, characterized in that, The preset specification mapping table is visualized and displayed; The preset specification mapping table is edited in response to a mapping editing instruction of a user. The model specification is managed by using a preset version management table; 4. The method according to any one of claims 1-2, characterized in that, The preset version management table is visualized and displayed; The preset version management table is edited in response to a specification editing instruction of a user. The application comprises the following steps: An application calling request sent by a terminal is acquired, the application calling request comprising a model package of a first model to be verified and a first machine learning platform and a first framework corresponding to the first model; 5. An apparatus for implementing model checking, characterized by The model package of the first model is parsed in a preset manner to acquire a model structure corresponding to the first model; A target model specification corresponding to the first machine learning platform and the first framework is determined according to a preset specification mapping table, wherein the preset specification mapping table comprises a mapping relationship among machine learning platforms, frameworks and model specifications, and the model specification is determined by parsing an original model specification; The model structure corresponding to the first model is verified by using the target model specification to acquire a verification result of the first model, and the verification result is fed back to the terminal; ​ The model checking unit is configured to check the model structure corresponding to the first model by using the target model specification, obtain a checking result of the first model, and feed back the checking result to the terminal. The model checking unit is specifically configured to identify configuration information in a model package of the first model, and update the target model specification according to the configuration information. The model structure corresponding to the first model is checked by using the updated target model specification. The model checking unit is specifically configured to determine an extended specification corresponding to the configuration information according to the configuration information and a preset specification-extended mapping table, and the preset specification-extended mapping table includes a mapping relationship between configuration information and an extended specification. The extended specification is added to the target model specification to generate an updated target model specification.

6. A server, characterized by The computer readable storage medium stores a computer program. The computer readable storage medium stores a computer program. The computer readable storage medium stores a computer program.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program.

8. A computer program product comprising a computer program, characterized in that, ​

Images