CN115719026A - System and method for artificial intelligence middling station and readable medium - Google Patents
System and method for artificial intelligence middling station and readable medium Download PDFInfo
- Publication number
- CN115719026A CN115719026A CN202110970350.0A CN202110970350A CN115719026A CN 115719026 A CN115719026 A CN 115719026A CN 202110970350 A CN202110970350 A CN 202110970350A CN 115719026 A CN115719026 A CN 115719026A
- Authority
- CN
- China
- Prior art keywords
- model
- module
- artificial intelligence
- developer
- task
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a system suitable for an artificial intelligence middle station, which comprises: a model engine module: the model engine manages model tasks and is in butt joint with a container cluster of heterogeneous computing resources; a model factory module: providing a model development assembly line, wherein the assembly line is used for guiding a model developer to prepare contents required by model training and reasoning and applying for executing tasks to a model engine module; a model market module: and the model is cooperated with an AIBox module of a docking station to form a cloud edge cooperation scheme, so that a model developer is helped to develop a developed model on the market, and the trained model is released to a model market module to be shared and displayed, or other people are selected to develop the model. The method takes model research and development as a core, forms a model engine module, a model factory module and a model market module, realizes flexible scheduling of AI resources and production of a production line type model, and is used for supporting research and development of enterprise machine learning and deep learning models.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a system and a method suitable for an artificial intelligence middle station and a readable medium.
Background
At present, AI model research and development need to carry out a large amount of model development, debugging and deployment, and research and development tasks need to flexibly use computational resources such as a graphics processor. Meanwhile, the model research and development process is complex, and a simple tool cannot well assist a non-AI professional developer to finish model training.
Through retrieval, patent document CN112347145A discloses an AI middlebox system, which comprises a basic database, an information screening module, a shared data storage module, a feature extraction module, a deep learning module, a model database, a result database, and a processor. The invention can break an information isolated island, the data in each system can be shared, the assistance between each system is promoted, and the data forms a benign closed loop between the data platform and the service system. The defects of the prior art are that although the work required by AI model research and development is provided, a standardized research and development flow is not formed, and the resource of the Chinese traditional computing power cannot be fully utilized.
Patent document CN110941421A discloses a development machine learning apparatus and a method of using the same, including: the machine learning platform is a platform for mining value information from mass data based on Huawei fusion instrumentation HD distributed storage and parallel computing technology; the deep learning platform is an enterprise-level deep learning modeling platform, and can enable client algorithm developers to efficiently manage data sets, develop algorithm codes, evaluate models and predict service release experience, and reduce deep learning modeling thresholds; and the reasoning platform is mainly used for completing multi-algorithm unified management and task containerization heterogeneous resource unified scheduling, and assisting the clients to realize cluster computing power sharing and reduce the operation and maintenance cost of the AI system. Although the prior art is assisted by a deep learning platform, a non-AI professional developer still cannot be assisted to complete model patrolling and pressing, and meanwhile, a standardized research and development process cannot be formed, and a mesobench computing resource cannot be fully utilized.
Therefore, there is a need to develop and design a method and system that can assist the AI engineer in research and development.
Disclosure of Invention
In view of the defects in the prior art, the invention aims to provide a system, a method and a readable medium suitable for an artificial intelligence middle desk, which are used for supporting machine learning and deep learning model research and development of enterprises.
The system suitable for the artificial intelligence middle station provided by the invention comprises the following components:
a model engine module: the model engine manages model tasks and is connected with a container cluster of heterogeneous computing resources;
a model factory module: providing a model development assembly line, wherein the assembly line is used for guiding a model developer to prepare contents required by model training and reasoning and applying for executing tasks to a model engine module;
a model market module: and forming a cloud edge cooperation scheme by cooperating with an AIBox module of the docking station. The model development device helps a model developer to develop a finished model on the market, and the trained model is released to a model market module for sharing and displaying, or the model is selected and purchased by others for model development.
Preferably, the model engine module includes task type, required computational resources, task dependent mirroring and task specific parameter information.
Preferably, the model engine module selects a host node with available computing power from the cluster nodes through a scheduling mechanism, pulls a mirror image package required by a task and deploys an application on the node, allows a model developer to train and deduce the task, mounts a disk into the training task according to rules, and persistently stores training results and running log information produced by the task.
Preferably, the container cluster interfaced by the model engine module can map the host node port with the container application port, allowing a user to access from the outside for viewing the training task interface or invoking inference tasks for scene recognition and prediction.
Preferably, in the model factory module, a model developer is required to upload a data set required by the model on a model development line, and labeling is performed according to the data type, and the labeling part is based on the data set type and supports multiple labeling modes according to different model scenes.
Preferably, the model factory module selects an algorithm required by model training, selects from a platform preset algorithm according to a required recognition scene, or automatically uploads a code for model training or reasoning to a platform code library to be made into a mirror image;
a developer uploads a model source code according to a code base address, a mirror image script is compiled according to a code needed environment and used for establishing a task of a model engine, and a model reasoning service is deployed on a cloud by utilizing a middle platform resource, so that the model can recognize the content aiming at a needed scene.
Preferably, the model market module displays models which are developed, applied and shared by all developers on the artificial intelligence middle desk and pass the auditing, guides the model developers to publish the models, and selects the models from the developed models to apply for sharing and displaying.
Preferably, the AI Box management module manages and controls an edge inference product AI Box associated with an artificial intelligence middle desk, a developer purchases the AI Box and registers in the middle desk, collects AI Box information and binds with a middle desk tenant, and the artificial intelligence middle desk generates a license file based on the AI Box information and issues the license to an AI Box device.
According to the design method suitable for the artificial intelligence middlings provided by the invention, the system suitable for the artificial intelligence middlings is adopted for design, and the design method comprises the following steps:
step S1: the model engine manages model tasks and is in butt joint with a container cluster of heterogeneous computing resources;
step S2: providing a model development assembly line, wherein the assembly line is used for guiding a model developer to prepare contents required by model training and reasoning and applying for executing tasks to a model engine module;
and step S3: and the model developer is helped to develop the finished model on the market, and the trained model is released to the model market module for sharing and displaying.
According to the present invention, there is provided a computer readable storage medium storing a computer program, which when executed by a processor implements the steps of the above-described design method for an artificial intelligence middlebox.
Compared with the prior art, the invention has the following beneficial effects:
1. according to the invention, the AI application is subjected to standardized packaging by adopting an artificial intelligence middle platform model engine and adopting a container technology, so that the problem of complex environment construction in AI research and development is solved.
2. According to the invention, the model research and development process is simplified through an artificial intelligence model factory, and the problems of complex research and development process, higher algorithm threshold and difficult data annotation in AI research and development are solved.
3. According to the invention, the business units can share and exchange AI capabilities through the model market, AI research and development are formed to create an ecology together, and the problem of non-intercommunication of the AI capabilities in an enterprise is solved.
4. The invention realizes data issuing to the edge terminal equipment under the condition of ensuring data safety through the AI Box management module to form an AI cloud edge cooperation scheme
5. The invention realizes the centralized distribution management of the training resources by the middlebox on the resources in the enterprise, solves the problems of difficult selection of computing resources and scarce computing resources required by AI in the enterprise, can be applied for use on the cloud when needed, and actively releases the resources to realize flexible scheduling when idle.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
FIG. 1 is a block diagram of a system suitable for use in an artificial intelligence center station in accordance with the present invention;
FIG. 2 is a schematic diagram of a task life cycle of a model engine in a system suitable for an artificial intelligence middlebox in the present invention;
FIG. 3 is a schematic diagram of a simulation plant operating process in the system for an artificial intelligence middlebox of the present invention;
fig. 4 is a schematic diagram of an AI Box license issuing process in the system suitable for the artificial intelligence middlebox in the present invention.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the invention.
As shown in fig. 1-4, the present invention provides a system suitable for an artificial intelligence middlebox, which takes model development as a core, and includes a model engine module, a model factory module and a model market module, so as to implement flexible scheduling of AI resources and pipelined model production. The method is used for supporting machine learning and deep learning model research and development of enterprises.
A model engine module: the model engine manages model tasks and interfaces a container cluster with heterogeneous computing resources such as a GPU. After applying for creating a task to a model engine, the engine selects a host node with available computing power from cluster nodes through a scheduling mechanism according to information such as task types, required computing power resources, task dependent mirror images and task specific parameters, pulls a mirror image package required by the task and deploys the application on the nodes, allows model training and reasoning tasks required by a model developer, mounts a disk into a training task according to rules, and persistently stores information such as training results and running logs generated by the task. During task execution, the container cluster can map the host node port and the container application port, and allows a user to access the container cluster from the outside so as to view a training task interface or call an inference task to perform scene recognition and prediction. Meanwhile, the middle platform is in butt joint with the task, so that logs or model training process changes in the task running process can be checked in real time, and parameters such as a Loss curve, learning rate changes and the like can be checked. After the task is executed, the task application sends a completion receipt to the model engine, the engine stores task products, process records and process logs, uploads the task products, the process records and the process logs to the object storage server for backup so as to check the logs and use the model products, closes the task and automatically recovers computing resources.
Model factory module: a model development pipeline is provided. The pipeline is used for guiding a model developer to prepare contents required by model training and reasoning and applying for executing tasks to the model engine module. In the assembly line, a model developer is required to upload a data set required by the model firstly, and labeling is carried out according to the data type. The pictures are uploaded through the middlebox and are backed up in the object storage server. The labeling part is based on the data set type, and can support various labeling modes according to different model scenes, such as the capabilities of label classification, target frame selection, character labeling and the like on pictures. Allowing the model developer to speak the annotation work to assign to any member of the team; and secondly, selecting an algorithm required by model training, selecting from a platform preset algorithm according to a required recognition scene, or automatically uploading a code for model training or reasoning to a platform code library to manufacture a mirror image. When the code needs to be uploaded, the middle station can automatically create a code base according to the algorithm information. And uploading the model source code of the developer according to the address of the code base, and compiling the mirror image script according to the environment required by the code. After uploading, a middle platform developer can manually trigger mirror image making, the system pulls an environment required by a source code according to a script and the source code, compiles an application into a task mirror image, and pushes the task mirror image to a middle platform mirror image warehouse, so that the model engine can conveniently use when establishing a task; after the algorithm is determined, a model developer is required to configure parameters for the training, a factory module supports typical and necessary parameter setting such as iteration rounds, batch training numbers, required labeling formats, resource templates and the like, and the developer is also supported to introduce self-defined key value pair parameters according to own code requirements and to introduce the parameters into model training task application as environment variables. In the parameter setting process, the classification of the data sets can be set, and the data sets are distributed into a training set, a verification set and a test set which are required by model training according to a certain proportion; then applying for creating and executing a training task to a model engine, and performing model training by using GPU resources; informing a model developer after the engine finishes executing the task, allowing the model developer to enter a next release link, assisting the model developer to select a reasoning template from the model, applying for creating a reasoning task to the model engine, and deploying a model reasoning service on the cloud by using the middlebox resources so that the model can identify the content aiming at the required scene. After training is finished, the model can also be applied to be released to a model market for sharing and displaying.
A model market module: and providing an AI Box management module which is used for the market module and the cloud edge to cooperate. The model market is used for helping a model developer to go online to develop a finished model and choose to purchase other people to develop the model. And all developers on the platform in the model market show complete research and development, apply for sharing and pass the audited model. And guiding a model developer to publish the model in the market, wherein the developer can select the model from the developed model to apply for sharing and displaying. When the application is carried out, model introduction materials need to be filled in, and after the information such as model names, application scenes, identification examples, performance description, model interface request modes and the like is improved, the application can be submitted to a market manager. The market manager can check and track the model information to judge whether the application is passed. The application can be displayed on the market in a public way after passing. The developer can also search the developed models of other people in the market, check the display data, apply for trial use and contact with the model owner for deployment and use. And the AI Box management module manages and controls an edge terminal inference product AI Box associated with the middle station. A developer can purchase an AI Box and register in the middle platform, and collects Box information to be bound with the middle platform tenants. The middle station generates a license file based on the AI Box information and issues a license to the AI Box device. When the AI Box needs to acquire the middlebox data, the AI Box needs to carry the information in the license. The central station will compare the license information, and issue the corresponding data if they are consistent, otherwise, stop.
The invention also provides a method suitable for the artificial intelligence middling station, which is designed by adopting the system suitable for the artificial intelligence middling station and comprises the following steps:
step S1: the model engine manages model tasks and is in butt joint with a container cluster of heterogeneous computing resources;
step S2: providing a model development assembly line, wherein the assembly line is used for guiding a model developer to prepare contents required by model training and reasoning and applying for executing tasks to a model engine module;
and step S3: and the model developer is helped to develop the finished model on the market, and the trained model is released to the model market module for sharing and displaying.
The present invention further provides a computer-readable storage medium storing a computer program, which when executed by a processor implements the steps of the above-mentioned design method for an artificial intelligence middlebox.
It is well within the knowledge of a person skilled in the art to implement the system and its various devices, modules, units provided by the present invention in a purely computer readable program code means that the same functionality can be implemented by logically programming method steps in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system and various devices, modules and units thereof provided by the present invention can be regarded as a hardware component, and the devices, modules and units included therein for implementing various functions can also be regarded as structures within the hardware component; means, modules, units for performing the various functions may also be regarded as structures within both software modules and hardware components for performing the method.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.
Claims (10)
1. A system adapted for use in an artificial intelligence midstation, comprising:
a model engine module: the model engine manages model tasks and is in butt joint with a container cluster of heterogeneous computing resources;
model factory module: providing a model development assembly line, wherein the assembly line is used for guiding a model developer to prepare contents required by model training and reasoning and applying for executing tasks to a model engine module;
a model market module: and the model is cooperated with an AIBox module of a docking station to form a cloud edge cooperation scheme, so that a model developer is helped to develop a developed model on the market, and the trained model is released to a model market module to be shared and displayed, or other people are selected to develop the model.
2. The system for an artificial intelligence midstation of claim 1, wherein the model engine module includes task type, required computational resources, task dependent mirroring, and task specific parametric information.
3. The system suitable for the artificial intelligence middlebox of claim 2, wherein the model engine module selects a host node with available computing power from cluster nodes through a scheduling mechanism, pulls a mirror image package required by a task and deploys an application on the node, allows a model developer to train and infer the task required by the model developer, mounts a disk into the training task according to rules, and persistently stores training results and running log information produced by the task.
4. The system for an artificial intelligence midstation of claim 1, wherein the container cluster interfaced by the model engine module maps host node ports with container application ports allowing external access by a user to view training task interfaces or invoke inference tasks for scene recognition and prediction.
5. The system suitable for the artificial intelligence middlebox of claim 1, wherein a model developer is required to upload a data set required by a model in the model factory module on a model development line, and labeling is performed according to the data type, and the labeling part is based on the data set type and supports multiple labeling modes according to different model scenes.
6. The system suitable for the artificial intelligence midstation as claimed in claim 1, wherein the model factory module selects an algorithm required by model training, selects from a platform preset algorithm according to a required recognition scene, or automatically uploads a code for model training or reasoning to a platform code base to be made into a mirror image;
a developer uploads a model source code according to a code base address, a mirror image script is compiled according to a code needed environment and used for establishing a task of a model engine, and a model reasoning service is deployed on a cloud by utilizing a middle platform resource, so that the model can recognize the content aiming at a needed scene.
7. The system for an artificial intelligence midstation as claimed in claim 1, wherein the model market module displays the models of all developers on the artificial intelligence midstation completing research and development, applying for sharing and passing auditing, the model market module guides the model developers to release the models, and the developers select the model application sharing display from the developed models.
8. The system suitable for the artificial intelligence middlebox according to claim 1, wherein the AI Box management module manages and controls an edge-side inference product AI Box associated with the artificial intelligence middlebox, a developer purchases the AI Box and registers in the middlebox, AI Box information is collected and bound with a middlebox tenant, and the artificial intelligence middlebox generates a license file based on the AI Box information and issues a license to an AI Box device.
9. A method for an artificial intelligence center, which is designed by using the system for an artificial intelligence center of any one of claims 1 to 8, comprising the steps of:
step S1: the model engine manages model tasks and is connected with a container cluster of heterogeneous computing resources;
step S2: providing a model development assembly line, wherein the assembly line is used for guiding a model developer to prepare contents required by model training and reasoning and applying for executing tasks to a model engine module;
and step S3: and the model developer is helped to go on the line to research and develop the finished model, and the trained model is released to the model market module to be shared and displayed.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method steps of claim 9, which are adapted for designing an artificial intelligence middesk.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110970350.0A CN115719026A (en) | 2021-08-23 | 2021-08-23 | System and method for artificial intelligence middling station and readable medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110970350.0A CN115719026A (en) | 2021-08-23 | 2021-08-23 | System and method for artificial intelligence middling station and readable medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115719026A true CN115719026A (en) | 2023-02-28 |
Family
ID=85253326
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110970350.0A Pending CN115719026A (en) | 2021-08-23 | 2021-08-23 | System and method for artificial intelligence middling station and readable medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115719026A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116560857A (en) * | 2023-06-29 | 2023-08-08 | 北京轻松筹信息技术有限公司 | AGI platform call management method and device, storage medium and electronic equipment |
| CN117524445A (en) * | 2023-10-19 | 2024-02-06 | 广州中康数字科技有限公司 | Medical field artificial intelligence engineering platform based on micro-service and containerization technology |
-
2021
- 2021-08-23 CN CN202110970350.0A patent/CN115719026A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116560857A (en) * | 2023-06-29 | 2023-08-08 | 北京轻松筹信息技术有限公司 | AGI platform call management method and device, storage medium and electronic equipment |
| CN116560857B (en) * | 2023-06-29 | 2023-09-22 | 北京轻松筹信息技术有限公司 | AGI platform call management method and device, storage medium and electronic equipment |
| CN117524445A (en) * | 2023-10-19 | 2024-02-06 | 广州中康数字科技有限公司 | Medical field artificial intelligence engineering platform based on micro-service and containerization technology |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107784152A (en) | Include the simulation of multiple simulators | |
| Riva et al. | Experiences with software product family evolution | |
| CN115719026A (en) | System and method for artificial intelligence middling station and readable medium | |
| Chen et al. | Strategic prototyping for developing big data systems | |
| JP2007305129A (en) | Business process federated repository | |
| CN111427549A (en) | Artificial intelligence reinforcement learning service platform | |
| Shah et al. | Exploring knowledge engineering strategies in designing and modelling a road traffic accident management domain | |
| Vieira et al. | Students' experimentation strategies in design: Is process data enough? | |
| Nordgren | Steps for proper simulation project management | |
| CN110489093A (en) | A kind of lasting building method, device, terminal and storage medium towards AI application | |
| US20230011461A1 (en) | Method and system for generating engineering diagrams in an engineering system | |
| CN109710303A (en) | The multi version parallel developing method and system of interactive voice product | |
| Bock et al. | Status report on automotive software development | |
| Ulrich et al. | Operator timing of task level primitives for use in computation-based human reliability analysis | |
| WO2021140176A1 (en) | Method and system for managing engineering workflow in a cloud computing environment | |
| KR102355787B1 (en) | Methods to gather screen requirements and to manage processes for the software development including web based services and mobile services | |
| CN114610368B (en) | Platform tool chain integration and maintenance method, system, equipment and medium | |
| CN116126710A (en) | Vehicle-mounted software simulation method, device, equipment and medium | |
| Mikkonen et al. | Exploring virtual reality as an integrated development environment for cyber-physical systems | |
| Hause | The digital twin throughout the se lifecycle | |
| Jalote et al. | Studying task processes for improving programmer productivity | |
| CN115330200A (en) | Apparatus, system and method for operating artificial intelligence model | |
| Dukkardt et al. | Informational system to support the design process of complex equipment based on the mechanism of manipulation and management for three-dimensional objects models | |
| CN114399804A (en) | Virtual laboratory teaching realization method, system, medium and equipment based on 3D simulation | |
| JP2022136983A (en) | Automatic generation of integrated test procedures using system test procedures |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |