Software framework for realizing algorithm deployment and algorithm deployment method
The software framework addresses the inefficiency of third-party model conversion by determining and calling target interfaces, enhancing deployment efficiency and reducing time through a unified interface module, thus improving algorithm deployment across diverse platforms.
Patent Information
- Application Number
- CN202310738867.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-20
- Publication Date
- 2024-12-20
AI Technical Summary
The need to convert third-party platform models during algorithm deployment leads to inefficiencies and prolonged deployment times, which is a significant challenge in the engineering application of algorithms across various platforms.
A software framework and algorithm deployment method that utilizes a calling interface class integrating an encapsulation interface module and a software interface module, which determines and calls target platform and model interfaces based on platform model information, thereby avoiding the need for model conversion and improving deployment capabilities.
This approach enhances algorithm deployment efficiency by eliminating the need for model conversion, improving compatibility and operability across different platforms, and significantly reducing deployment time.
Smart Images
Figure CN119166134A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a software framework for implementing algorithm deployment and an algorithm deployment method. Background Art
[0002] With the continuous development of deep learning technology and the improvement of platform hardware capabilities in recent years, algorithms have penetrated into all walks of life. The engineering application of the entire algorithm is based on the construction and deployment of the algorithm. However, when deploying the algorithm, a lot of time is always wasted on compatibility with third-party platforms, which includes model conversion and interface adaptation. Therefore, how to avoid the need to convert third-party platform models, improve algorithm deployment capabilities, and shorten algorithm deployment time have become technical problems that need to be solved urgently. Summary of the invention
[0003] The main purpose of the present invention is to provide a software framework and an algorithm deployment method for realizing algorithm deployment, aiming to solve the technical problem in the prior art that the third-party platform model needs to convert its algorithm deployment time.
[0004] To achieve the above object, the present invention provides a software framework for implementing algorithm deployment, the software framework comprising a calling interface class;
[0005] The calling interface class integrates a packaging interface module and a software interface module;
[0006] The encapsulation interface module is composed of platform interfaces encapsulated based on platform model information of different platforms, the software interface module contains platform software interfaces of the different platforms, and each platform interface is associated with a corresponding platform software interface;
[0007] The encapsulation interface module is used to determine the target platform interface according to the platform model information in the external trigger instruction, and determine the target platform software interface associated with the target platform interface;
[0008] The encapsulation interface module is also used to call the target platform software interface to respond to the external trigger instruction.
[0009] Optionally, the platform software interface includes multiple model interfaces under the corresponding platform;
[0010] The encapsulation interface module is further used to determine the corresponding target model interface in the target platform software interface according to the platform model information in the external trigger instruction;
[0011] The encapsulation interface module is also used to call the target model interface to call the corresponding model to respond to the external trigger instruction.
[0012] Optionally, the encapsulation interface module is further used to determine a corresponding model format according to the platform model information in the external trigger instruction, and create a model handle corresponding to the platform model information according to the model format;
[0013] The encapsulation interface module is further used to release the model handle after responding to the external trigger instruction.
[0014] Optionally, the encapsulation interface module is further used to determine framework configuration information according to the platform model information in the external trigger instruction, wherein the framework configuration information is the platform input information and the model input information corresponding to the platform model information;
[0015] The encapsulation interface module is also used to call the corresponding model interface in the target platform software interface according to the framework configuration information.
[0016] Optionally, the encapsulation interface module is further used to call the model interface to obtain a model input corresponding to the algorithm model, send the model input to the algorithm model for reasoning, and return the reasoning result as a model output of the algorithm model;
[0017] The encapsulation interface module is also used to determine that the response to the external trigger instruction is completed after returning the model output, and release the model handle.
[0018] Optionally, the encapsulation interface module is further used to obtain the platform interface corresponding to the newly added platform information according to the corresponding newly added platform information in the platform update instruction, and to re-encapsulate the platform interface and save it.
[0019] Optionally, the encapsulation interface module is further used to obtain the model interface corresponding to the model information under different platforms according to the corresponding model information in the model update instruction, and update the corresponding platform software interface according to the model interface.
[0020] In addition, to achieve the above purpose, the present invention provides an algorithm deployment method based on the above software framework, the method comprising the following steps:
[0021] The encapsulation interface module determines the target platform interface according to the platform model information in the external trigger instruction, and determines the target platform software interface associated with the target platform interface;
[0022] The encapsulation interface module calls the target platform software interface to respond to the external trigger instruction.
[0023] Optionally, the step of the encapsulation interface module calling the target platform software interface to respond to the external trigger instruction includes:
[0024] The encapsulation interface module determines the corresponding target model interface in the target platform software interface according to the platform model information in the external trigger instruction;
[0025] The encapsulation interface module calls the target model interface to call the corresponding model to respond to the external trigger instruction.
[0026] Optionally, the method further comprises: the encapsulation interface module determines a corresponding model format according to the platform model information in the external trigger instruction, and creates a model handle corresponding to the platform model information according to the model format;
[0027] The encapsulation interface module releases the model handle after responding to the external trigger instruction.
[0028] The present invention discloses a software framework and algorithm deployment method for realizing algorithm deployment, wherein the software framework includes a calling interface class; the calling interface class is integrated with an encapsulation interface module and a software interface module; the encapsulation interface module is composed of platform interfaces encapsulated based on platform model information of different platforms, the software interface module includes platform software interfaces of different platforms, and each platform interface is associated with a corresponding platform software interface; the encapsulation interface module determines a target platform interface according to platform model information in an external trigger instruction, and determines a target platform software interface associated with the target platform interface; the encapsulation interface module calls the target platform software interface to respond to the external trigger instruction. By calling the encapsulation interface module for algorithm deployment through the above-mentioned software framework, the problem of the need to convert the third-party platform model can be avoided, and the encapsulation interface module is called on different platforms for algorithm deployment, which can improve the algorithm deployment capability and shorten the algorithm deployment time. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 A structural block diagram of an embodiment of a software framework for implementing algorithm deployment of the present invention;
[0030] Figure 2 A schematic diagram of the structure of the external triggering process of the first embodiment of the software framework for implementing algorithm deployment of the present invention;
[0031] Figure 3 A schematic diagram of the structure of the platform update process of the first embodiment of the software framework for implementing algorithm deployment of the present invention;
[0032] Figure 4 This is a flow chart of the first embodiment of the algorithm deployment method of the present invention;
[0033] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0034] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0035] Reference Figure 1 , Figure 1 A structural block diagram of the first embodiment of the software framework for implementing the algorithm deployment of the present invention.
[0036] like Figure 1 As shown, the software framework for implementing algorithm deployment may include: a calling interface class A, in which an encapsulation interface module B and a software interface module C are integrated; the encapsulation interface module B is composed of platform interfaces encapsulated based on platform model information of different platforms, and the software interface module contains platform software interfaces of the different platforms, and each platform interface is associated with the corresponding platform software interface.
[0037] In the embodiment of the present application, the encapsulation interface module B is associated with the software interface module C. By calling the encapsulation interface module B to deploy the algorithm, the algorithm deployment capability is improved, the algorithm deployment time is shortened, and the compatibility and operability are improved. The software interface model includes platform software interfaces for different platforms, and the platform software solution interface is connected to the model core library under the corresponding platform.
[0038] The encapsulation interface module B is used to determine the target platform interface according to the platform model information in the external trigger instruction, and determine the target platform software interface associated with the target platform interface.
[0039] It can be understood that the external trigger instruction can be an algorithm deployment instruction based on the target platform. Algorithm deployment can be the process of running a trained algorithm model in a specific environment, that is, the process of using or calling a pre-configured algorithm model. The external trigger instruction includes but is not limited to: for example, if there are X algorithm and Y algorithm in the target platform, if X algorithm needs to be used, the external trigger instruction initiated at this time can be to deploy X algorithm in the target platform.
[0040] It should be understood that the platform model information can be platform information and / or model information. The platform information can be different operating systems, processor architectures, programming languages, etc. The model information can be the name, version number, input data format of the algorithm model, input data volume, output result format and result volume of the algorithm model, etc.
[0041] As an implementation manner, the encapsulation interface module B determines the target platform interface according to the platform information in the external trigger instruction, and determines the target platform software interface associated with the target platform interface according to the model information in the external trigger instruction.
[0042] Reference Figure 2The schematic diagram of the structure of the external triggering process of the first embodiment of the software framework for implementing algorithm deployment of the present invention, in which the calling interface class A of the framework includes interfaces such as platform and model selection and configuration, and interfaces for accessing the core libraries of models such as C, D, and E; the encapsulation interface module B of the framework is a subclass interface after secondary encapsulation of each platform interface (C interface, D interface, and e interface) with the help of the core software libraries of platforms such as C, D, and E. The software interfaces (C interface, D interface, E interface, etc.) to be called subsequently can be determined based on the configured platform and model format. The C interface, D interface, E interface, etc. are software interfaces for calling platforms such as C, D, and E, which respectively contain model interfaces of multiple models supported by each platform; C, D, E, etc. are the calling model core libraries under each platform. Among them, process 0 is to set the platform information and model information that need to be deployed, which is set through the encapsulation interface module B. When the external trigger process 1 calls the framework interface to create a model handle, the encapsulation interface module B identifies the platform and model information set by process 0 to determine which specific interface to access subsequently, and then accesses the interfaces of C, D, and E through the corresponding process (2, 3, 4, or 5), thereby realizing a framework that deploys a random combination of multiple platforms and multiple model formats.
[0043] The encapsulation interface module B is further used to call the target platform software interface to respond to the external trigger instruction.
[0044] As an implementation method, the encapsulation interface module B calls the target platform software interface to respond to the external trigger instruction through the model core library of the target platform.
[0045] Furthermore, in order to adapt to algorithm deployment in more platforms, the encapsulation interface module is also used to obtain the platform interface corresponding to the newly added platform information according to the corresponding newly added platform information in the platform update instruction, and save the platform interface after secondary encapsulation.
[0046] It should be noted that the platform update instruction may be an instruction for integrating a new platform into a software framework, and the platform interface corresponding to the newly added platform information is obtained according to the newly added platform information corresponding to the platform update instruction, and the platform interface is repackaged and saved.
[0047] Reference Figure 3 This is a schematic diagram of the structure of the platform update process of the first embodiment of the software framework for implementing algorithm deployment in the present invention. Figure 3 The F platform is newly added, and the platform interface f is re-encapsulated in the encapsulation interface module B. Based on the platform interface f, the F platform software interface is additionally called to call the model core library F under the platform.
[0048] It can be understood that the encapsulation interface module obtains the platform interface corresponding to the newly added platform information according to the corresponding newly added platform information in the platform update instruction, re-encapsulates the platform interface and saves it, and then additionally calls the corresponding platform software interface according to the saved platform interface.
[0049] Furthermore, in order to adapt to the deployment of more algorithm models in the platform, the encapsulation interface module is also used to obtain the model interface corresponding to the model information under different platforms according to the corresponding model information in the model update instruction, and update the corresponding platform software interface according to the model interface.
[0050] It should be noted that the model update instruction may be an instruction for integrating new models under different platforms in the software framework. The encapsulation interface module obtains the model interface corresponding to the model information under different platforms according to the model information corresponding to the model update instruction, and updates the corresponding platform software interface according to the model interface.
[0051] This embodiment proposes a software framework for implementing algorithm deployment, including: a calling interface class; an encapsulation interface module and a software interface module integrated in the calling interface class; the encapsulation interface module is composed of platform interfaces encapsulated based on platform model information of different platforms, the software interface module contains platform software interfaces of different platforms, and each platform interface is associated with a corresponding platform software interface; the encapsulation interface module determines the target platform interface according to the platform model information in the external trigger instruction, and determines the target platform software interface associated with the target platform interface; the encapsulation interface module calls the target platform software interface to respond to the external trigger instruction. By calling the encapsulation interface module for algorithm deployment through the above-mentioned software framework, the problem of the need to convert the third-party platform model can be avoided. Calling the encapsulation interface module for algorithm deployment on different platforms can improve the algorithm deployment capability and shorten the algorithm deployment time.
[0052] Based on the above Figure 1 The described embodiment proposes a second embodiment of the software framework for implementing algorithm deployment of the present invention.
[0053] In this embodiment, the platform software interface includes a variety of model interfaces under the corresponding platform.
[0054] The encapsulation interface module B is further used to determine the corresponding target model interface in the target platform software interface according to the platform model information in the external trigger instruction.
[0055] It should be noted that the target model interface may be a model interface selected in the target software interface according to the model information. The corresponding target model interface in the target platform software interface is determined according to the model information in the external trigger instruction.
[0056] As an implementation method, encapsulation interface module B can identify platform model information in external trigger instructions and obtain model information under the target platform. The model information includes model input format, output format, data type, calculation method and performance requirements, and determines the corresponding target model interface in the target platform software interface based on the model information.
[0057] The encapsulation interface module B is also used to call the target model interface to call the corresponding model to respond to the external trigger instruction.
[0058] It should be noted that the method of calling the target model interface to call the corresponding model can be that the encapsulation interface calls the target model interface to access the model core library, and the corresponding model is called based on the model core library to respond to the external trigger instruction.
[0059] It is understood that the model core library under the platform can refer to the optimized and tested core code library developed for the target algorithm model on the platform. These code libraries can include information such as the implementation details, parameter configuration, training data sets, etc. of the target algorithm model, and can be directly called to accelerate the deployment of the algorithm model.
[0060] Furthermore, in order to accelerate the deployment capability of the algorithm model, the encapsulation interface module B is also used to determine the corresponding model format according to the platform model information in the external trigger instruction, and create a model handle corresponding to the platform model information according to the model format.
[0061] It can be understood that the method of determining the corresponding model format according to the platform model information in the external trigger instruction can be to obtain the name information and content information of the model according to the platform model information, and determine the model format of the corresponding model (such as Joint, Pro, etc.) according to the name information and content information.
[0062] It should be understood that the model handle can be the context information of the model, and the model handle needs to be accessed when loading the model. When deploying the algorithm, different function interfaces need to be used to create the required model handle according to different model formats.
[0063] The encapsulation interface module B is further used to release the model handle after responding to the external trigger instruction.
[0064] It is understandable that the way to respond to external trigger instructions can be to first load the software framework, compile and generate a dynamic library or static library by loading the software framework for subsequent software interface calls. After initializing the software framework SDK, obtain the model information, determine the model format type based on the model information, and create the required model handle based on the model format type. Enter the platform information and model input information, set the function interface to obtain the platform information and model input information, and set it to the framework configuration information. After setting the framework configuration information, perform the reasoning work corresponding to the model. Release the model handle after completing the reasoning work to implement algorithm deployment.
[0065] In this embodiment, the platform software interface includes multiple model interfaces under the corresponding platform; the encapsulation interface module is also used to determine the corresponding target model interface in the target platform software interface according to the platform model information in the external trigger instruction; the encapsulation interface module is also used to call the target model interface to call the corresponding model to respond to the external trigger instruction. In the above manner, the encapsulation interface module is called to determine the corresponding target model interface in the target platform software according to the platform model information, and the encapsulation interface module can be called to deploy the algorithm on different platforms, thereby improving the ability of algorithm deployment, shortening the time of algorithm deployment, and improving compatibility and operability.
[0066] Based on the above Figure 1 The described embodiment proposes a third embodiment of the software framework for implementing algorithm deployment of the present invention.
[0067] In this embodiment, in order to accurately call the corresponding model interface, the encapsulation interface module B is also used to determine the framework configuration information according to the platform model information in the external trigger instruction, and the framework configuration information is the platform input information and model input information corresponding to the platform model information.
[0068] It can be understood that the platform information and model input information are determined according to the platform model information in the external trigger instruction, the model input information includes the model input size and the model input format (including CV_8UC1, CV_8UC2 and CV_8UC3, etc.), and the function interface is set to set the above platform information and model input information to the framework configuration information.
[0069] The encapsulation interface module B is further used to call the corresponding model interface in the target platform software interface according to the framework configuration information.
[0070] It should be understood that the way in which the encapsulation interface module B calls the corresponding model interface in the target platform software interface according to the set framework configuration can be that the encapsulation interface module B determines the model name and version number to be used according to the framework configuration information, searches for the model interface corresponding to the model name and version number in the software interface of the target platform, calls the interface and passes in the corresponding parameters and data.
[0071] Furthermore, the encapsulation interface module B is also used to call the model interface to obtain the model input corresponding to the algorithm model, send the model input to the algorithm model for reasoning, and return the reasoning result as the model output of the algorithm model.
[0072] It should be understood that reasoning can be the process of an algorithm performing corresponding reasoning work, analyzing, processing and inferring data according to predefined rules and logic after given input data, so as to draw conclusions or predict results. After the reasoning is completed, the reasoning result corresponding to the algorithm is returned as the model output of the algorithm model.
[0073] The encapsulation interface module B is further used to determine whether the response to the external trigger instruction is completed after returning the model output, and release the model handle.
[0074] It is understandable that after the inference is completed, the inference result corresponding to the algorithm is returned as the model output of the algorithm model, and after the model output is returned, it is determined that the response to the external trigger instruction is completed, and the model handle is released. Releasing the model handle realizes the algorithm deployment.
[0075] In this embodiment, the encapsulation interface module is also used to determine the framework configuration information according to the platform model information in the external trigger instruction, and the framework configuration information is the platform input information and model input information corresponding to the platform model information; the encapsulation interface module is also used to call the corresponding model interface in the target platform software interface according to the framework configuration information; the encapsulation interface module is also used to call the model interface to obtain the model input corresponding to the algorithm model, send the model input to the algorithm model for reasoning, and return the reasoning result as the model output of the algorithm model; the encapsulation interface module is also used to determine the completion of the response to the external trigger instruction after returning the model output, and release the model handle. In the above manner, the framework configuration information is determined and the reasoning work is performed, which can improve the ability of algorithm deployment and shorten the time of algorithm deployment.
[0076] Reference Figure 4 The software framework for implementing algorithm deployment of the present invention provides an algorithm deployment method. Figure 4This is a flow chart of the first embodiment of the algorithm deployment method of the present invention. The software framework for implementing the algorithm deployment may include: a calling interface class A, in which an encapsulation interface module B and a software interface module C are integrated; the encapsulation interface module B is composed of platform interfaces encapsulated based on platform model information of different platforms, and the software interface module contains platform software interfaces of the different platforms, and each platform interface is associated with the corresponding platform software interface.
[0077] In the embodiment of the present application, the encapsulation interface module B is associated with the software interface module C. By calling the encapsulation interface module B to deploy the algorithm, the algorithm deployment capability is improved, the algorithm deployment time is shortened, and the compatibility and operability are improved. The software interface model includes platform software interfaces for different platforms, and the platform software solution interface is connected to the model core library under the corresponding platform.
[0078] Step S10: Encapsulation interface module B: Determine the target platform interface according to the platform model information in the external trigger instruction, and determine the target platform software interface associated with the target platform interface.
[0079] It can be understood that the external trigger instruction can be an algorithm deployment instruction based on the target platform. Algorithm deployment can be the process of running a trained algorithm model in a specific environment, that is, the process of using or calling a pre-configured algorithm model. The external trigger instruction includes but is not limited to: for example, if there are X algorithm and Y algorithm in the target platform, if X algorithm needs to be used, the external trigger instruction initiated at this time can be to deploy X algorithm in the target platform.
[0080] It should be understood that the platform model information can be platform information and / or model information. The platform information can be different operating systems, processor architectures, programming languages, etc. The model information can be the name, version number, input data format of the algorithm model, input data volume, output result format and result volume of the algorithm model, etc.
[0081] As an implementation manner, a target platform interface is determined according to platform information in an external trigger instruction, and a target platform software interface associated with the target platform interface is determined according to model information in the external trigger instruction.
[0082] Reference Figure 2The schematic diagram of the structure of the external triggering process of the first embodiment of the software framework for implementing algorithm deployment of the present invention, in which the calling interface class A of the framework includes interfaces such as platform and model selection and configuration, and interfaces for accessing the core libraries of models such as C, D, and E; the encapsulation interface module B of the framework is a subclass interface after secondary encapsulation of each platform interface (C interface, D interface, and e interface) with the help of the core software libraries of platforms such as C, D, and E. The software interfaces (C interface, D interface, E interface, etc.) to be called subsequently can be determined based on the configured platform and model format. The C interface, D interface, E interface, etc. are software interfaces for calling platforms such as C, D, and E, which respectively contain model interfaces of multiple models supported by each platform; C, D, E, etc. are the calling model core libraries under each platform. Among them, process 0 is to set the platform information and model information that need to be deployed, which is set through the encapsulation interface module B. When the external trigger process 1 calls the framework interface to create a model handle, the encapsulation interface module B identifies the platform and model information set by process 0 to determine which specific interface to access subsequently, and then accesses the interfaces of C, D, and E through the corresponding process (2, 3, 4, or 5), thereby realizing a framework that deploys a random combination of multiple platforms and multiple model formats.
[0083] Step S20: the encapsulation interface module B calls the target platform software interface to respond to the external trigger instruction.
[0084] As an implementation method, the encapsulation interface module B calls the target platform software interface to respond to the external trigger instruction through the model core library of the target platform.
[0085] Furthermore, in order to adapt to algorithm deployment in more platforms, the algorithm deployment method also includes: obtaining a platform interface corresponding to the newly added platform information according to the corresponding newly added platform information in the platform update instruction, and re-packaging the platform interface for storage.
[0086] It should be noted that the platform update instruction may be an instruction for integrating a new platform into a software framework, and the platform interface corresponding to the newly added platform information is obtained according to the newly added platform information corresponding to the platform update instruction, and the platform interface is repackaged and saved.
[0087] Reference Figure 3 This is a schematic diagram of the structure of the platform update process of the first embodiment of the software framework for implementing algorithm deployment in the present invention. Figure 3 The F platform is newly added, and the platform interface f is re-encapsulated in the encapsulation interface module B. Based on the platform interface f, the F platform software interface is additionally called to call the model core library F under the platform.
[0088] It can be understood that the encapsulation interface module obtains the platform interface corresponding to the newly added platform information according to the corresponding newly added platform information in the platform update instruction, re-encapsulates the platform interface and saves it, and then additionally calls the corresponding platform software interface according to the saved platform interface.
[0089] Furthermore, in order to adapt to the deployment of more algorithm models in the platform, the algorithm deployment method also includes: obtaining the model interface corresponding to the model information under different platforms according to the corresponding model information in the model update instruction, and updating the corresponding platform software interface according to the model interface.
[0090] It should be noted that the model update instruction may be an instruction for integrating new models under different platforms in the software framework. The encapsulation interface module obtains the model interface corresponding to the model information under different platforms according to the model information corresponding to the model update instruction, and updates the corresponding platform software interface according to the model interface.
[0091] This embodiment proposes an algorithm deployment method, the method comprising: the encapsulation interface module determines the target platform interface according to the platform model information in the external trigger instruction, and determines the target platform software interface associated with the target platform interface; the encapsulation interface module calls the target platform software interface to respond to the external trigger instruction. Calling the encapsulation interface module for algorithm deployment through the above-mentioned software framework can avoid the problem of the need to convert the third-party platform model, and calling the encapsulation interface module for algorithm deployment on different platforms can improve the algorithm deployment capability and shorten the algorithm deployment time.
[0092] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.
[0093] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.
[0094] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A software framework for implementing algorithm deployment, characterized in that: The software framework includes a calling interface class; The calling interface class integrates a packaging interface module and a software interface module; The encapsulation interface module is composed of platform interfaces encapsulated based on platform model information of different platforms, the software interface module contains platform software interfaces of the different platforms, and each platform interface is associated with a corresponding platform software interface; The encapsulation interface module is used to determine the target platform interface according to the platform model information in the external trigger instruction, and determine the target platform software interface associated with the target platform interface; The encapsulation interface module is also used to call the target platform software interface to respond to the external trigger instruction.
2. The software architecture for implementing algorithm deployment as claimed in claim 1, characterized in that: The platform software interface includes multiple model interfaces under the corresponding platform; The encapsulation interface module is further used to determine the corresponding target model interface in the target platform software interface according to the platform model information in the external trigger instruction; The encapsulation interface module is also used to call the target model interface to call the corresponding model to respond to the external trigger instruction.
3. The software architecture for implementing algorithm deployment as claimed in claim 2, characterized in that: The encapsulation interface module is further used to determine the corresponding model format according to the platform model information in the external trigger instruction, and create a model handle corresponding to the platform model information according to the model format; The encapsulation interface module is further used to release the model handle after responding to the external trigger instruction.
4. The software architecture for implementing algorithm deployment as claimed in claim 3, characterized in that: The encapsulation interface module is further used to determine framework configuration information according to the platform model information in the external trigger instruction, wherein the framework configuration information is the platform input information and the model input information corresponding to the platform model information; The encapsulation interface module is also used to call the corresponding model interface in the target platform software interface according to the framework configuration information.
5. The software architecture for implementing algorithm deployment as claimed in claim 4, characterized in that: The encapsulation interface module is further used to call the model interface to obtain the model input corresponding to the algorithm model, send the model input to the algorithm model for reasoning, and return the reasoning result as the model output of the algorithm model; The encapsulation interface module is further used to determine whether the response to the external trigger instruction is completed after returning the model output, and release the model handle.
6. A software framework for implementing algorithm deployment as claimed in any one of claims 1 to 5, characterized in that: The encapsulation interface module is further used to obtain the platform interface corresponding to the newly added platform information according to the corresponding newly added platform information in the platform update instruction, and to save the platform interface after re-encapsulation.
7. A software framework for implementing algorithm deployment as claimed in any one of claims 1 to 5, characterized in that: The encapsulation interface module is also used to obtain the model interface corresponding to the model information under different platforms according to the corresponding model information in the model update instruction, and update the corresponding platform software interface according to the model interface.
8. An algorithm deployment method based on a software framework for implementing algorithm deployment according to any one of claims 1 to 7, characterized in that: The method comprises: The encapsulation interface module determines the target platform interface according to the platform model information in the external trigger instruction, and determines the target platform software interface associated with the target platform interface; The encapsulation interface module calls the target platform software interface to respond to the external trigger instruction.
9. The method according to claim 8, characterized in that The step of the encapsulation interface module calling the target platform software interface to respond to the external trigger instruction includes: The encapsulation interface module determines the corresponding target model interface in the target platform software interface according to the platform model information in the external trigger instruction; The encapsulation interface module calls the target model interface to call the corresponding model to respond to the external trigger instruction.
10. The method according to claim 9, characterized in that The method further comprises: The encapsulation interface module determines a corresponding model format according to the platform model information in the external trigger instruction, and creates a model handle corresponding to the platform model information according to the model format; The encapsulation interface module releases the model handle after responding to the external trigger instruction.