Model resource processing method, device, computer equipment and storage medium

Abstract

The present application discloses a method, device, computer equipment and storage medium for processing model resources, the method comprising: obtaining an initial version model and multiple target version data sets; performing multiple training operations on the initial version model based on the multiple target version data sets to obtain a new version model corresponding to each training operation; renaming the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain a second model name of the new version model; generating a target model pedigree between the initial version model, the target version data set and the new version model according to the first model name, the data set name and the second model name. This embodiment can use the pedigree to intuitively display the relationship between models and between models and data sets, which is convenient for tracing back to the previous version model and realizing effective management of models and data sets.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular to a method, device, computer equipment and storage medium for processing model resources. Background Art

[0002] As actual business expands, it is necessary to continuously update training data sets and retrain AI models to improve model performance or adapt models to new scenarios. However, as data sets and models are continuously updated, the increase in data volume can easily lead to confusion between data sets and models, so data sets and models need to be managed.

[0003] At present, the common method of managing datasets and models is to establish a strong association between the current version of the dataset and the model, so as to avoid confusion between the current version of the dataset and the model. However, in the case of poor model training results and the need for version backtracking, the existing methods are difficult to quickly backtrack to the correct version of the model, and there is still the problem of easy confusion between the dataset and the model, which has an adverse impact on the efficiency of model optimization. Summary of the invention

[0004] The present application provides a method, apparatus, computer device and storage medium for processing model resources to solve the technical problem that data sets and models are easily confused.

[0005] In order to solve the above technical problems, the present application embodiment provides a method for processing model resources, including:

[0006] Obtain an initial version model and multiple target version datasets;

[0007] Based on multiple target version data sets, the initial version model is trained multiple times to obtain a new version model corresponding to each training operation;

[0008] Rename the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain a second model name of the new version model;

[0009] A target model genealogy between the initial version model, the target version dataset and the new version model is generated according to the first model name, the data set name and the second model name.

[0010] This embodiment uses the target version data set to perform multiple training operations on the initial version model to obtain a new version model for each training operation to determine the training relationship between models; and renames the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain the second model name of the new version model, so that the second model name of the new version model has the model information of the previous version model and the data set information of the target version data set, and then the relationship between the new version model and the previous version model and the target version data set can be obtained according to the second model name; finally, according to the first model name, the data set name and the second model name, a target model pedigree between the initial version model, the target version data set and the new version model is generated, so that the pedigree can be used to intuitively display the relationship between models and between models and data sets, so that the relationship between data sets and models is clear, so that it is convenient to query the corresponding previous version model according to the pedigree relationship when the model training effect is not good, and trace back to the previous version model to achieve effective management of models and data sets.

[0011] In one embodiment, based on multiple target version data sets, multiple training operations are performed on the initial version model to obtain a new version model corresponding to each training operation, including:

[0012] The initial version model is used as the prior version model, and a plurality of first data sets are used to perform a training operation on the prior version model to obtain a plurality of subsequent version models, wherein each first data set is one of the plurality of target version data sets;

[0013] The later version model is used as the new earlier version model, and multiple second data sets are used to train the new earlier version model to obtain multiple new later version models, each second data set is one of the multiple target version data sets, and the new version model includes all the later version models.

[0014] In this embodiment, the initial version model is used as the prior version model, a data set is selected to train the initial version model, and a subsequent version model is obtained. The subsequent version model is then used as a new prior version model, and a data set is selected to train the prior version model. This is repeated over and over again to obtain multiple new version models, so as to improve the selectivity of the model during model optimization training, thereby training a model that meets the conditions to the maximum extent possible.

[0015] In one embodiment, the new version model is renamed according to the first model name of the initial version model and the data set name of the target version data set to obtain the second model name of the new version model, including:

[0016] Determine a first version number of the initial version model according to the first model name of the initial version model;

[0017] Determine the second version number of the target version data set according to the data set name of the target version data set;

[0018] The first version number and the second version number are combined to obtain a second model name.

[0019] This embodiment determines the first version number according to the first model name, and determines the second version number according to the data set name, so that the second model name of the new version model can have the model version and data set version of the previous version model, thereby facilitating the establishment of a relationship with the previous version model or the target version data set based only on the second model name, making the model relationship more intuitive and clear.

[0020] In one embodiment, generating a target model genealogy between an initial version model, a target version data set, and a new version model according to a first model name, a data set name, and a second model name includes:

[0021] Determine a first model version relationship between the initial version model and the new version model according to the first model name and the second model name;

[0022] Determine, according to the second model name, a second model version relationship between the multiple new version models;

[0023] Determine the model data relationship between the target version data set and the new version model according to the data set name and the second model name;

[0024] A target model pedigree is established based on the first model version relationship, the second model version relationship and the model data relationship.

[0025] This embodiment determines model relationships according to different model names, determines model-data relationships according to different model names and dataset names, and determines dataset relationships according to different dataset names, so as to construct a model genealogy according to model relationships, model-data relationships, and dataset relationships, thereby facilitating the intuitive display of the three relationships, making the model and dataset clear during the model training process to avoid confusion, and facilitating the adjustment of the model version or dataset version during model optimization training.

[0026] In one embodiment, before obtaining the initial version model and the plurality of target version data sets, the method further includes:

[0027] The initial version data set is used as the previous version data set, and a change operation is performed on the previous version data set to obtain the subsequent version data set. The change operation includes an add operation, a delete operation, or a modify operation.

[0028] The later version data set is used as the new earlier version data set, and the new earlier version data set is changed to obtain a new later version data set. The target version data set includes the initial version data set and all later version data sets.

[0029] This embodiment obtains multiple target version data sets by performing a change operation on the previous version data set to expand the initial version data set, thereby using multiple version data sets to optimize the model and improve the model optimization effect.

[0030] In one embodiment, after the later version data set is used as the new earlier version data set and the new earlier version data set is modified to obtain the new later version data set, the following steps are further included:

[0031] According to the first name of the previous version data set, rename the later version data set to obtain the second name of the later version data set;

[0032] A change sequence between the previous version data set and the later version data set is determined according to the change operation, the first name and the second name, where the change sequence is used to represent an inheritance relationship or a transition relationship between the previous version data set and the later version data set.

[0033] This embodiment determines the relationship between the data sets by changing the sequence and the second name, so that when the model is subsequently optimized and trained, the data set version can be adjusted according to the training results to improve the model optimization effect.

[0034] In one embodiment, based on multiple target version data sets, multiple training operations are performed on the initial version model to obtain a new version model corresponding to each training operation, and the method further includes:

[0035] Using the preset test data set, multiple new version models are tested to obtain the performance indicators of each new version model.

[0036] This embodiment tests the new version models to compare the performance indicators of each new version model, making the model optimization effect more intuitive.

[0037] In a second aspect, an embodiment of the present application provides a model resource processing device, including:

[0038] An acquisition module is used to acquire an initial version model and multiple target version data sets;

[0039] A training module is used to perform multiple training operations on the initial version model based on multiple target version data sets to obtain a new version model corresponding to each training operation;

[0040] A renaming module, used to rename the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain a second model name of the new version model;

[0041] A generation module is used to generate a target model pedigree between an initial version model, a target version data set and a new version model according to a first model name, a data set name and a second model name.

[0042] In a third aspect, an embodiment of the present application provides a computer device, including a processor and a memory, wherein the memory is used to store a computer program, and when the processor executes the computer program, the steps of the method for processing model resources of the first aspect are implemented.

[0043] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for processing model resources of the first aspect.

[0044] It should be noted that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flowchart of a method for processing model resources provided in an embodiment of the present application;

[0046] Figure 2 An example diagram of the target model pedigree provided in the embodiment of the present application;

[0047] Figure 3 A schematic diagram of the dataset pedigree provided for the embodiment of the present application;

[0048] Figure 4 A schematic diagram of the structure of a device for processing model resources provided in an embodiment of the present application;

[0049] Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0051] As described in the background technology, common methods for managing datasets and models mainly establish a strong association between the current version of the dataset and the model, thereby avoiding confusion between the current version of the dataset and the model. However, in the case where the model training effect is not good and version backtracking is required, the existing methods are difficult to quickly backtrack to the correct version of the model, and there is still the problem of easy confusion between the dataset and the model, which has an adverse effect on the efficiency of model optimization.

[0052] To this end, the embodiments of the present application provide a method, apparatus, computer equipment and storage medium for processing model resources. By adopting a target version data set, the initial version model is trained multiple times to obtain a new version model for each training operation to determine the training relationship between the models; and the new version model is renamed according to the first model name of the initial version model and the data set name of the target version data set to obtain a second model name of the new version model, so that the second model name of the new version model has the model information of the previous version model and the data set information of the target version data set, and then the relationship between the new version model and the previous version model and the target version data set can be obtained according to the second model name; finally, according to the first model name, the data set name and the second model name, a target model pedigree between the initial version model, the target version data set and the new version model is generated, so that the pedigree can be used to intuitively display the relationship between the models and between the models and the data sets, so that the relationship between the data sets and the models is clear, so that it is convenient to query the corresponding previous version model according to the pedigree relationship when the model training effect is not good, and trace back to the previous version model to achieve effective management of the model and the data set.

[0053] Please refer to Figure 1 , Figure 1 The flowchart of a method for processing model resources provided in an embodiment of the present application is shown below. The method for processing model resources in this embodiment can be applied to computer devices, including but not limited to computing devices such as smart phones, laptops, tablet computers, desktop computers, supercomputers, physical servers, and cloud servers. Figure 1 As shown, the method for processing model resources includes steps S101 to S104, which are described in detail as follows:

[0054] Step S101, obtaining an initial version model and multiple target version data sets.

[0055] In this step, the target version dataset is a dataset used to train the model. The target version dataset can be used to train the initial version model, or the target version dataset can be used to train other version models obtained by training the initial version model.

[0056] Optionally, the target version dataset is divided into a training set, a validation set, and a test set, where the training set serves as a data sample set for model fitting, the validation set is used to adjust the model's hyperparameters and perform a preliminary evaluation of the model's performance, and the test set is used to evaluate the generalization ability of the final model.

[0057] Optionally, the target version dataset is divided into a basic dataset and a difficult dataset. The basic dataset is a sample dataset composed of normal samples, which is relatively large in number; the difficult dataset is a sample dataset composed of difficult samples, which contains samples or extreme scenes that are difficult for the model to distinguish. For example, in face recognition, a normal face data sample should be similar to an ID photo, with the face facing the screen, clear facial features without obstruction, and moderate lighting, which is easy for the algorithm to recognize. However, for images with the face facing the screen in profile, or wearing a mask and sunglasses, or in dim lighting, which are not easy for the algorithm to recognize, such data are difficult samples.

[0058] Step S102: Based on the plurality of target version data sets, multiple training operations are performed on the initial version model to obtain a new version model corresponding to each training operation.

[0059] In this step, based on the multiple target version data sets, the initial version model is trained multiple times, including: using the target version data set to train the initial version model to obtain a secondary version model; using the target version data set to train the secondary version model to obtain a tertiary version model; continuing to use the version data set to train the tertiary version model, and so on, until the final version model is obtained. Among them, the secondary version model, the tertiary version model to the final version model are all new version models.

[0060] Step S103: rename the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain a second model name of the new version model.

[0061] In this step, the initial version model and the target version dataset are named in advance using the version number to obtain the first model name of the initial version model and the dataset name of the target version dataset. The new version model is renamed using the first model name and the dataset name, including: renaming the secondary version model using the first model name and the dataset name to obtain the secondary model name of the secondary version model; renaming the tertiary version model using the secondary model name and the dataset name to obtain the tertiary model name of the tertiary version model; continuing to rename the tertiary version model using the tertiary model name and the dataset name, and so on, until the final model name of the final version model is obtained. Among them, the secondary model name, the tertiary model name to the final model name are all the second model names, and the dataset name renamed each time should be consistent with the dataset name of the target version dataset used to train the corresponding version of the model.

[0062] Step S104: generating a target model pedigree between the initial version model, the target version data set and the new version model according to the first model name, the data set name and the second model name.

[0063] In this step, the target model pedigree is used to characterize the model relationship between models and the model data relationship between models and data sets. According to the first model name, the data set name and the second model name, the relationship between the three is determined, and the target model pedigree is generated based on the relationship between the three.

[0064] As dataset versions are constantly updated, models are also constantly updated. Generally, one dataset version corresponds to one or more model versions (different pre-trained models or hyperparameters). If the model versions are not effectively managed, when problems arise with the new model in actual business, it will be impossible to roll back to the old model due to confusion in the model version. Establishing pedigree management of the model is also conducive to a more intuitive understanding of the model optimization route.

[0065] In one embodiment, Figure 1 Based on the illustrated embodiment, step S102 specifically includes:

[0066] Using the initial version model as a prior version model, and using a plurality of first data sets to perform a training operation on the prior version model to obtain a plurality of subsequent version models, wherein each of the first data sets is one of the plurality of target version data sets;

[0067] The later version model is used as a new earlier version model, and a plurality of second data sets are used to perform training operations on the new earlier version model to obtain a plurality of new later version models, each of the second data sets being one of the plurality of target version data sets, and the new version model includes all the later version models.

[0068] In this embodiment, the target version data set is used to train the prior version model to obtain the subsequent version model, and the subsequent version model is used as the new prior version model for training. The first data set and the second data set may be the same or different, that is, the data set for training the previous prior version model and the data set for training the next prior version model may be the same or different. Exemplarily, the target version data set includes data sets D0 to D2, the initial version model is M0, and MO is obtained by training based on D0. Take M0 as the prior version model and use D1 for training to obtain M1-0; take M1-0 as the new prior version model and use D2 for training to obtain M2-1; take M2-1 as the new prior version model and continue to use D2 for training to obtain M2-2.

[0069] In one embodiment, Figure 1 Based on the illustrated embodiment, step S103 specifically includes:

[0070] Determining a first version number of the initial version model according to a first model name of the initial version model;

[0071] Determining a second version number of the target version data set according to the data set name of the target version data set;

[0072] The first version number and the second version number are combined to obtain the second model name.

[0073] In this embodiment, the combination method can be direct splicing or replacement, etc. Exemplarily, the first version number is M0 and the second version number is D1. For the target version data set D1, the initial version model M0 is trained and the model name of the secondary version model obtained can be obtained by splicing M0-D1 or by replacing M1-0; for the target version data set D1, the secondary version model is trained and the model name of the tertiary version model obtained can be obtained by splicing M0-D1-D1 or by replacing M1-1.

[0074] In one embodiment, Figure 1 Based on the illustrated embodiment, the above step S104 specifically includes:

[0075] Determining a first model version relationship between the initial version model and the new version model according to the first model name and the second model name;

[0076] Determining, according to the second model name, a second model version relationship between the plurality of new version models;

[0077] Determining a model data relationship between the target version data set and the new version model according to the data set name and the second model name;

[0078] The target model pedigree is established according to the first model version relationship, the second model version relationship and the model data relationship.

[0079] In this embodiment, the model relationship between models and the model-dataset relationship between models and data sets are determined by the version numbers in the model names and data set names. Figure 2 In the example diagram of the target model genealogy shown, D represents the dataset and M represents the model; the initial version model M0 is trained on the original dataset D0, model M1-0 is obtained by using M0 as the pre-trained model and fine-tuning on the dataset D1; model M2-0 is obtained by using M0 as the pre-trained model and fine-tuning on the dataset D2; model M2-1 is obtained by using M1-0 as the pre-trained model and fine-tuning on the dataset D2; model M3-0 is obtained by using M0 as the pre-trained model and fine-tuning on the dataset D3; model M3-1 is obtained by using M1-0 as the pre-trained model and fine-tuning on the dataset D3; model M3-2 is obtained by using M2-0 as the pre-trained model and fine-tuning on the dataset D3; model M3-3 is obtained by using M2-1 as the pre-trained model and fine-tuning on the dataset D3, and so on.

[0080] In one embodiment, Figure 1 Based on the embodiment shown, before the above step S101, it also includes:

[0081] Taking the initial version data set as the previous version data set, performing a change operation on the previous version data set to obtain a subsequent version data set, wherein the change operation includes an add operation, a delete operation or a modify operation;

[0082] The later version data set is used as the new earlier version data set, and a change operation is performed on the new earlier version data set to obtain a new later version data set, wherein the target version data set includes the initial version data set and all later version data sets.

[0083] In this embodiment, after collecting the original data set of a specific task, as the business progresses, the data set is continuously added, deleted, modified, and other operations are performed to continuously improve the performance of the model from the source of the data. The add operation (ADD) means adding one or more pictures and annotations, which is mainly used to expand the data set, such as returning new data or difficult samples from the field to the original data set for retraining; the delete operation (DEL) means deleting one or more pictures and annotations, which is mainly used to delete invalid samples or error samples, or to delete easy samples in the basic data set and retain difficult samples to construct a difficult data set; the modify operation (MOD) means modifying the annotation of one or more pictures, which is mainly used to modify the annotation information and improve the purity of the data set, such as fine-tuning the deviation rectangle box annotated by the detection task.

[0084] Optionally, after taking the later version data set as new earlier version data and performing a change operation on the new earlier version data set to obtain a new later version data set, the method further includes:

[0085] According to the first name of the previous version data set, rename the later version data set to obtain the second name of the later version data set;

[0086] A change sequence between the previous version data set and the later version data set is determined according to the change operation, the first name and the second name, wherein the change sequence is used to represent an inheritance relationship or a transition relationship between the previous version data set and the later version data set.

[0087] In this embodiment, each dataset genealogy has an original dataset D0, and the remaining datasets are obtained from the dataset D0 through change operations, and the change sequence of the change operations is recorded. For example, Figure 3 The schematic diagram of the dataset genealogy shown in FIG. 1 shows that the change sequence of the dataset D3 - 1 is: ADD->DEL->MOD, where the operation starting point is the original dataset D0.

[0088] Generally speaking, the data sets before and after the ADD and DEL operations are in a parent-child relationship, and the data sets before and after the MOD operation are in a transition relationship. If the data set version D2 is obtained through a change sequence on D1, a transition relationship is established between D2 and D1, denoted as D1→D2. If D1 contains D2, that is, all samples in D2 can be found in D1 and have the same annotations, then an inheritance relationship is established between D2 and D1, D1 is the parent, D2 is the child, denoted as D2∈D1. Transition relationships and inheritance relationships are transitive, for example, D1→D2→D3, then D1→D3; D1∈D2, D2∈D3, then D1∈D3.

[0089] In one embodiment, Figure 1 On the basis of the illustrated embodiment, after the above step S104, the method further includes: using a preset test data set to perform a test operation on the multiple new version models to obtain a performance indicator of each new version model.

[0090] In this embodiment, under the premise of determining the network structure and loss function, the main relevant factors for model training are data sets and hyperparameters, and the main relevant factors for model evaluation are test sets and test indicators. The model version is determined at the end of training, so it is mainly affected by the data set and hyperparameters. Here we mainly discuss the impact of the pre-trained model in the data set and hyperparameters on the model version.

[0091] The training of different model versions includes a set of hyperparameter settings, including learning rate, number of iterations, optimization method, loss function, etc., which affect the performance of the final model and can be combined with the pedigree relationship of the data set to reproduce the model. Recording the performance indicators of the model version under the test set, such as accuracy, recall rate, throughput, etc., is conducive to making decisions when selecting the model.

[0092] In order to execute the model resource processing method corresponding to the above method embodiment, to achieve the corresponding functions and technical effects. Figure 4 , Figure 4 The structural block diagram of a model resource processing device provided in an embodiment of the present application is shown. For the convenience of explanation, only the parts related to the present embodiment are shown. The model resource processing device provided in an embodiment of the present application includes:

[0093] An acquisition module 401 is used to acquire an initial version model and multiple target version data sets;

[0094] A training module 402 is used to perform multiple training operations on the initial version model based on the multiple target version data sets to obtain a new version model corresponding to each training operation;

[0095] A renaming module 403, configured to rename the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain a second model name of the new version model;

[0096] The generating module 404 is used to generate a target model pedigree between the initial version model, the target version data set and the new version model according to the first model name, the data set name and the second model name.

[0097] In one embodiment, the training module 402 includes:

[0098] A first training unit is used to use the initial version model as a previous version model, and use multiple first data sets to perform a training operation on the previous version model to obtain multiple subsequent version models, wherein each of the first data sets is one of the multiple target version data sets;

[0099] The second training unit is used to use the later version model as a new earlier version model, and use multiple second data sets to perform training operations on the new earlier version model to obtain multiple new later version models, each of which is one of the multiple target version data sets, and the new version model includes all the later version models.

[0100] In one embodiment, the renaming module 403 includes:

[0101] A first determining unit, configured to determine a first version number of the initial version model according to a first model name of the initial version model;

[0102] A second determining unit, configured to determine a second version number of the target version data set according to the data set name of the target version data set;

[0103] A combining unit is used to combine the first version number and the second version number to obtain the second model name.

[0104] In one embodiment, the generating module 404 includes:

[0105] A third determining unit, configured to determine a first model version relationship between the initial version model and the new version model according to the first model name and the second model name;

[0106] a fourth determining unit, configured to determine a second model version relationship between a plurality of the new version models according to the second model name;

[0107] a fifth determining unit, configured to determine a model data relationship between the target version data set and the new version model according to the data set name and the second model name;

[0108] An establishing unit is used to establish the target model pedigree according to the first model version relationship, the second model version relationship and the model data relationship.

[0109] In one embodiment, the processing device further includes:

[0110] A first change module is used to use the initial version data set as the previous version data set, perform a change operation on the previous version data set to obtain a later version data set, wherein the change operation includes an addition operation, a deletion operation or a modification operation;

[0111] The second change module is used to use the later version data set as the new earlier version data set, perform a change operation on the new earlier version data set, and obtain a new later version data set, wherein the target version data set includes the initial version data set and all later version data.

[0112] In one embodiment, the processing device further includes:

[0113] A second renaming module, used to rename the later version data set according to the first name of the previous version data set to obtain a second name of the later version data set;

[0114] A determination module is used to determine a change sequence between the previous version data set and the later version data set according to the change operation, the first name and the second name, wherein the change sequence is used to characterize an inheritance relationship or a transition relationship between the previous version data set and the later version data set.

[0115] In one embodiment, the processing device further includes:

[0116] The testing module is used to perform testing operations on the multiple new version models using a preset test data set to obtain performance indicators of each new version model.

[0117] The above-mentioned model resource processing device can implement the model resource processing method of the above-mentioned method embodiment. The optional items in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The rest of the contents of the embodiment of this application can refer to the contents of the above-mentioned method embodiment, and will not be repeated in this embodiment.

[0118] Figure 5 This is a schematic diagram of the structure of a computer device provided in one embodiment of the present application. Figure 5 As shown, the computer device 5 of this embodiment includes: at least one processor 50 ( Figure 5 Only one is shown in the figure) a processor, a memory 51, and a computer program 52 stored in the memory 51 and executable on the at least one processor 50, and when the processor 50 executes the computer program 52, the steps in any of the above method embodiments are implemented.

[0119] The computer device 5 may be a computing device such as a smart phone, a tablet computer, a desktop computer, a cloud server, etc. The computer device may include but is not limited to a processor 50 and a memory 51. Those skilled in the art will appreciate that Figure 5It is only an example of the computer device 5 and does not constitute a limitation on the computer device 5. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0120] The processor 50 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0121] In some embodiments, the memory 51 may be an internal storage unit of the computer device 5, such as a hard disk or memory of the computer device 5. In other embodiments, the memory 51 may also be an external storage device of the computer device 5, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 5. Further, the memory 51 may also include both an internal storage unit of the computer device 5 and an external storage device. The memory 51 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory 51 may also be used to temporarily store data that has been output or is to be output.

[0122] In addition, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0123] An embodiment of the present application provides a computer program product. When the computer program product is run on a computer device, the computer device implements the steps in the above-mentioned method embodiments when executing the computer device.

[0124] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.

[0125] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0126] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for processing model resources, characterized in that: include: Obtain an initial version model and multiple target version datasets; Based on the multiple target version data sets, multiple training operations are performed on the initial version model to obtain a new version model corresponding to each training operation; Rename the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain a second model name of the new version model; Generate a target model pedigree between the initial version model, the target version data set and the new version model according to the first model name, the data set name and the second model name; Before obtaining the initial version model and multiple target version data sets, the method further includes: Taking the initial version data set as the previous version data set, performing a change operation on the previous version data set to obtain a subsequent version data set, wherein the change operation includes an add operation, a delete operation or a modify operation; Using the later version data set as new earlier version data, performing a change operation on the new earlier version data set to obtain a new later version data set, wherein the target version data set includes the initial version data set and all later version data sets; After taking the later version data set as the new earlier version data set and performing a change operation on the new earlier version data set to obtain a new later version data set, the method further includes: According to the first name of the previous version data set, rename the later version data set to obtain the second name of the later version data set; A change sequence between the previous version data set and the later version data set is determined according to the change operation, the first name and the second name, wherein the change sequence is used to represent an inheritance relationship or a transition relationship between the previous version data set and the later version data set.

2. The method for processing model resources according to claim 1, characterized in that: The performing multiple training operations on the initial version model based on the multiple target version data sets to obtain a new version model corresponding to each training operation includes: Using the initial version model as a prior version model, and using a plurality of first data sets to perform a training operation on the prior version model to obtain a plurality of subsequent version models, wherein each of the first data sets is one of the plurality of target version data sets; The later version model is used as a new earlier version model, and a plurality of second data sets are used to perform training operations on the new earlier version model to obtain a plurality of new later version models, each of the second data sets being one of the plurality of target version data sets, and the new version model includes all the later version models.

3. The method for processing model resources according to claim 1, characterized in that: The step of renaming the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain the second model name of the new version model includes: Determining a first version number of the initial version model according to a first model name of the initial version model; Determining a second version number of the target version data set according to the data set name of the target version data set; The first version number and the second version number are combined to obtain the second model name.

4. The method for processing model resources according to claim 1, characterized in that: The step of generating a target model pedigree between the initial version model, the target version data set and the new version model according to the first model name, the data set name and the second model name comprises: Determining a first model version relationship between the initial version model and the new version model according to the first model name and the second model name; Determining, according to the plurality of second model names, a second model version relationship between the plurality of new version models; Determining a model data relationship between the target version data set and the new version model according to the data set name and the second model name; The target model pedigree is established according to the first model version relationship, the second model version relationship and the model data relationship.

5. The method for processing model resources according to claim 1, characterized in that: After performing multiple training operations on the initial version model based on the multiple target version data sets to obtain a new version model corresponding to each training operation, the method further includes: Using a preset test data set, a test operation is performed on a plurality of the new version models to obtain a performance indicator of each of the new version models.

6. A model resource processing device, characterized in that: include: An acquisition module is used to acquire an initial version model and multiple target version data sets; A training module, used to perform multiple training operations on the initial version model based on the multiple target version data sets to obtain a new version model corresponding to each training operation; A renaming module, used to rename the new version model according to the first model name of the initial version model and the data set name of the target version data set to obtain a second model name of the new version model; A generating module, configured to generate a target model pedigree between the initial version model, the target version data set and the new version model according to the first model name, the data set name and the second model name; The processing device further includes: A first change module is used to use the initial version data set as the previous version data set, perform a change operation on the previous version data set to obtain a later version data set, wherein the change operation includes an addition operation, a deletion operation or a modification operation; A second change module is used to use the later version data set as new earlier version data, perform a change operation on the new earlier version data set, and obtain a new later version data set, wherein the target version data set includes the initial version data set and all later version data; The processing device further includes: A second renaming module, used to rename the later version data set according to the first name of the previous version data set to obtain a second name of the later version data set; A determination module is used to determine a change sequence between the previous version data set and the later version data set according to the change operation, the first name and the second name, wherein the change sequence is used to characterize an inheritance relationship or a transition relationship between the previous version data set and the later version data set.

7. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program, and when the processor executes the computer program, the steps of the method for processing model resources as described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the steps of the method for processing model resources as described in any one of claims 1 to 5.

Images