Model development management method and device and computing equipment

Abstract

The invention discloses a model development management method and device and computing equipment, and the method comprises the steps: receiving a training task creation operation triggered at a client, the training task creation operation comprising an operation of inputting training basic information, and the training basic information comprising a training data set, a training code version and training parameters; creating a training task of the model based on the training basic information; in response to a starting request of the client for the training task, starting the training container through the trainer; and executing the training task through the training container: executing the code of the training code version through the training container based on the training parameters so as to train the model based on the training data set and obtain a trained model corresponding to the training task. On the basis, efficient and convenient model development and management can be realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a model development management method, a model development management device, and a computing device. Background Technology

[0002] With the rapid development of artificial intelligence technology, the development process of machine learning models has become increasingly complex. Traditional model development processes typically suffer from the following pain points: 1) Complex environment configuration: There are many versions of deep learning frameworks and dependent libraries. Environment configuration is time-consuming and labor-intensive, and it is difficult to ensure the consistency of training and inference environments, which can easily lead to compatibility issues.

[0003] 2) Inconvenient task management: There is a lack of effective recording methods for training and inference tasks, and it is impossible to view historical operation records and results.

[0004] 3) Disorganized resource management: Dataset, model version, and code version management are disorganized, making it difficult to trace and reproduce experimental results of specific versions during team collaboration.

[0005] 4) Lack of process monitoring: The training process is usually monitored by printing logs, lacking an intuitive and visual real-time monitoring interface. Developers cannot view key indicators such as training progress and loss function change curves in real time, leading to a delay in problem detection.

[0006] 5) Inconvenient access to output results: The training output model files, logs, and inference output result files (such as images) need to be manually searched and downloaded by users, which is a cumbersome process.

[0007] Therefore, a model development management method is needed to solve the problems existing in the above technical solutions. Summary of the Invention

[0008] Therefore, the present invention provides a model development management method and a model development management device to solve or at least alleviate the problems mentioned above.

[0009] According to one aspect of the present invention, a model development management method is provided, executed on a server, comprising: receiving a training task creation operation triggered on a client, the training task creation operation including an operation of inputting basic training information, the basic training information including a training dataset, a training code version, and training parameters; creating a training task for a model based on the basic training information; responding to a client's request to start the training task, starting a training container through a trainer; and executing the training task through the training container, comprising: executing the code of the training code version based on the training parameters through the training container, so as to train the model based on the training dataset to obtain a trained model corresponding to the training task.

[0010] Optionally, the model development management method according to the present invention further includes: receiving an inference task creation operation triggered on a client, the inference task creation operation including an operation of inputting basic inference information, the basic inference information including an inference dataset, an inference code version, and inference parameters; creating an inference task for the model based on the basic inference information; starting an inference container through an inference engine in response to a client's request to start the inference task; and executing the inference task through the inference container, including: executing the code of the inference code version based on the inference parameters through the inference container, so as to perform inference on the model based on the inference dataset and obtain an inference result.

[0011] Optionally, the model development management method according to the present invention further includes: receiving management operations triggered on the client interface for datasets, code, training tasks, inference tasks, models, or runtime environments, and executing the management operations; wherein the management operations include one or more of list viewing operations, creation operations, deletion operations, modification operations, query operations, and file download operations.

[0012] Optionally, the model development management method according to the present invention further includes: periodically monitoring the state of the training task and the state of the inference task through a task monitor, so as to update the state of the training task or the state of the inference task when a change in the state of the training task or the state of the inference task is detected.

[0013] Optionally, in the model development management method according to the present invention, obtaining the trained model corresponding to the training task includes: after training the model, obtaining checkpoints from the output path of the training task, obtaining the trained model based on the checkpoints, and saving it.

[0014] Optionally, in the model development management method according to the present invention, starting the training container through the trainer includes: creating a trainer, starting the training container through the trainer based on the running environment corresponding to the training task, and passing the training parameters corresponding to the training task to the training container.

[0015] Optionally, in the model development management method according to the present invention, based on the basic training information, creating a training task for model training includes: generating a unique training ID corresponding to the training task; associating the basic training information with the training ID and storing it in the training table of the database system; and creating a folder corresponding to the training ID in the training directory as the output path of the training task.

[0016] Optionally, in the model development management method according to the present invention, the client includes a browser; the server is adapted to interact with the client via OpenAPI.

[0017] According to one aspect of the present invention, a model development management apparatus is provided, deployed on a server, suitable for executing the method described above, the apparatus comprising: The receiving module is adapted to receive a training task creation operation triggered by the client. The training task creation operation includes an operation to input basic training information, which includes the training dataset, training code version, and training parameters. A module is created to create a training task for the model based on the aforementioned basic training information; The startup module is adapted to start the training container via the trainer in response to a client's request to start the training task. An execution module, adapted to execute the training task through the training container, includes: executing the code of the training code version through the training container based on the training parameters, so as to train the model based on the training dataset and obtain the trained model corresponding to the training task.

[0018] According to one aspect of the present invention, a computing device is provided, comprising: at least one processor; and a memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, the program instructions including instructions for performing the model development management method as described above.

[0019] According to one aspect of the present invention, a computer program product is provided, comprising computer program instructions, wherein the computer program instructions, when executed by a processor, implement the method as described above.

[0020] According to one aspect of the present invention, a readable storage medium storing program instructions is provided, which, when read and executed by a computing device, causes the computing device to perform the model development management method as described above.

[0021] According to the technical solution of the present invention, a model development and management method is provided. Users can input basic training information on a client interface to trigger the creation of a training task and initiate the training task. The server can then start a training container through a trainer and execute the training task to obtain the trained model corresponding to the training task. Based on this, the present invention can achieve efficient and convenient model development and management.

[0022] Furthermore, this invention enables unified and visual management of datasets, code, training tasks, inference tasks, models, and runtime environments, integrating the entire model development process and improving model development efficiency.

[0023] Furthermore, by storing basic task information and output data in a structured manner, this invention enables one-click reproduction of previously run training or inference tasks.

[0024] Furthermore, this invention enables visual monitoring of the model training or inference process and allows for previewing of the inference results.

[0025] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0026] To achieve the foregoing and related objectives, certain illustrative aspects are described herein in conjunction with the following description and accompanying drawings. These aspects indicate various ways in which the principles disclosed herein may be practiced, and all aspects and their equivalents are intended to fall within the scope of the claimed subject matter. The foregoing and other objectives, features, and advantages of the invention will become more apparent from the following detailed description, taken in conjunction with the accompanying drawings. Throughout the invention, the same reference numerals generally refer to the same parts or elements.

[0027] Figure 1 A schematic diagram of a computing device 100 provided according to an embodiment of the present invention is shown; Figure 2 A flowchart illustrating a model development management method 200 provided according to an embodiment of the present invention is shown; Figure 3 A schematic diagram of a training task list interface according to some embodiments of the present invention is shown; Figure 4 A schematic diagram of a training task creation interface according to some embodiments of the present invention is shown; Figure 5 A schematic diagram of a training task details interface according to some embodiments of the present invention is shown; Figure 6 A schematic diagram of a training state interface according to some embodiments of the present invention is shown; Figure 7 A schematic diagram of a model development management device 700 provided according to an embodiment of the present invention is shown. Detailed Implementation

[0028] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0029] To address the problems existing in the current machine learning model development process, this invention proposes a model development management method that can integrate the entire model development process, provide efficient, convenient, and visual management of the entire model development lifecycle, and improve model development efficiency.

[0030] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] Figure 1 A schematic diagram of a computing device 100 according to an embodiment of the present invention is shown. Figure 1 As shown, in a basic configuration, computing device 100 includes at least one processing unit 102 and system memory 104. According to one aspect, depending on the configuration and type of the computing device, the processing unit 102 may be implemented as a processor. System memory 104 includes, but is not limited to, volatile memory (e.g., random access memory), non-volatile memory (e.g., read-only memory), flash memory, or any combination of such memories. According to one aspect, system memory 104 includes an operating system 105.

[0032] According to one aspect, operating system 105 is, for example, suitable for controlling the operation of computing device 100. Furthermore, examples are practiced in conjunction with graphics libraries, other operating systems, or any other applications, and are not limited to any particular application or system. Figure 1 The basic configuration is illustrated by the components within the dashed lines. According to one aspect, the computing device 100 has additional features or functions. For example, according to one aspect, the computing device 100 includes additional data storage devices (removable and / or non-removable), such as disks, optical discs, or magnetic tapes. This additional storage... Figure 1 The image is shown by removable storage device 109 and non-removable storage device 110.

[0033] As stated above, according to one aspect, program module 103 is stored in system memory 104. According to one aspect, program module 103 may include one or more applications. The present invention does not limit the type of application; for example, applications may include: email and contact applications, word processing applications, spreadsheet applications, database applications, slideshow applications, drawing or computer-aided applications, web browser applications, etc.

[0034] According to one aspect, program module 103 may include a plurality of program instructions suitable for executing the model development management method 300 of the present invention, such that computing device 100 is configured to execute the model development management method 300 of the present invention.

[0035] According to one aspect, program module 103 may include model development management device 700, which may be configured to execute the model development management method 300 of the present invention.

[0036] According to one aspect, examples can be practiced on circuits including discrete electronic components, packaged or integrated electronic chips containing logic gates, circuits utilizing microprocessors, or on a single chip containing electronic components or a microprocessor. For example, it can be practiced via wherein... Figure 1 Each or many of the components shown can be implemented as an example by integrating a System-on-a-Chip (SOC) on a single integrated circuit. According to one aspect, such an SOC device may include one or more processing units, graphics units, communication units, system virtualization units, and various application functions, all integrated (or “burned in”) as a single integrated circuit onto a chip substrate. When operating via the SOC, the functions described herein can be operated via dedicated logic integrated on a single integrated circuit (chip) with other components of the computing device 100. Embodiments of the invention can also be implemented using other techniques capable of performing logical operations (e.g., AND, OR, and NOT), including but not limited to mechanical, optical, fluid, and quantum technologies. Additionally, embodiments of the invention can be implemented within a general-purpose computer or in any other circuit or system.

[0037] According to one aspect, computing device 100 may also have one or more input devices 112, such as a keyboard, mouse, pen, voice input device, touch input device, etc. It may also include output devices 114, such as a display, speaker, printer, etc. The foregoing devices are examples and other devices may also be used. Computing device 100 may include one or more communication connections 116 that allow communication with other computing devices 118. Examples of suitable communication connections 116 include, but are not limited to: RF transmitter, receiver and / or transceiver circuitry; Universal Serial Bus (USB), parallel and / or serial ports.

[0038] As used herein, the term computer-readable medium includes computer storage medium. Computer storage medium can include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information (e.g., computer-readable instructions, data structures, or program module 103). System memory 104, removable storage device 109, and non-removable storage device 110 are examples of computer storage media (i.e., memory storage). Computer storage media can include random access memory (RAM), read-only memory (ROM), electrically erasable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape, magnetic tape, disk storage or other magnetic storage devices, or any other article of manufacture that can be used to store information and is accessible by computing device 100. According to one aspect, any such computer storage medium can be part of computing device 100. Computer storage media does not include carrier waves or other transmitted data signals.

[0039] According to one aspect, the communication medium is implemented by computer-readable instructions, data structures, program modules 103, or other data in a modulated data signal (e.g., a carrier wave or other transmission mechanism), and includes any information transmission medium. According to one aspect, the term "modulated data signal" describes a signal having one or more sets of characteristics or altered in a manner that encodes information in the signal. By way of example and not limitation, the communication medium includes wired media such as wired networks or direct wired connections, and wireless media such as acoustic, radio frequency (RF), infrared, and other wireless media.

[0040] In an embodiment of the present invention, a computing device 100 is configured to execute the model development management method 300 of the present invention. The computing device 100 includes one or more processors and one or more readable storage media storing program instructions that, when configured to be executed by one or more processors, cause the computing device to execute the model development management method 300 of the present invention.

[0041] Figure 2 A schematic flowchart of a model development management method 200 according to an embodiment of the present invention is shown. The model development management method 200 can be executed in a computing device (such as the aforementioned computing device 100).

[0042] In this embodiment of the invention, the computing device 100 for executing the model development management method 300 of the present invention can be a server. The server can communicate with one or more clients. In some embodiments, the client includes a browser, which can communicate with the server.

[0043] In some embodiments, the server can interact with the client via OpenAPI.

[0044] like Figure 2 As shown, the model development management method 200 includes at least the following steps 210-240.

[0045] Step 210: The server can receive training task creation operations triggered by the user on the client. Here, the training task creation operation triggered by the user on the client includes the user inputting basic training information on the client. In other words, the user can input basic training information on the client and thus trigger the training task creation operation.

[0046] In this embodiment of the invention, the basic training information includes the training dataset, the training code version, and the training parameters. In some embodiments, the basic training information may further include the runtime environment (Docker image) configured for the training code version.

[0047] In some embodiments, the user can first click the "Create Training Task" button on the client interface (training task list interface), then enter the basic training information for the training task to be created on the subsequently displayed "Create Training Task" interface, and then click the "OK" button on the "Create Training Task" interface to save the basic training information and trigger the training task creation operation. Thus, in step 210, the server can receive the training task creation operation triggered by the user on the client.

[0048] Step 220: The server can create a training task for the model based on the basic training information. It should be understood that the basic training information here is the same as the basic training information corresponding to the training task.

[0049] It should be noted that the server can execute steps 210-220 above multiple times to create multiple training tasks.

[0050] In some embodiments, the server is coupled to a database system. When creating a training task for the model based on basic training information, the server first generates a unique training ID corresponding to the training task. Specifically, the server can generate a UUID as the unique training ID for the training task. Subsequently, the basic training information (including training dataset, training code version, training parameters, runtime environment, etc.) can be associated with the training ID and stored in the training table of the database system, thereby generating a record corresponding to the training ID (training task) in the training table. Furthermore, the server can create a folder corresponding to the training ID (training task) in the training directory (specifically, a folder with the training ID as the filename), and use the folder corresponding to the training ID as the output path of the training task (i.e., the output path of the training code version), so as to store the output data of the training task based on the output path of the training task.

[0051] It should be understood that the training table of the database system can store records corresponding to one or more training tasks that have been created (i.e., records corresponding to the training IDs of the training tasks).

[0052] Step 230: The server can receive the client's request to start the training task, and then respond to the client's request by starting the training container through the trainer. Then, step 240 can be executed to perform the training task through the training container.

[0053] It should be noted that the client's request to start the training task received in step 230 can be a request from the client to start any training task that has already been created.

[0054] Step 240: The server can train the model by executing training tasks (training tasks initiated by client requests) through the training container. Specifically, the server can execute the training code version corresponding to the training task based on the training parameters (i.e., the training parameters in the basic training information) of the training task through the training container, so as to train the model based on the training dataset corresponding to the training task and obtain the trained model corresponding to the training task.

[0055] In this embodiment of the invention, the code is a script program used for model training and inference. The code version may include basic code configuration information (such as code version number, runtime environment, and code file), and may also include training parameters and inference parameters.

[0056] In some embodiments, the server starts the training container via the trainer as follows: First, the server can create a trainer to start the training container. Then, it can start the training container based on the runtime environment (image) corresponding to the training task, passing the training parameters corresponding to the training task to the training container. Simultaneously, the file path of the training dataset, the file path of the training code version, and the output path of the training task can be mounted into the training container. Furthermore, the training container can execute the training task (executing the training code version based on the training parameters to train the model based on the training dataset). Here, the runtime environment corresponding to the training task is the runtime environment configured for the training code version of the training task.

[0057] In some embodiments, after the server starts the training container through the trainer, it can also store the training container ID in the record corresponding to the training task (training ID) in the training table of the database system.

[0058] In some embodiments, the server can save the training state (including the number of training epochs and steps) in real time during the execution of the training task (training code version of the code) through the training container. Users can request to view the training state, training progress, and loss curve from the client. The server can respond to the client's request to view the training state, training progress, and loss curve by returning the corresponding training state value (including the number of training epochs and steps), training progress information, and loss curve to the client.

[0059] In some embodiments, during the process of the server training the model by executing training tasks through training containers, the user can also trigger a stop request for the training task (the training task being executed) on the client side. The server can respond to the client's stop request for the training task by stopping the training container corresponding to the training task through the docker stop command.

[0060] In some embodiments, after the model training is complete, the server can obtain checkpoints (checkpoint files) from the output path of the training task. Based on the checkpoints, the trained model (the trained model corresponding to the training task) can be obtained and saved. That is, checkpoints can be saved as the trained model. It should be noted that checkpoints are used to save the model state at different stages. Therefore, the trained model can be obtained by directly loading the contents of the checkpoints, so that the trained model can be used for inference or deployment.

[0061] In some embodiments, the basic training information corresponding to the training task may also include the training code name. In step 240, the server can use the training container to execute the code corresponding to the training code name and training code version based on the training parameters corresponding to the training task (i.e., the training parameters in the basic training information), so as to train the model based on the training dataset corresponding to the training task and obtain the trained model corresponding to the training task.

[0062] According to the model development management method 200 in the embodiments of the present invention, the creation and initiation of inference tasks can be carried out in a manner similar to the creation and initiation of training tasks in the above embodiments.

[0063] Specifically, the server can also receive inference task creation operations triggered by users on the client, and then create inference tasks for the model based on basic inference information.

[0064] Here, the inference task creation operation triggered by the user on the client includes the user inputting basic inference information on the client. In other words, the user can input basic inference information on the client and then trigger the inference task creation operation.

[0065] In this embodiment of the invention, the basic inference information includes the inference dataset, the inference code version, and the inference parameters. In some embodiments, the basic inference information may also include the runtime environment (Docker image) configured for the inference code version.

[0066] In some embodiments, the user can first click the "Create Inference Task" button on the client's interface (Inference Task List Interface), then enter the basic inference information for the inference task to be created on the subsequently displayed "Create Inference Task" interface, and then click the "OK" button on the "Create Inference Task" interface to save the basic inference information and trigger the inference task creation operation. Thus, in step 210, the server can receive the inference task creation operation triggered by the user on the client.

[0067] In some embodiments, when the server creates an inference task based on basic inference information, it can first generate a unique inference ID for the inference task. Specifically, the server can generate a UUID as the unique inference ID for the inference task. Subsequently, the basic inference information (including inference dataset, inference code version, inference parameters, runtime environment, etc.) can be associated with the inference ID and stored in the inference table of the database system, thereby generating a record corresponding to the inference ID (inference task) in the inference table. Furthermore, the server can create a folder corresponding to the inference ID (inference task) in the inference directory (specifically, a folder named after the inference ID), and use this folder as the output path for the inference task (i.e., the output path of the inference code version), so that the output data of the inference task can be stored based on the output path of the inference task.

[0068] It should be understood that the inference table of a database system can store records corresponding to one or more inference tasks that have been created (i.e., records corresponding to the inference ID of the inference task).

[0069] Furthermore, the server can receive client requests to start inference tasks, and then respond to these requests by launching the inference container through the inference engine. The server can then execute the inference task (the inference task requested by the client) through the inference container. Specifically, the server can use the inference container to execute the inference code version corresponding to the inference task based on the inference parameters (i.e., the inference parameters in the basic inference information), so as to perform inference on the model based on the inference dataset corresponding to the inference task (controlling the model to perform inference based on the inference dataset) and obtain the inference result.

[0070] It should be noted that the client's request to start an inference task can be a request to start any inference task that has already been created.

[0071] In some embodiments, the server starts the inference container via the inferencer as follows: First, the server creates an inferencer to start the inference container. Then, it starts the inference container based on the runtime environment (image) corresponding to the inference task, passing inference parameters corresponding to the inference task to the inference container. Simultaneously, the file path of the inference dataset, the file path of the inference code version, and the output path of the inference task can be mounted into the inference container. Then, the inference container executes the inference task (executing the inference code version based on the inference parameters to perform inference on the model based on the inference dataset). Here, the runtime environment corresponding to the inference task is the runtime environment configured for the inference code version corresponding to the inference task.

[0072] In some embodiments, after the server starts the inference container through the inference engine, it can also store the inference container ID in the record corresponding to the inference task (inference ID) in the inference table of the database system.

[0073] In some embodiments, the server can save the inference state (including the number of steps) in real time during the execution of the inference task (inference code version) through the inference container. Users can request to view the inference state and progress from the client, and the server can respond to the client's request by returning the corresponding inference state value and progress information. Furthermore, after the server completes the inference task and inference of the model, users can also request to view the inference results from the client, and the server can respond to the client's request by returning the inference results.

[0074] In some embodiments, while the server is performing inference tasks on the model through an inference container, the user can also trigger a stop request for the inference task (the inference task being executed) on the client. The server can respond to the client's stop request for the inference task by stopping the inference container corresponding to the inference task using the docker stop command.

[0075] In some embodiments, the basic inference information corresponding to an inference task may also include an inference code name. The server can use an inference container to execute the code corresponding to the inference task and the inference code version based on the inference parameters corresponding to the inference task (i.e., the inference parameters in the basic inference information), so as to perform inference on the model based on the inference dataset corresponding to the inference task.

[0076] In some embodiments, a task monitor is deployed on the server side. This monitor periodically monitors the status of the training and inference tasks and updates the status of either task when a change is detected. The status of the training and inference tasks may include, for example, creation, running, termination, stop, and exit.

[0077] In some embodiments, users can trigger management operations on datasets, code, training tasks, inference tasks, models, or runtime environments via the client interface. Management operations include one or more of the following: list viewing, creation, deletion, modification, querying, and file download. Upon receiving a management operation triggered by the user on the client interface, the server executes the operation. Based on this, the present invention enables unified and visual management of datasets, code, training tasks, inference tasks, models, and runtime environments, integrating the entire model development process and improving model development efficiency.

[0078] Specifically, for the creation of datasets, code, and runtime environments, the server can receive dataset files, code files, runtime environment images, and corresponding basic information imported by the user through the client interface, and save the dataset files, code files, and runtime environment images in the corresponding directories, and save the basic information in the response data table of the database system.

[0079] For model creation, users can input basic model information (including model name, code version, training task name, and checkpoint information) on the client interface to trigger the model creation operation. The server can respond to the model creation operation from the client, creating the model based on the basic model information. Specifically, it can generate a unique model ID and associate the basic model information with the model ID, storing them in the model table of the database system. Simultaneously, it can create a folder corresponding to the model ID in the model directory (specifically, a folder named after the model ID). Then, based on the basic model information, it can retrieve the corresponding checkpoint from the output path of the corresponding training task (i.e., the folder corresponding to the training ID in the training directory) and copy the contents of that checkpoint to the folder corresponding to the model ID, so that the model can be created based on the checkpoint.

[0080] In some embodiments, management operations for training tasks (list viewing, creation, deletion, modification, querying, and file download operations) can be implemented in the following ways: Before executing step 210, the client can request access to the model development management interface from the server (sending an access request to the model development management interface to the server). The server can respond to the client's access request and return the model development management interface to the client for display. The model development management interface contains multiple menus, such as code, dataset, training, model, inference, and runtime environment menus. Afterward, the user can click the training menu on the model development management interface displayed on the client to trigger a list view of training tasks, displaying the training task list interface on the client.

[0081] Figure 3 A schematic diagram of a training task list interface according to some embodiments of the present invention is shown. Figure 3 As shown, the training task list interface displays a training task list from the server (the training task list generated by the server based on the training table). The training task list includes one or more entries of created training tasks. Each training task entry has corresponding information such as name, dataset (training dataset), code name, code version (training code version), source, status, and creation time.

[0082] In addition, there is a create button on the training task list interface (i.e., the training task creation button). Users can click the training task creation button on the training task list interface displayed on the client to trigger the display of the training task creation interface. Figure 4 A schematic diagram of an interface for creating training tasks according to some embodiments of the present invention is shown. For example... Figure 4 As shown, users can enter the basic training information of the training task to be created on the client's Create Training Task interface, including the name of the training task, description, dataset (training dataset), source, code name (training code name), code version (training code version), and training parameters. Then, they can click the OK button on the Create Training Task interface to save the basic training information and trigger the training task creation operation.

[0083] like Figure 3 As shown, each training task entry on the training task list interface also has corresponding operation buttons, including edit and delete buttons. Users can click the delete button for any training task (even created ones) on the training task list interface to delete that training task. Users can click the edit button for any training task (even created ones) on the training task list interface to modify that training task. The client displays a modify training task interface (similar to the create training task interface described above), where users can then modify the basic training information of the training task.

[0084] In addition, users can click one or more filter buttons (such as those for name, source, status, and creation time) in the header of the training task list on the training task list interface to set query conditions, so as to trigger query operations on training tasks based on the query conditions. Subsequently, the server can respond to the query operation, retrieve the corresponding records from the training table of the database system based on the query conditions, and return them to the client for display.

[0085] In addition, users can click on the name of any training task (the name in the training task entry) on the training task list interface to enter the training task details interface corresponding to that name. Figure 5 A schematic diagram of a training task details interface according to some embodiments of the present invention is shown. Figure 5 As shown, the training task details interface contains multiple tabs, specifically the Information tab, Parameters tab, Status tab, Log tab, and File tab. Users can click the File tab in the training task details interface to trigger the display of the file list corresponding to the training task, and then click the download button in the file list to trigger the file download operation for the training task. Subsequently, the server receives the file download operation from the client, retrieves the corresponding training task file from the training directory based on the training ID and file path corresponding to the file download operation, and returns it to the client.

[0086] like Figure 5 As shown, the training task details interface also includes a start button. In step 230, the user can click the name of the training task to be started (the name in the training task entry) on the training task list interface to enter the corresponding training task details interface. Subsequently, the user can click the start button on the training task details interface to trigger a start request for the training task. Then, the server can respond to the client's start request for the training task, start the training container through the trainer, and then execute the training task through the training container to train the model.

[0087] Additionally, during the training process of the model by the server executing training tasks through the training container, the user can click the status tab in the training task details interface to trigger the display of the training status interface (which requests the training status value, training progress information, and loss curve corresponding to the training task from the server and generates the training status interface to be displayed on the client). Figure 6 A schematic diagram of a training state interface according to some embodiments of the present invention is shown. For example... Figure 6As shown, the training status interface can display the training status values ​​(including training rounds and steps), training progress information, and loss curves corresponding to the training task, thereby enabling visual monitoring of the model training process. It should be noted that for inference tasks, the server can also respond to the client's request to view the inference results by sending the model's inference results (e.g., images generated by inference) to the client for visual display.

[0088] Similarly, management operations (including list viewing, creation, deletion, modification, querying, and file download operations) for inference tasks, datasets, code (code versions), models, and runtime environments can be implemented in a similar manner to the management operations for training tasks described in the above embodiments. The difference lies in the specific information related to the object triggered and displayed on various interfaces of the client during the management operation, depending on the object being managed. Here, the specific implementation methods for various management operations of inference tasks, datasets, code, models, and runtime environments will not be repeatedly described. Furthermore, it should be noted that this invention is not limited to the specific implementation methods in the above embodiments; any method that enables visual management of datasets, code, training tasks, inference tasks, models, or runtime environments is acceptable.

[0089] It should also be noted that, in this embodiment of the invention, in order to distinguish between training tasks and inference tasks, the dataset, code version, code name, and container corresponding to the training task are respectively referred to as "training dataset," "training code version," "training code name," and "training container." Similarly, the dataset, code version, code name, and container corresponding to the inference task are respectively referred to as "inference dataset," "inference code version," "inference code name," and "inference container."

[0090] Figure 7 A schematic diagram of a model development management device 700 according to an embodiment of the present invention is shown. The model development management device 700 can be deployed on a server (e.g., the aforementioned computing device 100), and the model development management device 700 is configured to execute the model development management method 200 of the present invention.

[0091] like Figure 7 As shown, in an embodiment of the present invention, the model development management device 700 includes a receiving module 710, a creation module 720, a startup module 730, and an execution module 740 that are connected in sequence.

[0092] The receiving module 710 can receive the training task creation operation triggered by the client. The training task creation operation includes the operation of inputting basic training information, which includes the training dataset, training code version, and training parameters.

[0093] The creation module 720 can create training tasks for the model based on basic training information.

[0094] The startup module 730 can respond to the client's request to start the training task and start the training container through the trainer.

[0095] The execution module 740 can execute training tasks through the training container, including: executing the training code version of the code based on the training parameters through the training container, so as to train the model based on the training dataset and obtain the trained model corresponding to the training task.

[0096] It should be noted that the receiving module 710, the creation module 720, the starting module 730, and the execution module 740 are respectively used to execute the aforementioned steps 210 to 240. Here, the specific execution logic of each unit can be found in the description of steps 210 to 240 in the previous method 200, and will not be repeated here.

[0097] According to the model development and management method 300 in this embodiment of the invention, a user can input basic training information on a client interface to trigger a training task creation operation and initiate the training task. The server can then start a training container through the trainer and execute the training task through the training container to obtain the trained model corresponding to the training task. Based on this, the present invention can achieve efficient and convenient model development and management.

[0098] Furthermore, this invention enables unified and visual management of datasets, code, training tasks, inference tasks, models, and runtime environments, integrating the entire model development process and improving model development efficiency.

[0099] Furthermore, by storing basic task information and output data in a structured manner, this invention enables one-click reproduction of previously run training or inference tasks.

[0100] Furthermore, this invention enables visual monitoring of the model training or inference process and allows for previewing of the inference results.

[0101] The various techniques described herein can be implemented in combination with hardware or software, or a combination thereof. Thus, the methods and apparatus of the present invention, or certain aspects or portions thereof, can take the form of program code (i.e., instructions) embedded in a tangible medium, such as a removable hard disk, USB flash drive, floppy disk, CD-ROM, or any other machine-readable storage medium, wherein when the program is loaded into and executed by a machine such as a computer, the machine becomes an apparatus for practicing the present invention.

[0102] When the program code is executed on a programmable computer, the mobile terminal generally includes a processor, a processor-readable storage medium (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device. The memory is configured to store program code; the processor is configured to execute the model development management method of the present invention according to instructions in the program code stored in the memory.

[0103] By way of example, and not limitation, readable media include readable storage media and communication media. Readable storage media stores information such as computer-readable instructions, data structures, program modules, or other data. Communication media generally embodies computer-readable instructions, data structures, program modules, or other data in the form of modulated data signals such as carrier waves or other transmission mechanisms, and includes any information delivery medium. Any combination of the above is also included within the scope of readable media.

[0104] In the specification provided herein, the algorithms and displays are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used with the examples of this invention. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0105] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0106] Similarly, it should be understood that, in order to streamline this disclosure and aid in understanding one or more of the various aspects of the invention, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof.

[0107] Those skilled in the art will understand that modules, units, or components of the devices disclosed in the examples herein can be arranged in the devices described in this embodiment, or alternatively, can be located in one or more devices different from the devices in this example. The modules in the foregoing examples can be combined into a single module or, in addition, can be divided into multiple sub-modules.

[0108] Unless otherwise specified, the use of ordinal numbers such as “first,” “second,” “third,” etc., to describe ordinary objects merely indicates different instances of similar objects and is not intended to imply that the objects being described must have a given order in time, space, ordering, or any other manner.

Claims

1. A model development management method, executed on the server side, comprising: Receive a training task creation operation triggered on the client, the training task creation operation includes an operation to input basic training information, the basic training information includes training dataset, training code version, and training parameters; Based on the aforementioned basic training information, a training task for the model is created; In response to the client's request to start the training task, the training container is started through the trainer; Executing the training task through the training container includes: executing the code of the training code version through the training container based on the training parameters, so as to train the model based on the training dataset and obtain the trained model corresponding to the training task.

2. The method as described in claim 1, wherein, Also includes: Receive an inference task creation operation triggered on the client, the inference task creation operation including an operation to input basic inference information, the basic inference information including inference dataset, inference code version, and inference parameters; Based on the aforementioned basic information, a reasoning task for the model is created; In response to the client's request to start the inference task, the inference container is started via the inferencer; Executing the inference task through the inference container includes: executing the code of the inference code version through the inference container based on the inference parameters, so as to infer the model based on the inference dataset and obtain the inference result.

3. The method as described in claim 1 or 2, wherein, Also includes: Receive management operations triggered on the client interface for datasets, code, training tasks, inference tasks, models, or runtime environments, and execute the management operations. The management operations include one or more of the following: list viewing, creation, deletion, modification, query, and file download.

4. The method according to any one of claims 1-3, wherein, Also includes: The status of the training task and the inference task are monitored periodically by the task monitor so that the status of the training task or the inference task can be updated when a change is detected.

5. The method according to any one of claims 1-4, wherein, Obtaining the trained model corresponding to the training task includes: After the model is trained, checkpoints are obtained from the output path of the training task, and the trained model is obtained and saved based on the checkpoints.

6. The method according to any one of claims 1-5, wherein, Starting the training container via the trainer includes: A trainer is created, and a training container is started based on the runtime environment corresponding to the training task through the trainer, and the training parameters corresponding to the training task are passed to the training container.

7. The method according to any one of claims 1-6, wherein, Based on the aforementioned basic training information, a training task for model training is created, including: Generate a unique training ID corresponding to the training task, and associate the basic training information with the training ID and store it in the training table of the database system; Create a folder corresponding to the training ID in the training directory as the output path for the training task.

8. The method according to any one of claims 1-7, wherein, The client includes a browser; The server is adapted to interact with the client via OpenAPI.

9. A model development management device, deployed on a server, adapted to execute the method as described in any one of claims 1-8, the device comprising: The receiving module is adapted to receive a training task creation operation triggered by the client. The training task creation operation includes an operation to input basic training information, which includes the training dataset, training code version, and training parameters. A module is created to create a training task for the model based on the aforementioned basic training information; The startup module is adapted to start the training container via the trainer in response to a client's request to start the training task. An execution module, adapted to execute the training task through the training container, includes: executing the code of the training code version through the training container based on the training parameters, so as to train the model based on the training dataset and obtain the trained model corresponding to the training task.

10. A computing device, comprising: At least one processor; and A memory storing program instructions, wherein the program instructions are configured to be processed by the at least one processor, the program instructions including instructions for processing the method as described in any one of claims 1-9.

11. A computer program product comprising computer program instructions, wherein, When the computer program instructions are executed by the processor, they implement the method as described in any one of claims 1-9.

Images