Multi-model combination development platform, development method and related devices
By using a unified API interface and driver scheduling in a multi-model combination development platform, the problems of complex manual configuration and inconsistent interfaces in multi-model combination are solved, and an efficient and flexible data processing workflow is achieved.
Patent Information
- Application Number
- CN202411896102.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-20
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2044-12-20
AI Technical Summary
Existing technologies suffer from problems such as complex manual configuration, high risk of human error, and low development efficiency due to inconsistent model interfaces during multi-model combination.
A multi-model combination development platform is provided, which manages multiple candidate components through a unified API interface, schedules component processing flow using drivers, manages data transmission through data pipelines, and constructs a directed acyclic graph to ensure sequential processing.
It lowers the barrier to development of multi-model combinations, improves development efficiency, reduces human error, ensures the continuity and flexibility of data transmission, and adapts to diverse business needs.
Smart Images

Figure CN119810545B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of data processing technology, and in particular to the fields of artificial intelligence, large models, deep learning, and model combinatorial development. Background Technology
[0002] With the development of artificial intelligence (AI) technology, its applications are becoming increasingly widespread across various fields, making it a significant driving force for development. As model scale continues to expand and technology matures, different neural network models are demonstrating superior processing capabilities for specific tasks. Summary of the Invention
[0003] This disclosure provides a multi-model combination development platform, development method, and related apparatus.
[0004] According to one aspect of this disclosure, a multi-model combination development platform is provided, comprising:
[0005] The component set includes multiple candidate components, each encapsulating corresponding data processing operations; the multiple candidate components have a unified API (Application Programming Interface) to support the construction of a processing flow based on at least one neural network model by defining the relationships between the input and output interfaces of different target components among the multiple candidate components; at least one neural network model is determined based on multiple target components.
[0006] A driver is used to schedule multiple target components to process data sequentially according to a processing flow.
[0007] Data pipelines are used to provide data to multiple target components and store data involved in the processing flow.
[0008] According to one aspect of this disclosure, a multi-model combination development method is provided, comprising:
[0009] In response to the association between the input and output interfaces of multiple target components defined by coding, the connection relationship between the multiple target components is determined; wherein, the multiple target components are components of multiple candidate components provided by the multi-model combination development platform, and the multiple candidate components have a unified API;
[0010] In response to the operation of adding multiple target components to the driver provided by the multi-model combination development platform, a processing flow based on at least one neural network model is built based on the driver; the at least one neural network model is determined based on multiple target components.
[0011] According to another aspect of this disclosure, a multi-model combination development apparatus is provided, comprising:
[0012] The determination module is used to determine the connection relationship between multiple target components in response to the association relationship between the input and output interfaces of multiple target components defined by coding; wherein, the multiple target components are components of multiple candidate components provided by the multi-model combination development platform, and the multiple candidate components have a unified application programming interface (API);
[0013] A module is built to respond to the operation of adding multiple target components to the driver provided by the multi-model combination development platform, and to build a processing flow based on at least one neural network model based on the driver; the at least one neural network model is determined based on multiple target components.
[0014] According to another aspect of this disclosure, an electronic device is provided, comprising:
[0015] At least one processor; and
[0016] The memory is communicatively connected to the at least one processor; wherein,
[0017] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described in the present disclosure.
[0018] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform any of the methods according to embodiments of this disclosure.
[0019] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the methods according to embodiments of this disclosure.
[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description
[0021] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:
[0022] Figure 1 This is a schematic diagram of the structure of a multi-model combination development platform provided according to an embodiment of the present disclosure;
[0023] Figure 2 This is a schematic diagram of a processing flow constructed by a driver according to an embodiment of the present disclosure;
[0024] Figure 3 This is another structural schematic diagram of a multi-model combination development platform provided according to an embodiment of the present disclosure;
[0025] Figure 4 This is a flowchart illustrating a multi-model combination development method provided according to an embodiment of the present disclosure;
[0026] Figure 5 This is another schematic diagram of a multi-model combination development method provided according to an embodiment of the present disclosure;
[0027] Figure 6 This is a schematic diagram illustrating the structure of multiple components depending on a data pipeline being executed sequentially, according to an embodiment of this disclosure;
[0028] Figure 7 This is a flowchart illustrating the sequential execution of multiple models, taking image processing as an example, according to an embodiment of the present disclosure.
[0029] Figure 8 This is a schematic diagram of a process for extracting key information from an image using a multi-model combination process according to an embodiment of the present disclosure.
[0030] Figure 9 This is a schematic diagram of the structure of a multi-model combination development apparatus provided according to an embodiment of the present disclosure;
[0031] Figure 10 This is another schematic diagram of a multi-model combination development apparatus provided according to an embodiment of the present disclosure;
[0032] Figure 11 This is a block diagram of an electronic device used to implement the multi-model combination development method of the embodiments of this disclosure. Detailed Implementation
[0033] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0034] As artificial intelligence technology continues to develop across various fields, business needs are also expanding and diversifying. This has led to a demand for combining multiple neural network models to address complex business challenges. However, in the actual process of combining multiple models, many significant problems have gradually emerged.
[0035] On the one hand, the process of combining multiple models often relies on complex manual configuration and integration operations. Developers need to spend a lot of time and effort coordinating parameter settings between different models. This manual operation is not only inefficient, but also prone to human error, making the entire development process full of uncertainty and risk.
[0036] On the other hand, the lack of a unified interface between different models makes it difficult for developers to adapt and adjust the interface of each model individually when combining models.
[0037] In view of this, embodiments of this disclosure provide a multi-model ensemble development platform. This platform can provide a unified API to help developers quickly build models and call them for inference and prediction, thereby lowering the barrier to entry for model ensemble development. Figure 1 This is a schematic diagram of the development platform system provided by the present invention, as shown below. Figure 1 As shown, it includes:
[0038] The component set 101 includes multiple candidate components, each of which encapsulates corresponding data processing operations. The multiple candidate components have a unified application development interface (API) to support the construction of a processing flow based on at least one neural network model by defining the relationship between the input and output interfaces of different target components among the multiple candidate components. The at least one neural network model is determined based on the multiple target components.
[0039] Among them, data processing operations refer to the specific functions possessed by each candidate component to perform specific processing on the data. For example, some candidate components can perform data format conversion, some can extract features from the data, and some can perform different data processing operations such as cropping or rotating images.
[0040] Based on a unified application programming interface (API) shared by multiple candidate components, the relationships between input and output interfaces of different target components can be defined to meet different task requirements. This allows for the construction of a processing flow for at least one neural network model using multiple target components. For example, target component A can first preprocess the raw data, then send the preprocessed data to target component B for feature extraction, and finally pass the data with extracted features to target component C for subsequent analysis.
[0041] Understandably, candidate components with no input or output are also supported. For example, a neural network inference data processor involves three operations, each of which implements a corresponding candidate component, including: a first component that copies data from memory to GPU memory, an inference component, and a second component that copies data from GPU memory to memory. The first component, which copies data to GPU memory, processes data in the form of pointers through the inference engine's interface, and therefore has no output. Similarly, the second component, which copies data to memory, has no input, while the inference component responsible for execution has neither input nor output.
[0042] Furthermore, candidate components support nesting. A candidate component can call other candidate components to complete data processing operations. The number of nesting levels can be set as needed, and this disclosure does not limit this.
[0043] Understandably, each component in the component set can be referred to as a candidate component, and after the processing flow is built, the component used in the processing flow is the target component.
[0044] Driver 102 is used to schedule multiple target components to process data sequentially according to the processing flow;
[0045] As the name suggests, a processing flow based on at least one neural network model, determined by multiple target components, does not have the ability to run itself. Therefore, the driver is responsible for providing the power to run the processing flow and enabling it to actually operate, thereby ensuring that each target component can participate in the data processing in an orderly manner. Thus, the driver 102 is the key to ensuring the effective execution of the entire processing flow built based on the target components.
[0046] Data pipeline 103 is used to provide data to multiple target components and store data involved in the processing flow.
[0047] The data involved in the processing flow may include input data and / or output data of any target component.
[0048] In the processing flow of a neural network model, the data pipeline is the infrastructure for data transmission. It can store data output by the target component and also provide data to the target component.
[0049] Throughout the entire process, once a target component completes its specific data processing task, the resulting intermediate data is stored in the data pipeline so that it can be provided to the target components of subsequent processes when needed, to meet the requirements of subsequent processing steps.
[0050] Whether it's the output data of the target component or the intermediate data generated during the processing flow, the data pipeline identifies the data source when it's stored, and provides relevant API access interfaces for the target components in the processing flow to access the data. In implementation, for any given target component, its output data and its component identifier are associated and stored in the data pipeline.
[0051] In practice, the data pipeline can be a dictionary in memory.
[0052] In this embodiment of the disclosure, based on a unified application API, developers do not need to learn the complex usage operations such as different calling methods and parameter settings for each component when using these target components. They can operate them according to the unified API to build a processing flow based on at least one neural network model. The driver can schedule multiple target components to process data sequentially according to the processing flow, ensuring the entire data processing process proceeds in an orderly manner and avoiding the problem of chaotic data transmission caused by manually coding the processing flow of target components. The data pipeline stores this data and manages it uniformly, ensuring the integrity of data in the processing flow and increasing the probability of the entire processing flow operating normally. In summary, the multi-model combination development platform provided by this embodiment of the disclosure can reduce the workload of developers, improve the efficiency of multi-model combination development, and meet diverse business needs by encapsulating a rich set of components. In conclusion, facing the diversity of different model input and output data types, a unified API enables unified management, improves system flexibility, can cope with complex and ever-changing real-world business scenarios, and meets users' needs for diverse data processing and model applications.
[0053] In this embodiment of the disclosure, in order to improve the running efficiency of the processing flow, the input stream and output stream of multiple candidate components are pipeline objects. The pipeline object of the input stream is used to obtain the input stream from the data pipeline, and the pipeline object of the output stream is used to store the output stream into the data pipeline.
[0054] In other words, the input stream pipeline object can interact with the data pipeline based on the relevant data reading API provided by the data pipeline, and extract data that meets the requirements of the component as the input stream data of the component.
[0055] Similarly, the output stream pipeline object can interact with the data pipeline based on the relevant write data API provided by the data pipeline, and store the output stream data in the data pipeline in an orderly manner according to the data pipeline's storage specifications and requirements.
[0056] In this embodiment of the disclosure, the input and output streams of the candidate components are designed as pipeline objects and a tight interaction mechanism is established with the data pipeline, so that the entire data processing flow can be carried out smoothly and continuously, avoiding the risk of data transmission chaos and data loss, while also improving data reusability and the flexibility and scalability of the entire processing flow.
[0057] In this embodiment of the disclosure, the data pipeline is also used to remove data that is not needed in the remaining processing flow from the data pipeline based on the processing progress of the processing flow.
[0058] In other words, the data pipeline sets a lifecycle for each piece of data in the processing flow, and automatically releases the data when it is no longer needed by the subsequent target components, thereby saving memory usage.
[0059] by Figure 2 For example, the output data of target component A and target component B serve as the input data of target component C. After target component C outputs its data, it is then processed by target component D. Once the processing progress indicates that the operation of target component C is complete, the remaining flow only includes target component D. Accordingly, the data of target components A and B are no longer used in the processing flow, and at this point, the output data of target components A and B can be removed from the data pipeline.
[0060] In this embodiment of the disclosure, the data pipeline is also used to remove data that is not needed in the remaining processing flow from the data pipeline based on the processing progress of the processing flow. This can reduce the complexity of data management in the data pipeline and ensure that the data remaining in the data pipeline is valid data that is closely related to the current and subsequent processing steps. This enables the entire data processing flow to continue to advance efficiently and stably, while saving the storage resources of the data pipeline.
[0061] In this embodiment of the disclosure, the driver schedules multiple target components to process data sequentially according to the processing flow. A directed acyclic graph can be constructed based on the processing flow. In the directed acyclic graph, nodes represent multiple target components, and edges represent data flows transmitted between different target components. The driver schedules multiple target components to process data sequentially according to the processing flow based on the directed acyclic graph.
[0062] During implementation, developers define the connections between different target components to define the processing flow. The driver can map multiple target components as nodes in a graph based on the pre-defined processing flow, and the data flows between different target components are represented as edges connecting these nodes. This visualizes the originally abstract processing flow as an intuitive directed acyclic graph structure, thereby determining the execution order and dependencies between the target components. This allows the driver to automatically schedule multiple predefined target components and obtain end-to-end data processing results.
[0063] The data flow is configured through the relevant APIs between the target components.
[0064] In this embodiment of the disclosure, the driver can abstract the processing flow into a directed acyclic graph based on the unified interface dependency relationship between target components, so that the driver can manage the entire data processing flow in an orderly manner, ensuring that each target component receives the correct data input at the correct time and obtains the data processing result end-to-end.
[0065] In this embodiment of the disclosure, in order to improve the flexibility and convenience of developing multi-model combinations, the multiple candidate components include at least one of the following types of components:
[0066] (1) Basic component, which encapsulates the smallest unit of data processing operation. This basic component may include multiple preprocessing components and / or multiple postprocessing components.
[0067] The preprocessing component is used to perform preprocessing operations on the input data;
[0068] For example, performing preprocessing on input data refers to transforming the raw input data into a form more suitable for subsequent processing steps, laying a good foundation for subsequent data processing and analysis.
[0069] Taking image processing as an example, the preprocessing component may perform image cropping to remove unnecessary edges or irrelevant backgrounds; it may also perform grayscale conversion, transforming a color image into a grayscale image to reduce data volume while preserving key image features, making it suitable for the input requirements of subsequent processing models. Furthermore, preprocessing operations may include data format conversion. For instance, if the image processing component can only process raw image data, and a component provides an image address, the preprocessing component can retrieve image metadata from that address and store it in the data pipeline for use by the image processing component. Additionally, preprocessing operations may include data format conversion. For example, converting the image to a NumPy ndarray (multidimensional array) format and storing it in the data pipeline for use by subsequent components. Whether to convert the format can be flexibly determined by the driver based on the input requirements of the target component; this disclosure does not limit this.
[0070] Post-processing components are used to perform post-processing operations on the input data.
[0071] Post-processing operations are often based on the data results obtained in the previous stages, and further optimize, transform, or extract more valuable information from them.
[0072] Taking post-processing of image processing models as an example, post-processing refers to a series of optimization and improvement operations performed on the output results after the model has completed its main image analysis task. The purpose of post-processing is to improve the accuracy and stability of the detection results and improve the final output quality. This may include non-maximum suppression (NMS), confidence filtering, coordinate correction, and class probability transformation.
[0073] Confidence filtering refers to filtering out detection results with lower scores based on a confidence threshold, and only retaining those detection boxes with high confidence.
[0074] Coordinate correction refers to mapping the coordinates of the detection boxes output by the model back to the original image size, because the model usually makes predictions on scaled or resized images.
[0075] Category probability transformation refers to converting the class probabilities output by the model into specific class labels, usually by selecting the class with the highest probability as the final class prediction.
[0076] In this embodiment, the preprocessing component performs preprocessing operations on the input data, optimizing and standardizing the data to suit different processing needs. The postprocessing component performs postprocessing operations on the input data, further optimizing and improving the model output data for specific application scenarios. The combination of the preprocessing and postprocessing components enables flexible model development, improving development efficiency and convenience.
[0077] (2) Predictive model component, which encapsulates multiple models to complete a specific task.
[0078] In practice, these multiple models perform the same specific task. Completing this same specific task requires combining multiple data processing operations.
[0079] For example, multiple neural network models for image classification tasks can be encapsulated within a single predictive model component. Developers can then determine which model to use to build the processing workflow based on their needs.
[0080] Therefore, it can also be understood that a prediction model component defines multiple basic components.
[0081] (3) Model production line component, used to call multiple prediction model components in a preset execution order.
[0082] The model production line component is an end-to-end predictive inference system composed of multiple predictive model components.
[0083] In this embodiment, multiple data processing operations are encapsulated into predictive model components, which can independently complete a single target task, making them easier to manage and use. The model production line calls multiple predictive model components in a preset execution order, enabling more complex predictive logic and solving more complex problems. By providing candidate components of different granularities, the required processing flow can be flexibly organized according to the purpose, improving the development efficiency of multi-model combinations.
[0084] To accommodate more task requirements, such as Figure 3 As shown, the multi-model combination development platform provided in this embodiment of the disclosure also includes a data sampler 104, which can support a variety of input data types, such as image files, audio files, text files, video streams, audio streams, and other streaming data.
[0085] The data sampler is used to sample the input data to obtain at least one batch of data, convert the at least one batch of data into the target data type, and input it into the components of the processing flow in batches until sampling is complete. If required, the data sampler can also perform validity checks on the input data of the target object, and if the checks pass, sample the input data to obtain at least one batch of data.
[0086] For example, for image files, it checks whether the image resolution, size, color mode, etc., meet preset standards, and whether the image file is complete and undamaged; for audio files, it checks whether the audio sampling rate, number of channels, duration, etc., are compliant, and whether there are any abnormalities such as truncation or excessive noise in the audio data; for text files, it checks whether the text encoding format is correct, whether there are any garbled characters or other situations that may affect subsequent processing, and whether the text content meets the basic requirements such as the corresponding syntax and semantic rules.
[0087] If the detection passes, the input data is sampled to obtain at least one batch of data, which is then converted into a target data type that meets the requirements of the target component. Afterwards, all the necessary data is delivered in batches to the corresponding target components in the processing flow according to the set sampling rules.
[0088] In addition, when needed, the data sampler can convert at least one batch of data into the target data type and then input it into the corresponding target component in the processing flow.
[0089] In this embodiment, a data sampler is used to sample various data types of input and perform batch input operations, eliminating the need for developers to code the data sampling process and thus improving development efficiency. Furthermore, the encapsulated data sampler is compatible with the API interface of the target component, providing a high-quality, compliant data foundation for subsequent processing. This ensures the accuracy and reliability of the entire data processing flow, enabling its application in different fields and scenarios.
[0090] In addition, such as Figure 3 As shown, the multi-model combination development platform provided in this embodiment of the disclosure also includes a result wrapper 105 for providing at least one specific operation for the output data of the processing flow.
[0091] Results wrappers are used to wrap the outputs of prediction models and model production lines, thereby providing users with a rich and unified interface, such as accessing, printing, persistent storage, visualization, and serialization of prediction results.
[0092] In summary, based on the design of the above target components, the multi-model combination development platform provided in this disclosure embodiment can support the execution logic of model serialization and / or parallelization in a low-code (i.e., low-code) manner. Based on modular design, the tool provides good scalability and supports user-defined implementation of multi-model combination, thereby solving complex practical problems.
[0093] Based on the same concept, this disclosure also provides a multi-model combination development method, which can be applied to the multi-model combination development platform described above. Specifically, as follows... Figure 4 As shown, it includes the following:
[0094] S401, in response to the association between the input and output interfaces of multiple target components defined by encoding, determines the connection relationship between the multiple target components; wherein, the multiple target components are components among multiple candidate components provided by the multi-model combination development platform, and the multiple candidate components have a unified application development interface (API).
[0095] That is, based on the association between the input and output interfaces of the multiple target components defined in the multi-model combination development platform described above, the connection relationship between the multiple target components is determined.
[0096] S402, in response to the operation of adding multiple target components to the driver provided by the multi-model combination development platform, a processing flow based on at least one neural network model is built based on the driver; the at least one neural network model is determined based on multiple target components.
[0097] That is, the target component can be used to invoke the neural network model for inference operations, so at least one neural network model can be determined based on the target component.
[0098] After determining the connection relationships between multiple target components, the target components are added to the instantiated driver, and the driver schedules the multiple target components to process the data sequentially according to the processing flow.
[0099] In this embodiment, based on a unified application programming interface (API), the connection relationships between target components are determined by defining component interface relationships through coding. Then, with the help of a driver, a neural network model processing flow is built based on the target components. This allows for the flexible combination of different target components to perform diverse tasks with minimal code. Therefore, this embodiment can improve the development efficiency of multi-model combinations to meet users' needs for diverse data processing and model applications.
[0100] In this disclosure, in response to the association between input and output interfaces of multiple target components defined by encoding, the connection relationship between the multiple target components can be determined as follows: Figure 5 As shown, it includes:
[0101] S501, in response to the operation of assigning component names of multiple target components to multiple custom component names through encoding, determines the mapping relationship between multiple custom component names and multiple target components.
[0102] For example, define target components A, B, and C as "ProcessorA", "ProcessorB", and "ProcessorC" respectively. Then, create three custom component names through coding, such as "processor-a", "processor-b", and "processor-c". Next, assign the target component name "ProcessorA" to "processor-a", "ProcessorB" to "processor-b", and "ProcessorC" to "processor-c". This establishes the mapping relationship between multiple custom component names and multiple target components.
[0103] S502, for any target component name among a plurality of custom component names, in response to the operation of assigning the output stream of at least one custom component name among the plurality of custom component names to the input stream of the target component name, determine the connection relationship between the target component corresponding to the target component name and the target component of at least one custom component name.
[0104] That is, based on the established mapping relationship between custom component names and target components, the connection relationship between the target component corresponding to the target component name and the target components corresponding to those custom component names is determined by assigning the output stream of at least one other custom component name to the input stream of the target component name.
[0105] like Figure 2 As shown, the output stream of target component A can be used as the input stream of target component B, and the output streams of target component A and target component B can be used as the input streams of target component C. The connection relationship between each target component can be determined by defining the relationship between the input streams and the output streams through a unified API interface.
[0106] In this embodiment, the mapping relationship between target components and custom component names is determined, and then the connection relationship between each target component is determined through encoding. This provides an orderly and flexible method, offering an effective means of building complex data processing workflows. Furthermore, defining the relationship between input and output streams through a unified API interface can effectively reduce the amount of code required for development and improve the efficiency of developing multi-model combinations.
[0107] In this embodiment of the disclosure, as described above, the input streams and output streams of the multiple target components are pipe objects, wherein the pipe object of the input stream is used to obtain the input stream from the data pipe, and the pipe object of the output stream is used to store the output stream into the data pipe.
[0108] In this embodiment of the disclosure, the input and output streams of the candidate components are designed as pipeline objects and a tight interaction mechanism is established with the data pipeline, so that the entire data processing flow can be carried out smoothly and continuously, avoiding the risk of data transmission chaos and data loss, while also improving data reusability and the flexibility and scalability of the entire processing flow.
[0109] In this embodiment of the disclosure, for the processing flow of the neural network model, in response to the operation of running the processing flow, a directed acyclic graph is constructed based on the processing flow by a driver. In the directed acyclic graph, nodes are multiple target components and edges represent data flows between different target components. The driver schedules multiple target components to process data sequentially according to the directed acyclic graph.
[0110] Based on the preceding explanation, the driver schedules multiple target components to process data sequentially according to a processing flow, which can be achieved by constructing a directed acyclic graph (DAG). After successfully constructing the DAG, the driver uses the graph's structure to schedule multiple target components to process data sequentially, ensuring the entire process is orderly and follows a strict logical order. Figure 2As shown, the driver starts with target component A as the starting node of the directed acyclic graph (DAG), and sequentially activates the target components represented by each node according to the DAG. Specifically, for the components to be run in the current processing progress, data is obtained from the data pipeline according to the input stream required by the component to be run, and input to the component to be run. Then, the output stream of the component to be run is stored in the data pipeline according to the pipeline object of the component to be run.
[0111] In this embodiment, by scheduling multiple target components to process data sequentially according to a directed acyclic graph (DAG) using a driver, the structural characteristics of the DAG can be utilized to effectively avoid potential chaos in the data processing flow, ensuring the accuracy, stability, and efficiency of the entire data processing process, and providing a reliable guarantee for achieving complex data processing tasks. Furthermore, using the driver provided by the multi-model combination development platform, only the target components need to be defined in the driver to start running, without requiring additional code development, thus improving development efficiency.
[0112] In addition, as described above, during the execution of the driver's processing flow, the processing progress can be monitored based on the directed acyclic graph; and the remaining processing flow can be determined based on the processing progress.
[0113] If there is residual data in the data pipeline that is not needed in the processing flow and needs to be cleaned up, remove the data from the data pipeline.
[0114] In summary, by using directed acyclic graphs to monitor the progress of the processing flow, determine the remaining processing flow, and clean up useless data in the data pipeline, a closely linked and interconnected data management and process monitoring system is formed, which helps to improve the operational efficiency and resource utilization efficiency of the entire data processing flow.
[0115] In this embodiment, the raw input data is processed by a data sampler provided by a multi-model combination development platform. This data sampler supports various input data types, such as image files, audio files, text files, streaming data like video streams and audio streams.
[0116] The data sampler is used to sample the input data to obtain at least one batch of data, convert the at least one batch of data into the target data type, and input it into the components of the processing flow in batches until sampling is complete. If required, the data sampler can also perform validity checks on the input data of the target object, and if the checks pass, sample the input data to obtain at least one batch of data.
[0117] In addition, when needed, the data sampler can convert at least one batch of data into the target data type and then input it into the corresponding target component in the processing flow.
[0118] In this embodiment, a data sampler is used to sample various data types of input and perform batch input operations, eliminating the need for developers to code the data sampling process and thus improving development efficiency. Furthermore, the encapsulated data sampler is compatible with the API interface of the target component, providing a high-quality, compliant data foundation for subsequent processing. This ensures the accuracy and reliability of the entire data processing flow, enabling its application in different fields and scenarios.
[0119] In this embodiment of the disclosure, for the final output data, a specific operation to be performed on the output data can be determined in response to the operation of the output data association result wrapper that uses the encoding method as the processing flow.
[0120] The specific operation includes at least one of the following:
[0121] (1) Access the output data;
[0122] This means acquiring the generated output data so that it can be viewed, further processed, or used.
[0123] (2) Print out the data;
[0124] The output data will be presented in a visual and easily viewable format.
[0125] (3) Persistently store the output data;
[0126] The output data will be saved to a storage medium that can retain it for a long time.
[0127] (4) Visualize the output data;
[0128] (5) Serialize the output data.
[0129] The output data structure will be converted into a format that can be stored or transmitted.
[0130] By performing the specific operations described above on the output data, the output data can be viewed, displayed, stored long-term, intuitively understood, and easily transmitted and reused, thus meeting the data utilization needs of different application scenarios in many ways.
[0131] In this embodiment of the disclosure, in response to the operation of the output data association result wrapper that uses the encoding method as the processing flow, the output can be packaged, thereby providing users with a rich and unified interface, which is helpful for data management.
[0132] In this disclosure embodiment, the candidate components described above include at least one of the following types of components:
[0133] (1) Basic components, which encapsulate the smallest unit of data processing operations; the basic components include multiple preprocessing components and / or multiple postprocessing components, wherein:
[0134] Preprocessing components are used to perform preprocessing operations on the input data;
[0135] Post-processing components are used to perform post-processing operations on the input data.
[0136] In this embodiment, the preprocessing component performs preprocessing operations on the input data, which optimizes and standardizes the raw input data. The post-processing component performs post-processing operations on the input data, further meeting the requirements for the final presentation format and quality of the data in specific application scenarios. The combination of the preprocessing and post-processing components enables the entire data processing flow to effectively transform raw input data into high-quality data results that meet application needs.
[0137] (2) Predictive model components, which encapsulate multiple data processing operations and can independently complete a single target task;
[0138] (3) Model production line component, used to call multiple prediction model components in a preset execution order.
[0139] In this embodiment, multiple data processing operations are encapsulated into predictive model components, which can independently complete a single target task, making them easier to manage and use. The model production line calls multiple predictive model components in a preset execution order, enabling the implementation of more complex predictive logic and the solution of more complex problems.
[0140] In summary, taking a process involving multiple target components as an example, combined with... Figure 6 The present disclosure provides a data pipeline, driver, and target component according to embodiments thereof.
[0141] like Figure 6 As shown, the constructed processing flow includes prediction model component 1 and prediction model component 2. The driver schedules each component to perform data processing tasks according to the processing flow. The output of prediction model component 1 is provided to prediction model component 2.
[0142] by Figure 6 Taking prediction model component 1 as an example, it includes target component A, target component B, target component C, and target component D. The output stream of target component A is provided to target components B and C. The output stream of target component C is provided to target component D. The output stream of target component D is provided to prediction model component 2.
[0143] like Figure 6 As shown, target component A processes the input stream from the data pipeline and then stores its output stream in the data pipeline. Subsequently, target components B and C can both obtain target component A's output stream from the data pipeline as input streams for data processing. After processing the data and obtaining its output stream, target component B stores it in the data pipeline (not shown in the diagram) for use by target component C. After obtaining target component C's output stream, its output data is stored in the data pipeline for use by target component D.
[0144] The reading and writing of inputs and the operation of each target component can be scheduled by a driver.
[0145] like Figure 7 The diagram illustrates a processing flow built around an image processing scenario, scheduling and running various target components. The designed neural network model can include an image classification model, an OCR (Optical Character Recognition) model, and an OCR detection model. The processing flow includes:
[0146] S701 inputs the input data to the preprocessing component in batches through the data sampler.
[0147] S702 schedules the preprocessing component to perform preprocessing operations through the preprocessing component interface, and the preprocessing results are stored in the data pipeline.
[0148] For example, preprocessing components can be used to transform the format of the raw input data or perform cropping operations.
[0149] S703 schedules the image classification model through the model scheduling interface to obtain preprocessing results from the data pipeline, and performs image classification operations on the preprocessing results to obtain the classification results.
[0150] For example, whether the skewness of the classified image is 90° or some other degree.
[0151] S704 schedules the post-processing component to perform image correction based on the classification results through the post-processing component interface, and stores the correction results in the data pipeline.
[0152] S705 schedules the OCR detection model through the model scheduling interface to detect the correction results and obtain the detection results.
[0153] S706 schedules an OCR recognition model through a model scheduling interface to perform OCR recognition on the detection results and outputs the recognition results.
[0154] like Figure 8As shown, taking the extraction of key information from an image as an example, the process developed using a multi-model combination development platform in this embodiment of the disclosure is described, including the following:
[0155] S801, obtain the query statement (Query) provided by the target object.
[0156] For example, a request could be made to retrieve the content of a seal and the date information from an image.
[0157] S802 takes the image to be identified as a data file, aligns it, performs layout analysis, and obtains the analysis results. The layout analysis includes the following steps:
[0158] S8021 uses an image orientation classification model to classify images and obtain classification results.
[0159] S8022 uses a text image correction model to correct images based on classification results, resulting in a corrected image.
[0160] S8023 uses a panel area detection model to detect different regions in the corrected image, including stamp regions, text regions, and table regions.
[0161] S8024, The seal area is detected using a seal text detection component to obtain seal information.
[0162] S8025 uses a text detection component to detect the text region and obtain the text content.
[0163] S8026, The table recognition component is used to detect the table area to obtain the table content.
[0164] S8027 summarizes the seal information, text content, and table content using a summarization component to obtain the text recognition result.
[0165] S803, construct a retrieval vector based on the obtained user query.
[0166] S804 generates prompt words by calling the Prompt generator, which uses the text recognition results obtained from the retrieval vector and version analysis.
[0167] The S805 uses a large language model to process prompt words, extract key information, and output it.
[0168] Based on the same technical concept, this disclosure also proposes a multi-model combination development device 900, such as... Figure 9 As shown, it includes:
[0169] The determination module 901 is used to determine the connection relationship between multiple target components in response to the association relationship between the input and output interfaces of multiple target components defined by encoding; wherein, the multiple target components are components of multiple candidate components provided by the multi-model combination development platform, and the multiple candidate components have a unified application programming interface (API);
[0170] Module 902 is used to respond to the operation of adding multiple target components to the driver provided by the multi-model combination development platform, and build a processing flow based on at least one neural network model based on the driver; the at least one neural network model is determined based on multiple target components.
[0171] In some embodiments, such as Figure 10 As shown, module 901 includes:
[0172] The relationship determination unit 9011 is used to determine the mapping relationship between multiple custom component names and multiple target components in response to the operation of assigning component names of multiple target components to multiple custom component names by encoding.
[0173] The connection unit 9012 is used to determine the connection relationship between the target component corresponding to the target component name and the target component of at least one custom component name, in response to the operation of assigning the output stream of at least one custom component name to the input stream of the target component name, for any target component name among a plurality of custom component names.
[0174] In some embodiments, the input streams and output streams of the multiple target components are pipeline objects, wherein the pipeline object of the input stream is used to obtain the input stream from the data pipeline, and the pipeline object of the output stream is used to store the output stream into the data pipeline.
[0175] In some embodiments, such as Figure 10 As shown, it also includes a processing module 903, used for:
[0176] In response to the operation of running the processing flow, a directed acyclic graph is constructed based on the processing flow through the driver. In the directed acyclic graph, nodes are multiple target components and edges represent data flows between different target components.
[0177] The driver schedules multiple target components to process data sequentially according to a directed acyclic graph.
[0178] In some embodiments, such as Figure 10 As shown, it also includes a clearing module 904, used for:
[0179] The processing progress is monitored based on the directed acyclic graph.
[0180] Determine the remaining processing steps based on the processing progress;
[0181] If there is residual data in the data pipeline that is not needed in the processing flow and needs to be cleaned up, remove the data from the data pipeline.
[0182] In some embodiments, such as Figure 10 As shown, it also includes a sampling module 905, used for:
[0183] The data sampler provided by the multi-model combination development platform samples the input data of the target object to obtain at least one batch of data, and inputs it into the components in the processing flow in batches until the sampling is completed.
[0184] In some embodiments, such as Figure 10 As shown, it also includes an operation module 906, used for:
[0185] In response to the operation of associating the output data of the processing flow with a result wrapper using an encoded method, determine the specific operation to be performed on the output data.
[0186] In some embodiments, a particular operation includes at least one of the following:
[0187] Access output data, print output data, persist output data, visualize output data, and serialize output data.
[0188] In some embodiments, the plurality of candidate components includes at least one of the following types of components:
[0189] Basic components encapsulate the smallest unit of data processing operations;
[0190] Predictive model components encapsulate multiple models that perform a specific task;
[0191] The model production line component is used to call multiple prediction model components in a preset execution order.
[0192] In some embodiments, the base components include multiple preprocessing components and / or multiple postprocessing components, wherein:
[0193] Preprocessing components are used to perform preprocessing operations on the input data;
[0194] Post-processing components are used to perform post-processing operations on the input data.
[0195] The specific functions and examples of each module and submodule of the apparatus in this disclosure can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.
[0196] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0197] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.
[0198] Figure 11 A schematic block diagram of an example electronic device 1100 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0199] like Figure 11 As shown, device 1100 includes a computing unit 1101, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1102 or a computer program loaded from storage unit 1108 into random access memory (RAM) 1103. The RAM 1103 may also store various programs and data required for the operation of device 1100. The computing unit 1101, ROM 1102, and RAM 1103 are interconnected via bus 1104. Input / output (I / O) interface 1105 is also connected to bus 1104.
[0200] Multiple components in device 1100 are connected to I / O interface 1105, including: input unit 1106, such as keyboard, mouse, etc.; output unit 1107, such as various types of monitors, speakers, etc.; storage unit 1108, such as disk, optical disk, etc.; and communication unit 1109, such as network card, modem, wireless transceiver, etc. Communication unit 1109 allows device 1100 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0201] The computing unit 1101 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1101 performs the various methods and processes described above, such as the multi-model combination development method. For example, in some embodiments, the quantization perception training method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1108. In some embodiments, part or all of the computer program can be loaded and / or installed on device 1100 via ROM 1102 and / or communication unit 1109. When the computer program is loaded into RAM 1103 and executed by the computing unit 1101, one or more steps of the quantization multi-model combination development method described above can be performed. Alternatively, in other embodiments, computing unit 1101 may be configured to perform a multi-model combination development method by any other suitable means (e.g., by means of firmware).
[0202] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0203] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0204] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0205] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0206] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.
[0207] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.
[0208] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.
[0209] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A multi-model combination development platform system, comprising: The component collection includes multiple candidate components, each of which encapsulates corresponding data processing operations; The multiple candidate components share a unified application programming interface (API) to support the construction of a processing flow based on at least one neural network model by defining the relationships between the input and output interfaces of different target components within the multiple candidate components. The multiple candidate components include candidate components that support no input and / or no output. The at least one neural network model is determined based on multiple target components. No input means data is retrieved from GPU memory, and no output means generated data is stored in GPU memory. Candidate components with no output include a first component that copies data from memory to GPU memory, and candidate components with no input include a second component that copies data from GPU memory to memory. The first and second components process data in the form of pointers through the inference engine's interface. The multiple candidate components include basic components, predictive model components, and / or model production line components. The basic components encapsulate the smallest unit of data processing operations. The predictive model components encapsulate multiple models that complete specific tasks. The model production line components are used to call multiple predictive model components in a preset execution order. The candidate components support inference components with no input and no output. A driver is used to construct a directed acyclic graph based on the processing flow, wherein nodes in the directed acyclic graph are the multiple target components and edges represent data flows transmitted between different target components; Based on the directed acyclic graph, the multiple target components are scheduled to process the data sequentially according to the processing flow; A data pipeline is used to provide data to the multiple target components and store the data involved in the processing flow.
2. The system according to claim 1, wherein, The input and output streams of the plurality of candidate components are pipeline objects, wherein the pipeline object of the input stream is used to obtain the input stream from the data pipeline, and the pipeline object of the output stream is used to store the output stream into the data pipeline.
3. The system according to claim 1, wherein, The basic components include multiple preprocessing components and / or multiple postprocessing components, wherein: The preprocessing component is used to perform preprocessing operations on the input data; The post-processing component is used to perform post-processing operations on the input data.
4. The system according to any one of claims 1-3, wherein, The data pipeline is also used to remove data that is no longer needed in the remaining processing flow from the data pipeline based on the processing progress of the processing flow.
5. The system according to claim 1, further comprising: Data sampler, supporting multiple input data types; The data sampler is used to sample the input data of the target object to obtain at least one batch of data, and input the data in batches to the components in the processing flow until the sampling is completed.
6. The system according to claim 1, further comprising: Result wrapper for providing at least one specific operation for the output data of the processing flow.
7. A multi-model combined development method, comprising: In response to the association between input and output interfaces of multiple target components defined by encoding, the connection relationship between the multiple target components is determined; wherein, the multiple target components are components among multiple candidate components provided by the multi-model combination development platform system, and the multiple candidate components have a unified application programming interface (API); the multiple candidate components include candidate components that support no input and / or no output; no input means data is obtained from video memory, and no output means generated data is stored in video memory; the candidate components with no output include a first component that copies data from memory to video memory, and the candidate components with no input include a second component that copies data from video memory to memory; the first and second components process data in the form of pointers through the inference engine interface; the multiple candidate components include basic components, predictive model components, and / or model production line components; the basic components encapsulate the smallest unit of data processing operations; the predictive model components encapsulate multiple models that complete specific tasks; the model production line components are used to call multiple predictive model components according to a preset execution order; the candidate components support inference components with no input and no output. In response to the operation of adding the plurality of target components to the driver provided by the multi-model combination development platform system, a processing flow based on at least one neural network model is constructed based on the driver; the at least one neural network model is determined based on the plurality of target components; In response to the operation of running the processing flow, a directed acyclic graph is constructed by the driver based on the processing flow, in which nodes are the multiple target components and edges represent data flows between different target components; and the multiple target components are scheduled to process data sequentially according to the directed acyclic graph.
8. The method according to claim 7, wherein, The step of determining the connection relationship between the multiple target components in response to the association between the input and output interfaces of the multiple target components defined by encoding includes: In response to the operation of assigning the component names of the plurality of target components to the plurality of custom component names through encoding, the mapping relationship between the plurality of custom component names and the plurality of target components is determined; For any target component name among the plurality of custom component names, in response to the operation of assigning the output stream of at least one of the plurality of custom component names to the input stream of the target component name, the connection relationship between the target component corresponding to the target component name and the target component of the at least one custom component name is determined.
9. The method according to claim 7, wherein, The input and output streams of the multiple target components are pipe objects, wherein the pipe object of the input stream is used to obtain the input stream from the data pipe, and the pipe object of the output stream is used to store the output stream into the data pipe.
10. The method of claim 7, further comprising: The processing progress of the process is monitored based on the directed acyclic graph. The remaining processing steps are determined based on the processing progress. If there is data to be cleaned that is not needed in the remaining processing flow in the data pipeline, the data to be cleaned is removed from the data pipeline.
11. The method of claim 7, further comprising: The data sampler provided by the multi-model combination development platform system samples the input data of the target object to obtain at least one batch of data, and inputs it into the components in the processing flow in batches until the sampling is completed.
12. The method according to any one of claims 7-9, further comprising: In response to the operation of associating a result wrapper with the output data of the processing flow using an encoded method, a specific operation to be performed on the output data is determined.
13. The method according to claim 12, wherein, The specific operation includes at least one of the following: Access the output data, print the output data, persist the output data, visualize the output data, and serialize the output data.
14. The method according to claim 7, wherein, The basic components include multiple preprocessing components and / or multiple postprocessing components, wherein: The preprocessing component is used to perform preprocessing operations on the input data; The post-processing component is used to perform post-processing operations on the input data.
15. A multi-model combination development apparatus, comprising: A determination module is used to determine the connection relationships between multiple target components in response to the association relationships between input and output interfaces defined by encoding. These multiple target components are components among multiple candidate components provided by a multi-model combination development platform system, and these candidate components have a unified application programming interface (API). Among the multiple candidate components are candidate components that support no input and / or no output. No input means data is retrieved from video memory, and no output means generated data is stored in video memory. Candidate components with no output include a first component that copies data from memory to video memory, and candidate components with no input include a second component that copies data from video memory to memory. The first and second components process data in the form of pointers through the inference engine's interface. The multiple candidate components include basic components, predictive model components, and / or model production line components. The basic components encapsulate the smallest unit of data processing operations. The predictive model components encapsulate multiple models that complete specific tasks. The model production line components are used to call multiple predictive model components according to a preset execution order. The candidate components support inference components with no input and no output. A module is configured to respond to an operation of adding the plurality of target components to a driver provided by the multi-model combination development platform system, and to build a processing flow based on at least one neural network model based on the driver; the at least one neural network model is determined based on the plurality of target components; It also includes a processing module for: In response to the operation of running the processing flow, a directed acyclic graph is constructed by the driver based on the processing flow, in which nodes are the multiple target components and edges represent data flows between different target components; and the multiple target components are scheduled to process data sequentially according to the directed acyclic graph.
16. The apparatus according to claim 15, wherein, The determining module includes: A relationship determination unit is used to determine the mapping relationship between the multiple custom component names and the multiple target components in response to the operation of assigning the component names of the multiple target components to multiple custom component names by encoding. A connection unit is configured to, for any target component name among the plurality of custom component names, determine the connection relationship between the target component corresponding to the target component name and the target component of the at least one custom component name in response to an operation of assigning the output stream of at least one custom component name among the plurality of custom component names to the input stream of the target component name.
17. The apparatus according to claim 15, wherein, The input and output streams of the multiple target components are pipe objects, wherein the pipe object of the input stream is used to obtain the input stream from the data pipe, and the pipe object of the output stream is used to store the output stream into the data pipe.
18. The apparatus of claim 15, further comprising a cleaning module for: The processing progress of the process is monitored based on the directed acyclic graph. The remaining processing steps are determined based on the processing progress. If there is data to be cleaned that is not needed in the remaining processing flow in the data pipeline, the data to be cleaned is removed from the data pipeline.
19. The apparatus of claim 15, further comprising a sampling module, configured to: The data sampler provided by the multi-model combination development platform system samples the input data of the target object to obtain at least one batch of data, and inputs it into the components in the processing flow in batches until the sampling is completed.
20. The apparatus according to any one of claims 15-17, further comprising an operation module for: In response to the operation of associating a result wrapper with the output data of the processing flow using an encoded method, a specific operation to be performed on the output data is determined.
21. The apparatus according to claim 20, wherein, The specific operation includes at least one of the following: Access the output data, print the output data, persist the output data, visualize the output data, and serialize the output data.
22. The apparatus according to claim 15, wherein, The basic components include multiple preprocessing components and / or multiple postprocessing components, wherein: The preprocessing component is used to perform preprocessing operations on the input data; The post-processing component is used to perform post-processing operations on the input data.
23. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 7-14.
24. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 7-14.
25. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 7-14.
Citation Information
Patent Citations
Machine learning method and device based on assembly line, electronic equipment and storage medium
CN115904359A
Back-end extensible framework and method for deep learning model parsing, optimization and deployment
CN116306812A
Private computing job processing method and device, equipment, medium and product
CN117633876A
Configurable algorithm component packaging and combinatorial modeling method
CN118428457A
Convolutional neural network acceleration device based on pipeline design
CN118536561A
