Data processing method and device, terminal equipment and computer readable storage medium
By scoring and filtering the sample data sets, high-quality data is selected for training, which solves the problem of poor performance of pre-trained language models due to low quality of using label-free data, and achieves the effect of improving model performance.
Patent Information
- Application Number
- CN202311349046.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-17
- Publication Date
- 2025-04-18
AI Technical Summary
The existing pre-trained language models perform poorly in downstream tasks, mainly due to the low quality of the label-free data used during training, resulting in a degradation in model performance.
By scoring the sample data set, high-quality data is selected for model training to improve model performance. The specific steps include obtaining the sample data set, first scoring the data, filtering the target sample data based on the scoring results, and performing a second scoring for further screening.
By using high-quality sample data to train the model, the performance of the model can be effectively improved, avoiding the mixing of low-quality sample data during the training process, thereby improving the model's performance in downstream tasks.
Smart Images
Figure CN119848533A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a data processing method, apparatus, terminal device and computer-readable storage medium. Background Art
[0002] The concept of pre-trained models is not unfamiliar in the field of computer vision. Usually, a general model can be pre-trained on a large-scale image dataset, and then migrated to similar specific tasks. This reduces the demand for image samples while speeding up the development of the model. Similarly, the pre-trained language model is the application of the pre-training method in the field of natural language processing. In essence, it is the representation learning of natural language, which is to convert natural language into a data expression form that can be processed by machines. The role of the pre-trained language model is to incorporate a large amount of prior knowledge mastered by humans, and the method of incorporating prior knowledge is to train through a large amount of corpus.
[0003] The current training method of pre-trained language models is to train with unlabeled data to obtain a general language representation model. If the pre-trained model is used to perform some downstream tasks, the performance of the model in the downstream tasks will be unsatisfactory because the pre-trained data is unlabeled. Therefore, a method to improve the performance of the model is needed. Summary of the invention
[0004] The present application provides a data processing method for improving the performance of the model by screening out higher quality data for model training.
[0005] In a first aspect, the present application provides a data processing method, the method comprising:
[0006] Get a sample dataset;
[0007] Performing a first scoring on each sample data in the sample data set to obtain a first scoring result for each sample data;
[0008] According to the first scoring result of each of the sample data, target sample data is screened and obtained for use in model training.
[0009] In some embodiments of the present application, before obtaining the sample data set, the method further includes:
[0010] Get annotated datasets;
[0011] Adjusting a first parameter of a preset language model according to the labeled data set to obtain an instruction label determination model, wherein the instruction label determination model is used to determine an instruction label for each original sample data in the original sample data set to obtain sample data with each determined instruction label;
[0012] According to the sample data of each of the determined instruction labels, a second parameter adjustment is performed on the language model to obtain a first scoring model for performing a first scoring.
[0013] In some embodiments of the present application, the step of obtaining a sample data set includes:
[0014] Get the initial sample data set;
[0015] Generating an instruction label for each initial sample data in the initial sample data set;
[0016] Each of the initial sample data for generating the instruction tag is determined as a sample data set.
[0017] In some embodiments of the present application, the obtaining of the initial sample data set includes:
[0018] Get the sample data set to be processed;
[0019] Invalid sample data in the sample data set to be processed is deleted to obtain the initial sample data set.
[0020] In some embodiments of the present application, the step of screening target sample data according to the first scoring result of each sample data for model training includes:
[0021] Determine that the remaining sample data after screening is non-target sample data;
[0022] The non-target sample data and the target sample data are input into a preset target model to train the target model.
[0023] In some embodiments of the present application, the step of screening target sample data according to the first scoring result of each sample data for model training includes:
[0024] According to the first scoring result of each sample data, initial target sample data is obtained by screening each sample data in the sample data set;
[0025] Performing a second scoring on each of the initial target sample data to obtain a second scoring result;
[0026] According to the second scoring result, target sample data are screened from each of the initial target sample data for use in model training.
[0027] In some embodiments of the present application, before performing a second scoring on each of the initial target sample data to obtain a second scoring result, the method further includes:
[0028] Get multiple data to be scored;
[0029] Performing a first scoring on each of the data to be scored to obtain a first scoring result for each of the data to be scored;
[0030] According to the first scoring result of each of the data to be scored, each of the data to be scored is screened to obtain target scoring data;
[0031] According to each of the to-be-scored data and the target scoring data, a third parameter of a preset language model is adjusted to obtain a second scoring model;
[0032] By performing a second scoring on each of the initial target sample data according to the second scoring model, a second scoring result is obtained.
[0033] In a second aspect, the present application further provides a data processing device, the device comprising:
[0034] An acquisition module is used to obtain a sample data set;
[0035] A scoring module, used for performing a first scoring on each sample data in the sample data set to obtain a first scoring result for each sample data;
[0036] The screening module is used to screen the target sample data for model training according to the first scoring result of each sample data.
[0037] In some embodiments of the present application, the device further includes a parameter adjustment module, and the parameter adjustment module is specifically used to:
[0038] Get annotated datasets;
[0039] Adjusting a first parameter of a preset language model according to the labeled data set to obtain an instruction label determination model, wherein the instruction label determination model is used to determine an instruction label for each original sample data in the original sample data set to obtain sample data with each determined instruction label;
[0040] According to the sample data of each of the determined instruction labels, a second parameter adjustment is performed on the language model to obtain a first scoring model for performing a first scoring.
[0041] In some embodiments of the present application, the acquisition module is specifically used to:
[0042] Get the initial sample data set;
[0043] Generating an instruction label for each initial sample data in the initial sample data set;
[0044] Each of the initial sample data for generating the instruction tag is determined as a sample data set.
[0045] In some embodiments of the present application, the acquisition module is further used to:
[0046] Get the sample data set to be processed;
[0047] Invalid sample data in the sample data set to be processed is deleted to obtain the initial sample data set.
[0048] In some embodiments of the present application, the acquisition module is specifically used to:
[0049] Determine that the remaining sample data after screening is non-target sample data;
[0050] The non-target sample data and the target sample data are input into a preset target model to train the target model.
[0051] In some embodiments of the present application, the screening module is specifically used for:
[0052] According to the first scoring result of each sample data, initial target sample data is obtained by screening each sample data in the sample data set;
[0053] Performing a second scoring on each of the initial target sample data to obtain a second scoring result;
[0054] According to the second scoring result, target sample data are screened from each of the initial target sample data for use in model training.
[0055] In some embodiments of the present application, the parameter adjustment module is further used to:
[0056] Get multiple data to be scored;
[0057] Performing a first scoring on each of the data to be scored to obtain a first scoring result for each of the data to be scored;
[0058] According to the first scoring result of each of the data to be scored, each of the data to be scored is screened to obtain target scoring data;
[0059] According to each of the to-be-scored data and the target scoring data, a third parameter of a preset language model is adjusted to obtain a second scoring model;
[0060] By performing a second scoring on each of the initial target sample data according to the second scoring model, a second scoring result is obtained.
[0061] In a third aspect, the present application also provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any one of the data processing methods described.
[0062] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to implement the steps in any one of the data processing methods described.
[0063] The data processing method provided by the present application can screen out sample data with higher scores by scoring each sample data in the sample data set used for model training. Furthermore, sample data with higher quality can be determined based on the score of the sample data. Thus, if the model is trained with sample data of higher quality, the performance of the model can be effectively improved, and the mixing of sample data of lower quality during the training process can be avoided, thereby solving the problem that the performance of the model is reduced when the model is trained with sample data of lower quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0065] Figure 1 is a scenario schematic diagram of a data processing system provided in an embodiment of the present application;
[0066] Figure 2 This is a flow chart of an embodiment of a data processing method in an embodiment of the present application;
[0067] Figure 3 is a functional module schematic diagram of a data processing device in an embodiment of the present application;
[0068] Figure 4 It is a schematic diagram of the structure of the terminal device in the embodiment of the present application. DETAILED DESCRIPTION
[0069] 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 those skilled in the art without creative work are within the scope of protection of this application.
[0070] In the description of the present application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the feature. In the description of the present application, "plurality" means two or more, unless otherwise clearly and specifically defined.
[0071] In this application, the word "exemplary" is used to mean "used as an example, illustration or description". Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. At the same time, it is to be understood that in the specific implementation of this application, when user information, user data and other related data are involved, when the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data shall comply with relevant laws, regulations and standards of relevant countries and regions.
[0072] In order to enable any person skilled in the art to implement and use the present application, the following description is provided. In the following description, details are listed for the purpose of explanation. It should be understood that those of ordinary skill in the art will recognize that the present application can be implemented without using these specific details. In other examples, known structures and processes will not be elaborated in detail to avoid unnecessary details that make the description of the present application obscure. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the widest range of principles and features disclosed in the present application.
[0073] The present application provides a data processing method, apparatus, device and storage medium, which are described in detail below.
[0074] See also Figure 1 , Figure 1 Schematic diagram of a data processing system provided in an embodiment of the present application. The data processing system may include a terminal device 100 and a storage device 200. The storage device 200 may transmit data to the terminal device 100. Figure 1 The terminal device 100 in the embodiment can obtain the logs stored in the storage device 200 to execute the data processing method in the present application.
[0075] In the embodiment of the present application, the terminal device 100 includes but is not limited to a desktop computer, a portable computer, a network server, a PDA (Personal Digital Assistant), a tablet computer, a wireless terminal device, an embedded device, etc.
[0076] In an embodiment of the present application, communication between the terminal device 100 and the storage device 200 can be achieved through any communication method, including but not limited to mobile communications based on the 3rd Generation Partnership Project (3GPP), Long Term Evolution (LTE), and Worldwide Interoperability for Microwave Access (WiMAX), or computer network communications based on the TCP / IP Protocol Suite (TCP / IP) and User Datagram Protocol (UDP).
[0077] It should be noted that Figure 1 The scenario diagram of the data processing system shown is merely an example. The data processing system and scenario described in the embodiments of the present application are intended to more clearly illustrate the technical solution of the embodiments of the present application, and do not constitute a limitation on the technical solution provided in the embodiments of the present application. A person of ordinary skill in the art will appreciate that with the evolution of the data processing system and the emergence of new business scenarios, the technical solution provided in the embodiments of the present application is equally applicable to similar technical problems.
[0078] like Figure 2 As shown, Figure 2 This is a flow chart of an embodiment of a data processing method in an embodiment of the present application. The data processing method may include the following steps 201 to 203:
[0079] 201. Obtain a sample data set.
[0080] The sample data in the sample data set in the embodiment of the present application can be any type of data, such as text data, image data, etc., which is not limited in the specific embodiment of the present application. The sample data set is a data set formed by combining various sample data. Among them, the method of obtaining the sample data set can be to read the sample data set stored in the memory, or to directly obtain the sample data set sent by the user or other terminal, which is not limited in the specific embodiment of the present application.
[0081] 202. Perform a first scoring on each sample data in the sample data set to obtain a first scoring result for each sample data.
[0082] In the embodiment of the present application, when scoring each sample data in the sample data set, any scoring model can be used for scoring. For example: there are different types of reference data in the scoring model. After the sample data set is input into the scoring model, the scoring model can calculate the feature similarity between each sample data in the sample data set and different reference data, so as to score according to the feature similarity, and then the data set can be scored according to the degree of similarity. For example: the score of sample data with a similarity of 0%-20% can be 1 point; the score of sample data with a similarity of 21%-40% can be 2 points; the score of sample data with a similarity of 41%-60% can be 3 points; the score of sample data with a similarity of 61%-80% can be 4 points; the score of sample data with a similarity of 81%-100% can be 5 points. In this way, the scoring results of each sample data can be determined.
[0083] Alternatively, multiple types of reference features can be set in the scoring model. After the sample data set is input into the scoring model, a sample data in the sample data set is used for illustration. The scoring model can extract the data features in the sample data, and then decompose the extracted data features to obtain various sub-features. After that, the feature type of each sub-feature is determined, and the feature type of the sub-feature is compared with the feature type of each reference feature. The more the same feature type is, the higher the score of the current sample data is. For example: if there is one same feature type, it is 1 point; if there are two same feature types, it is 2 points, and so on, which will not be repeated here.
[0084] Among them, the reference data or reference features involved in the above description are used to determine the quality of the sample data. The higher the score, the closer the sample data is to the reference feature or reference data, and the higher the quality. Therefore, in actual situations, specific reference data or reference features can be set according to actual conditions. In addition, the scoring methods listed in the above description can be adjusted according to actual conditions, and the specific embodiments of the present application are not limited.
[0085] 203. According to the first scoring result of each sample data, target sample data is screened and obtained for model training.
[0086] The above description provides a scheme for how to score the sample data in the sample data set. Therefore, after obtaining the first scoring result of each sample data, the sample data with a higher score in the scoring result can be selected as the target sample data. For example: Continuing with the scheme described above, a score threshold of 3 can be set. If the score in the scoring result exceeds the score threshold 3, the sample data exceeding the score threshold 3 is determined to be the target sample data. It should be noted that the score threshold can be set according to the actual situation, and the specific embodiments of the present application are not limited. After the target sample data is obtained, the screened target sample data can be used for specific model training.
[0087] The data processing method provided by the present application can screen out sample data with higher scores by scoring each sample data in the sample data set used for model training. Furthermore, sample data with higher quality can be determined based on the score of the sample data. Thus, if the model is trained with sample data of higher quality, the performance of the model can be effectively improved, and the mixing of sample data of lower quality during the training process can be avoided, thereby solving the problem that the performance of the model is reduced when the model is trained with sample data of lower quality.
[0088] In order to better implement the embodiment of the present application, in one embodiment of the present application, before obtaining the sample data set, the method further includes:
[0089] Acquire a labeled data set; adjust a first parameter of a preset language model according to the labeled data set to obtain an instruction label determination model, the instruction label determination model is used to determine the instruction label of each original sample data in the original sample data set to obtain sample data of each determined instruction label; according to the sample data of each determined instruction label, adjust a second parameter of the language model to obtain a first scoring model for first scoring.
[0090] The above embodiment provides an implementation method of scoring by a scoring model, and the scoring model described in the above embodiment can be any existing scoring model. In order to enable the scoring model to better realize the scoring function, the present application provides a solution for determining the scoring model.
[0091] Specifically, in an embodiment of the present application, the labeled data set may be a data set with manually labeled labels and manually labeled scores, and the labeled data set includes multiple labeled data, and one labeled data corresponds to one manually labeled label and\or one manually labeled score. At the same time, the preset language model in the embodiment of the present application may be an LLAMA (Large Language Model Meta AI) language model, and the labeled data set is input into the LLAMA language model, and the label prediction is performed for each labeled data in the labeled data set through the language model, and it is determined whether the predicted label of the labeled data is different from the true label. If there is a situation where the predicted label is different from the true label, the model parameters of the LLAMA language model are adjusted until the predicted label of each labeled data obtained by the LLAMA language model is the same as the corresponding true label. Or determine the loss value of each labeled data in the labeled data set, determine whether the loss value of each labeled data is greater than a loss value threshold, and if there is at least one labeled data whose loss value is greater than the loss value threshold, adjust the model parameters of the LLAMA language model until the loss value of each labeled data obtained by the LLAMA language model is less than the loss value threshold. At this point, "adjusting the first parameter of the preset language model according to the labeled data set to obtain the instruction label determination model" is completed. It should be noted that if the model is adjusted based on the loss value, a specific loss function can be selected according to actual conditions, and the specific embodiments of the present application are not limited.
[0092] After the instruction label determination model is obtained, the instruction label determination model can predict the label of the data. At this time, when the original sample data without the label is obtained, the instruction label determination model can predict the instruction label of the original sample data without the label. However, some original sample data may get a situation without a predicted label after being predicted by the instruction label determination model. At this time, these data without predicted labels can be regarded as data of low quality. Therefore, the original sample data without predicted labels can be filtered to obtain the original sample data with predicted labels. After the original sample data with the predicted instruction label is obtained, the original sample data with the instruction label can be input into another untrained LLAMA language model for model training to obtain the first scoring model. Specifically, the training method can be the same as the above description, for example: the original sample data with the instruction label obtained by filtering is input into the LLAMA language model, and the label prediction is performed for the original sample data with the instruction label obtained by filtering through the language model, and it is determined whether the predicted label of the original sample data with the instruction label is different from the predicted label obtained by the instruction label determination model. If there is a situation where the predicted label is different from the predicted label obtained by the instruction label determination model, adjust the model parameters of the LLAMA language model until the predicted labels of each original sample data with instruction labels obtained by the LLAMA language model are the same as the predicted labels obtained by the corresponding instruction label determination model. At this time, "adjusting the second parameter of the language model according to the sample data of each determined instruction label" is completed. It should be noted that.
[0093] It should be noted that each original sample data in the original sample data set can be obtained from a preset memory, or can be the original sample data transmitted by the user or terminal, or can be the original sample data set composed of the original sample data obtained from the network, and the specific embodiments of the present application are not limited thereto.
[0094] In order to better implement the embodiment of the present application, in one embodiment of the present application, obtaining a sample data set includes:
[0095] Acquire an initial sample data set; generate an instruction label for each initial sample data in the initial sample data set; and determine each initial sample data for which an instruction label is generated as a sample data set.
[0096] The above embodiment provides a solution for directly scoring each sample data in a sample data set. In actual situations, the sample data immediately acquired may not have an instruction label. Therefore, in order to ensure that the sample data to be scored has a certain data quality, an initial sample data set may be acquired first. Each initial sample data in the initial sample data set is sample data that is not labeled with an instruction label. Therefore, after acquiring the initial sample data set, the instruction label of each initial sample data in the initial sample data set may be predicted first, thereby filtering out some initial sample data without instruction labels, and treating the set of initial sample data with instruction labels as the sample data set. Among them, the instruction labels of each initial sample data in the initial sample data set may be determined by the instruction label determination model listed in the above embodiment, and the specific details will not be repeated here.
[0097] In order to better implement the embodiment of the present application, in one embodiment of the present application, obtaining an initial sample data set includes:
[0098] Obtain a sample data set to be processed; delete invalid sample data in the sample data set to be processed to obtain an initial sample data set.
[0099] In actual situations, duplicate data may exist in the obtained initial sample data set. For example, when the initial sample data is obtained from different addresses, duplicate initial sample data may exist in different addresses. Therefore, duplicate data may exist in the aggregated initial sample data set. In order to avoid repeated calculations and waste of computing resources, it is necessary to remove duplicate data. In addition, in addition to duplicate data, there are also data with similar situations such as damaged data, and duplicate data, damaged data and other data can all be regarded as invalid sample data in the embodiments of the present application. These invalid sample data are not conducive to the training of the model, so they need to be eliminated.
[0100] Among them, in the embodiment of the present application, the sample data set to be processed may be a sample data set transmitted from a corresponding storage device, network, manual or other terminal. After the sample data set to be processed is obtained, the sample data in the sample data set to be processed can be compared two by two to remove duplicate data. When there are two identical sample data, the duplicate data can be removed. In addition, whether the data is damaged can be detected according to the type of sample data. For example: if the sample data to be processed is image data, it is detected whether the image data can be opened normally. If the image data cannot be opened normally, it is determined to be damaged data; if the sample data to be processed is text data, it can be detected whether there is garbled code in the text data, or whether the proportion of garbled code is large. If there is garbled code in the text data, and the proportion of garbled code is large, it can be determined that the current data is damaged data, and the damaged data can be deleted at this time. When the damaged data is deleted, the remaining data can be regarded as the initial sample data set.
[0101] In order to better implement the embodiment of the present application, in one embodiment of the present application, according to the first scoring result of each sample data, the target sample data is screened and obtained for model training, including:
[0102] Determine that the remaining sample data is non-target sample data; input the non-target sample data and the target sample data into a preset target model to train the target model.
[0103] In the above embodiment, a scheme is provided for training the model with the target sample data obtained after screening. However, in actual situations, there may be a situation where the target sample data is less. Therefore, if the target sample data is less, for example, after the target sample data is obtained, the number of target sample data can be determined. If the number of target sample data is less than the preset number threshold, the non-target sample data that was previously filtered out can be used as the training sample of the model at the same time. Although it may reduce the training efficiency of the model, it can make up for the problem of insufficient training samples.
[0104] In order to better implement the embodiment of the present application, in one embodiment of the present application, according to the first scoring result of each sample data, the target sample data is screened and obtained for model training, including:
[0105] According to the first scoring result of each sample data, initial target sample data is obtained by screening each sample data in the sample data set; a second scoring is performed on each initial target sample data to obtain a second scoring result; according to the second scoring result, target sample data is screened from each initial target sample data for use in model training.
[0106] The above embodiment provides a solution for determining target sample data according to the first scoring result. However, in order to further improve the training effect of the model, the embodiment of the present application also provides a solution for obtaining higher quality data through multi-level scoring and multi-level screening.
[0107] Specifically, when scoring each sample data in the sample data set, any scoring model can be used for scoring. For example: there are different types of reference data in the scoring model. After the sample data set is input into the scoring model, the scoring model can calculate the feature similarity between each sample data in the sample data set and different reference data, so as to score according to the feature similarity, and then score the data set according to the degree of similarity. For example: the score of sample data with a similarity of 0%-20% can be 1 point; the score of sample data with a similarity of 21%-40% can be 2 points; the score of sample data with a similarity of 41%-60% can be 3 points; the score of sample data with a similarity of 61%-80% can be 4 points; the score of sample data with a similarity of 81%-100% can be 5 points. In this way, the scoring results of each sample data can be determined.
[0108] Alternatively, multiple types of reference features can be set in the scoring model. After the sample data set is input into the scoring model, a sample data in the sample data set is used for illustration. The scoring model can extract the data features in the sample data, and then perform feature decomposition on the extracted data features to obtain various sub-features. Afterwards, the feature type of each sub-feature is determined, and the feature type of the sub-feature is compared with the feature type of each reference feature. The more the same feature types are, the higher the score of the current sample data. For example: if there is one same feature type, it is 1 point; if there are two same feature types, it is 2 points, and so on, which will not be repeated here. At this time, the sample data of the first scoring result can be obtained, which is the same as the above embodiment. At this time, a score threshold can be set, and the initial sample data set that is less than the score threshold is filtered out according to the score threshold to obtain the initial target sample data.
[0109] After the initial target sample data is obtained, another scoring model can be used to perform a second scoring on the initial target sample data. The reference data or reference features in the other scoring model are different from the reference data and reference features in the first scoring model. Therefore, the above scheme can still be used to continue scoring according to the initial target sample data to obtain a second scoring result. After the scores corresponding to each initial target sample data are determined, the target sample data can be screened according to another score threshold. The second scoring step and the second screening step are similar to the first scoring step and the first screening step, and will not be described in detail here.
[0110] In order to better implement the embodiment of the present application, in one embodiment of the present application, before performing a second scoring on each initial target sample data and obtaining the second scoring result, the method further includes:
[0111] Acquire multiple data to be scored; perform a first score on each data to be scored to obtain a first scoring result for each data to be scored; screen each data to be scored according to the first scoring result for each data to be scored to obtain target scoring data; perform a third parameter adjustment on a preset language model according to each data to be scored and the target scoring data to obtain a second scoring model; perform a second score on each initial target sample data according to the second scoring model to obtain a second scoring result.
[0112] The embodiment of the present application also provides a scheme for training a second scoring model. Specifically, after obtaining the data to be scored, the first scoring model trained in the above embodiment can be used for scoring, and then the target scoring data is screened out, and the target scoring data is the data with higher quality after the first round of screening. After the target scoring data is determined, the target scoring data and the data to be scored are sent to the LLAMA language model for training. Specifically, the LLAMA language model can be made to determine the difference between the target scoring data and the non-target scoring data, so that the LLAMA language model learns the difference characteristics between the two data of different quality. Therefore, the LLAMA language model can be made to better identify data with better quality.
[0113] In an embodiment of the present application, the LLAMA language model can be trained according to the relevant loss function. If the total loss obtained after the target scoring data and the non-target scoring data are input into the LLAMA language model is greater than the set loss value threshold, the model parameters of the LLAMA language model can be adjusted so that the total loss of the LLAMA language model is less than the loss value threshold to complete the training of the second scoring model.
[0114] In order to better implement the data processing method in the embodiment of the present application, in addition to the data processing method, the embodiment of the present application also provides a data processing device, such as Figure 3 As shown, the device 300 includes:
[0115] An acquisition module 301 is used to acquire a sample data set;
[0116] The scoring module 302 is used to perform a first scoring on each sample data in the sample data set to obtain a first scoring result for each sample data;
[0117] The screening module 303 is used to screen target sample data for model training according to the first scoring result of each sample data.
[0118] The data processing device provided by the present application can first obtain a sample data set through the acquisition module 301, and then score each sample data in the sample data set used for model training through the scoring module 302, so as to screen out sample data with higher scores. Furthermore, the screening module 303 can determine the sample data with higher quality according to the score of the sample data. At this point, if the model is trained with sample data of higher quality, the performance of the model can be effectively improved, and the mixing of sample data of lower quality during the training process can be avoided, thereby solving the problem that the performance of the model is reduced when the model is trained with sample data of lower quality.
[0119] In some embodiments of the present application, the device further includes a first training module, and the first training module is specifically used to:
[0120] Get annotated datasets;
[0121] Adjusting a first parameter of a preset language model according to the labeled data set to obtain an instruction label determination model, wherein the instruction label determination model is used to determine an instruction label for each original sample data in the original sample data set to obtain sample data with each determined instruction label;
[0122] According to the sample data of each determined instruction label, a second parameter adjustment is performed on the language model to obtain a first scoring model for performing a first scoring.
[0123] In some embodiments of the present application, the acquisition module 301 is specifically used to:
[0124] Get the initial sample data set;
[0125] Generate an instruction label for each initial sample data in the initial sample data set;
[0126] Each initial sample data for generating an instruction label is determined as a sample data set.
[0127] In some embodiments of the present application, the acquisition module 301 is further used to:
[0128] Get the sample data set to be processed;
[0129] The invalid sample data in the sample data set to be processed is deleted to obtain the initial sample data set.
[0130] In some embodiments of the present application, the screening module 303 is specifically used to:
[0131] Determine that the remaining sample data after screening is non-target sample data;
[0132] The non-target sample data and the target sample data are input into a preset target model to train the target model.
[0133] In some embodiments of the present application, the screening module 303 is further used to:
[0134] According to the first scoring result of each sample data, initial target sample data is obtained by screening each sample data in the sample data set;
[0135] Performing a second scoring on each initial target sample data to obtain a second scoring result;
[0136] According to the second scoring result, target sample data are screened from each initial target sample data to be used for model training.
[0137] In some embodiments of the present application, the device further includes a second training module, and the second training module is specifically used to:
[0138] Get multiple data to be scored;
[0139] Performing a first scoring on each data to be scored, and obtaining a first scoring result for each data to be scored;
[0140] According to the first scoring result of each data to be scored, each data to be scored is screened to obtain target scoring data;
[0141] According to each of the to-be-scored data and the target scoring data, a third parameter of the preset language model is adjusted to obtain a second scoring model;
[0142] By performing a second scoring on each initial target sample data according to the second scoring model, a second scoring result is obtained.
[0143] The present application also provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps of any data processing method in the present application. The terminal device integrates any data processing method provided in the present application. Figure 4 As shown, it shows a schematic diagram of the structure of the terminal device involved in the embodiment of the present application, specifically:
[0144] The terminal device may include one or more processing core processors 401, one or more computer-readable storage media memories 402, a power supply 403, an input unit 404 and other components. Those skilled in the art will appreciate that Figure 4The terminal device structure shown in the figure does not constitute a limitation on the terminal device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. Among them:
[0145] The processor 401 is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device. By running or executing software programs and / or modules stored in the memory 402, and calling data stored in the memory 402, the processor 401 executes various functions of the terminal device and processes data, thereby monitoring the terminal device as a whole. Optionally, the processor 401 may include one or more processing cores; the processor 401 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 gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. Preferably, the processor 401 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, the user interface and the application program, etc., and the modem processor mainly processes wireless communication. It is understandable that the above-mentioned modem processor may not be integrated into the processor 401.
[0146] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 402 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.
[0147] The terminal device also includes a power supply 403 for supplying power to each component. Preferably, the power supply 403 can be logically connected to the processor 401 through a power management system, so as to manage charging, discharging, power consumption management and other functions through the power management system. The power supply 403 can also include any components such as one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, and power status indicators.
[0148] The terminal device may further include an input unit 404, which may be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.
[0149] Although not shown, the terminal device may further include a display unit, etc., which will not be described in detail herein. Specifically in this embodiment, the processor 401 in the terminal device will load the executable files corresponding to the processes of one or more application programs into the memory 402 according to the following instructions, and the processor 401 will run the application programs stored in the memory 402, thereby realizing various functions, such as:
[0150] Get a sample dataset;
[0151] Performing a first scoring on each sample data in the sample data set to obtain a first scoring result for each sample data;
[0152] According to the first scoring result of each sample data, the target sample data is screened and obtained for model training.
[0153] A person of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.
[0154] To this end, an embodiment of the present application provides a computer-readable storage medium, which may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in any data processing method provided in the embodiment of the present application. For example, the computer program is loaded by a processor to execute the following steps:
[0155] Get a sample dataset;
[0156] Performing a first scoring on each sample data in the sample data set to obtain a first scoring result for each sample data;
[0157] According to the first scoring result of each sample data, the target sample data is screened and obtained for model training.
[0158] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the detailed description of other embodiments above, and will not be repeated here.
[0159] In specific implementation, the above units or structures can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units or structures can refer to the previous method embodiments, which will not be repeated here.
[0160] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.
[0161] The above is a detailed introduction to a data processing method and device provided in an embodiment of the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for technical personnel in this field, according to the idea of the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.
Claims
1. A data processing method, characterized in that: The method comprises: Get a sample dataset; Performing a first scoring on each sample data in the sample data set to obtain a first scoring result for each sample data; According to the first scoring result of each of the sample data, target sample data is screened and obtained for use in model training.
2. The data processing method according to claim 1, characterized in that: Before obtaining the sample data set, the method further includes: Get annotated datasets; Adjusting a first parameter of a preset language model according to the labeled data set to obtain an instruction label determination model, wherein the instruction label determination model is used to determine an instruction label for each original sample data in the original sample data set to obtain sample data with each determined instruction label; According to the sample data of each of the determined instruction labels, a second parameter adjustment is performed on the language model to obtain a first scoring model for performing a first scoring.
3. The data processing method according to claim 1, characterized in that: The obtaining of the sample data set comprises: Get the initial sample data set; Generating an instruction label for each initial sample data in the initial sample data set; Each of the initial sample data for generating the instruction tag is determined as a sample data set.
4. The data processing method according to claim 3, characterized in that: The obtaining of the initial sample data set comprises: Get the sample data set to be processed; Invalid sample data in the sample data set to be processed is deleted to obtain the initial sample data set.
5. The data processing method according to claim 3, characterized in that: The step of screening target sample data according to the first scoring result of each sample data for model training includes: Determine that the remaining sample data after screening is non-target sample data; The non-target sample data and the target sample data are input into a preset target model to train the target model.
6. The data processing method according to claim 1, characterized in that: The step of screening target sample data according to the first scoring result of each sample data for model training includes: According to the first scoring result of each sample data, initial target sample data is obtained by screening each sample data in the sample data set; Performing a second scoring on each of the initial target sample data to obtain a second scoring result; According to the second scoring result, target sample data are screened from each of the initial target sample data for use in model training.
7. The data processing method according to claim 6, characterized in that: Before performing a second scoring on each of the initial target sample data to obtain a second scoring result, the method further includes: Get multiple data to be scored; Performing a first scoring on each of the data to be scored to obtain a first scoring result for each of the data to be scored; According to the first scoring result of each of the data to be scored, each of the data to be scored is screened to obtain target scoring data; According to each of the to-be-scored data and the target scoring data, a third parameter of a preset language model is adjusted to obtain a second scoring model; By performing a second scoring on each of the initial target sample data according to the second scoring model, a second scoring result is obtained.
8. A data processing device, characterized in that: The device comprises: An acquisition module is used to obtain a sample data set; A scoring module, used for performing a first scoring on each sample data in the sample data set to obtain a first scoring result for each sample data; The screening module is used to screen the target sample data for model training according to the first scoring result of each sample data.
9. A terminal device, characterized in that: The terminal device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps in the data processing method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps in the data processing method according to any one of claims 1 to 7.