Model training method and device, service execution method and device, medium and electronic equipment

Abstract

The invention discloses a model training method and device, a business execution method and device, a medium and electronic equipment, and the method comprises the steps: obtaining a target sample set, determining the input information corresponding to the training sample and the guide information corresponding to the input information according to the training sample for each training sample in the target sample set, and carrying out the operation of the input information; inputting the input information and the guide information into a to-be-trained model, so that the to-be-trained model determines initial logic information based on which an output result corresponding to the input information is output according to the input information and the guide information, and adjusts the initial logic information through the to-be-trained model according to a standard output result corresponding to the input information, and adjusting the logic information to obtain adjusted logic information, determining an output result of the to-be-trained model for the input information according to the adjusted logic information, and training the to-be-trained model according to the output result.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a model training method, a business execution method, a device, a medium and an electronic device. Background Art

[0002] With the continuous development of computer technology, technologies such as artificial intelligence and big data analysis have been applied in many business scenarios to improve users' various business experiences while ensuring the security of users' personal information and privacy data.

[0003] In the current field of artificial intelligence, large language models are becoming the mainstream direction of future development. Users can quickly get corresponding answers by inputting questions into large language models.

[0004] However, currently large language models cannot give accurate answers during the inference process, which brings inconvenience to users when using large language models. Summary of the invention

[0005] This specification implements a model training method, a business execution method, a device, a medium and an electronic device to partially solve the problem that the large language model in the prior art cannot give accurate answers, thereby causing inconvenience to users.

[0006] The embodiments of this specification adopt the following technical solutions:

[0007] The embodiment of this specification provides a model training method, including:

[0008] Acquire a target sample set, wherein the training samples contained in the target sample set include at least one of text, picture and audio;

[0009] For each training sample in the target sample set, determine input information corresponding to the training sample and guidance information corresponding to the input information according to the training sample, wherein the guidance information is used to guide the model to be trained to output logical information based on which the output result is obtained, and the logical information is used to represent the logical reasoning process based on which the output result of the model to be trained is based;

[0010] Inputting the input information and the guiding information into the model to be trained, so that the model to be trained determines the initial logic information based on which the output result corresponding to the input information is output according to the input information and the guiding information;

[0011] According to the standard output result corresponding to the input information, the initial logic information is adjusted by the model to be trained to obtain adjusted logic information;

[0012] According to the adjusted logic information, the output result of the model to be trained for the input information is determined, so as to train the model to be trained according to the output result.

[0013] Furthermore, in some implementations, obtaining the target sample set specifically includes:

[0014] Obtain an initial sample set;

[0015] For each initial sample in the initial sample set, input information corresponding to the initial sample is input into the model to be trained to obtain an output result corresponding to the initial sample, and the uncertainty corresponding to the initial sample is determined according to the output result corresponding to the initial sample;

[0016] The initial samples whose uncertainty is not less than the first preset threshold are added to the target sample set.

[0017] Furthermore, in some implementations, obtaining the target sample set specifically includes:

[0018] Determine a historical sample set used in the history to train the model to be trained;

[0019] For each historical sample in the historical sample set, new information corresponding to the historical sample is determined based on the historical sample, and the historical sample is adjusted based on the new information to add the adjusted sample to the target sample set, and the new information has not been used in the training of the model to be trained in history.

[0020] Furthermore, in some implementations, obtaining the target sample set specifically includes:

[0021] Obtain an initial sample set;

[0022] For each initial sample in the initial sample set, supplementary information matching the initial sample is determined based on the initial sample, and the supplementary information is integrated into the initial sample to add the integrated sample to the target sample set.

[0023] Furthermore, in some implementations, obtaining the target sample set specifically includes:

[0024] Obtain an initial sample set;

[0025] For each initial sample in the initial sample set, determine the information dimension quantity corresponding to the initial sample according to the initial sample, wherein the larger the information dimension quantity is, the more dimensions of information the initial sample contains;

[0026] Selecting, from the initial sample set, initial samples whose information dimension is not less than a second preset threshold as target samples;

[0027] The target sample is added to the target sample set.

[0028] Furthermore, in some implementations, determining input information corresponding to the training sample according to the training sample specifically includes:

[0029] Generate a query instruction according to the training sample;

[0030] According to the query instruction, query the associated information matching the training sample from a preset information database;

[0031] According to the association information and the training sample, input information corresponding to the training sample is determined.

[0032] The embodiment of this specification provides a service execution method, including:

[0033] Receiving input information from a user, wherein the input information includes at least one of text, picture, and audio;

[0034] Inputting the input information into a pre-trained target model so that the target model outputs an output result for the input information, wherein the target model is trained by the above-mentioned model training method;

[0035] Execute the service according to the output result.

[0036] The embodiment of this specification provides a model training device, including:

[0037] An acquisition module, used to acquire a target sample set, wherein the training samples contained in the target sample set include at least one of text, picture and audio;

[0038] A determination module, for determining, for each training sample in the target sample set, input information corresponding to the training sample and guidance information corresponding to the input information according to the training sample, wherein the guidance information is used to guide the output of the model to be trained to obtain logic information based on which the output result is obtained, and the logic information is used to represent the logic reasoning process based on which the output result of the model to be trained is based;

[0039] An input module, used for inputting the input information and the guiding information into the model to be trained, so that the model to be trained determines the initial logic information based on which the output result corresponding to the input information is output according to the input information and the guiding information;

[0040] An adjustment module, configured to adjust the initial logic information through the model to be trained according to a standard output result corresponding to the input information, so as to obtain adjusted logic information;

[0041] A training module is used to determine the output result of the model to be trained for the input information according to the adjusted logic information, so as to train the model to be trained according to the output result.

[0042] The embodiment of this specification provides a service execution device, including:

[0043] A receiving module, used to receive input information from a user, wherein the input information includes at least one of text, picture and audio;

[0044] An input module, used for inputting the input information into a pre-trained target model so that the target model outputs an output result for the input information, wherein the target model is trained by the above-mentioned model training method;

[0045] An execution module is used to execute the business according to the output result.

[0046] An embodiment of the present specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned model training method and business execution method are implemented.

[0047] An embodiment of the present specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned model training method and business execution method when executing the program.

[0048] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0049] In the model training method and business execution method provided in the embodiments of the present specification, a target sample set is obtained, wherein the training samples contained in the target sample set include at least one of text, pictures and audio, and then, for each training sample in the target sample set, the input information corresponding to the training sample and the guidance information corresponding to the input information are determined according to the training sample, the guidance information is used to guide the model to be trained to output the logical information based on which the output result is obtained, and the logical information is used to represent the logical reasoning process based on which the output result of the model to be trained is based, and then, the input information and the guidance information are input into the model to be trained, so that the model to be trained determines the initial logical information based on which the output result corresponding to the input information is output according to the input information and the guidance information, and according to the standard output result corresponding to the input information, the initial logical information is adjusted by the model to be trained to obtain the adjusted logical information, and according to the adjusted logical information, the output result of the model to be trained for the input information is determined, so as to train the model to be trained according to the output result.

[0050] It can be seen from the above method that in the process of model training, the model is guided by guiding information to give logical information for outputting the output results, and by adjusting the logical information, the logical reasoning based on which the trained model outputs the results is more reasonable, thereby effectively realizing that the trained model can output accurate results based on accurate and reasonable logical reasoning, which in turn brings great convenience to users who use the model and significantly improves the business efficiency of users who use the model to execute business. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A flowchart of a model training method provided in an embodiment of this specification;

[0052] Figure 2 A schematic diagram of a model training architecture provided in an embodiment of this specification;

[0053] Figure 3 A flowchart of a service execution method provided in an embodiment of this specification;

[0054] Figure 4 A schematic diagram of a model training device provided in an embodiment of this specification;

[0055] Figure 5 A schematic diagram of a service execution device provided in an embodiment of this specification;

[0056] Figure 6 A method corresponding to the embodiment of this specification is provided Figure 1 or Figure 3 Schematic diagram of an electronic device. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0058] At present, Large Language Model (LLM) refers to a type of complex artificial intelligence model trained with large-scale text data sets. It adopts deep learning architecture, especially the Transformer architecture widely used in the field of Natural Language Processing (NLP). This type of model has a very high number of parameters, usually reaching tens of billions or more, which gives the large language model powerful language understanding and generation capabilities.

[0059] By learning the language rules and patterns in massive text data, large language models can demonstrate superior performance in various natural language processing tasks, including but not limited to answering, text generation, translation, summary generation, dialogue interaction, sentiment analysis, etc., and can demonstrate high flexibility and creativity in these tasks. Due to its strong generalization ability and high adaptability to language structure, large language models have become an important pillar of modern natural language processing technology.

[0060] However, current large language models also have many problems, such as hallucination problems (referring to the fact that large language models may produce seemingly reasonable but actually inaccurate or completely fictitious information when outputting results) and outdated parameter memory, which will cause the output results generated by large language models to be inaccurate, thus causing a certain degree of inconvenience to users in the process of using large language models.

[0061] In order to solve the above problems, the embodiments of this specification provide a model training method and a business execution method. By adjusting the logical information based on the output results of the model to be trained during the model training process, the output results obtained by the model to be trained through reasonable and accurate adjusted logical information can be more accurate. Through this model training method, the trained model can obtain more accurate output results in the subsequent business execution process through the logical reasoning process "learned" during the training process, thereby greatly facilitating the user to use the trained model.

[0062] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0063] Figure 1 A flowchart of a model training method provided in an embodiment of this specification.

[0064] S100: Acquire a target sample set, wherein the training samples included in the target sample set include at least one of text, picture and audio.

[0065] In order to train the model to be trained so that the trained model can output more accurate results in the subsequent business execution process, in the embodiment of this specification, a target sample set used for training the model can be obtained.

[0066] Among them, the execution entity for executing model training can be a server, or it can be a model training system composed of terminal devices such as desktop computers, laptops and servers, or special training equipment for model training, etc. For the sake of ease of explanation, the model training method provided in the embodiments of this specification is described below only with the server as the execution entity.

[0067] The model to be trained mentioned above may refer to a large language model. Of course, the model training method provided in the embodiments of this specification may also be applicable to other types of artificial intelligence models. The embodiments of this specification do not explicitly limit the model to be trained.

[0068] The target sample set mentioned above contains a large number of training samples, and for any training sample, its sample form can refer to text, picture, audio, etc. For example, the training sample can be a question text, which is used to ask a specific question to the model to be trained and generate a corresponding answer; for another example, the training sample can be a picture, and the model to be trained can process the picture accordingly according to the prompt language; for another example, the training sample can be audio, and the model to be trained can perform tasks such as speech-to-text conversion and speech conversion (converting one person's speech into another person's speech) on the audio. In addition, a training sample can also contain multiple forms of content. For example, a training sample can contain a picture, a text describing the picture, and a prompt text describing the next task that the model needs to perform; for another example, a training sample can contain audio, and a prompt text describing the next task that the model needs to perform.

[0069] In addition, for a training sample, if the training sample includes a picture, the picture may be a single picture or a video composed of a plurality of continuous pictures.

[0070] In order to improve the training effect of the model so that the trained model can accurately output results in subsequent business execution, in the embodiments of this specification, some suitable and characteristic training samples can be prepared for the model to be trained. The specific training samples can be roughly divided into the following five categories:

[0071] 1. Ordinary samples

[0072] The so-called ordinary samples may refer to samples with annotated labels or mixed samples with annotated labels and without labels. There are no strict quality requirements for ordinary samples, as long as they meet the actual training standards and requirements.

[0073] 2. Uncertain Samples

[0074] The so-called uncertain samples refer to samples that cannot be well identified or clearly determined for the model to be trained. For example, after a sample is input into the model, the model can classify the sample to determine the probability that the sample belongs to each category. For samples with high certainty, when the model classifies it, the probability that the sample belongs to a certain category is significantly higher than the probability that the sample belongs to other categories. For samples with low certainty, when the model classifies it, the probability that the sample belongs to a certain category may not be significantly different from the probability that the sample belongs to other categories. Therefore, such samples with low certainty belong to the above-mentioned uncertain samples.

[0075] 3. High-value samples

[0076] The so-called high-value samples refer to samples that contain relatively rich information content. For example, for two samples, the content of sample A includes name, age, and address, and the content of sample B includes name, age, address, position, hobbies, and education. It can be obviously seen that the content contained in sample B is richer than that contained in sample A. Therefore, sample B is a high-value sample compared to sample A.

[0077] 4. Error Correction Sample

[0078] The so-called error correction samples refer to samples used to correct the existing knowledge "learned" by the model. Compared with the historical samples used to train the model, these samples are new samples, and the knowledge they contain is more updated and accurate than the knowledge contained in the historical samples. For example, after the model is trained using historical samples and used for a period of time, the new knowledge generated during this period of time can be used as the above-mentioned error correction samples.

[0079] 5. Strengthening samples

[0080] The so-called enhanced samples can be understood as adding some relevant information matching the sample itself to any of the above four types of samples, so as to further improve the knowledge of the sample itself, so that the model can deepen its understanding of various types of "knowledge" during the training process.

[0081] For the first type of samples mentioned above, the server can obtain them from a preset sample library. For example, the server queries the required samples from the preset sample library through the received query instructions, and adds the queried samples to the target sample set. Among them, most of the samples queried from the preset sample library can be historical samples.

[0082] For the second type of samples mentioned above, the server can obtain various required samples from the preset sample library to construct an initial sample set. For each initial sample in the initial sample set, the server can input the input information corresponding to the initial sample into the above-mentioned model to be trained to obtain the output result corresponding to the initial sample, and then determine the uncertainty corresponding to the initial sample based on the output result corresponding to the initial sample.

[0083] The input information corresponding to the initial sample mentioned here can refer to the initial sample itself, or it can refer to the information composed of the initial sample combined with the corresponding prompt sentence. For example, suppose an initial sample is a picture, and the prompt sentence is: "Please generate a landscape painting according to the style of this picture". Since both this picture and this prompt sentence need to be input into the model to be trained, the picture + prompt sentence is the input information.

[0084] In the embodiments of the present specification, there may be multiple ways to determine the uncertainty corresponding to the initial sample. For example, the uncertainty corresponding to the initial sample may be determined by determining the minimum confidence corresponding to the initial sample. Specifically, when the trained model determines the classification category corresponding to the initial sample, it may also output the confidence of the probability that the initial sample belongs to each classification category. The confidence corresponding to the classification category with the highest probability may be used as the basis for determining the uncertainty corresponding to the initial sample. That is, the lower the confidence of the classification category with the highest probability, the higher the uncertainty corresponding to the initial sample.

[0085] For another example, after the input information corresponding to the initial sample is input into the model to be trained, the information entropy corresponding to the initial sample can be determined based on the probability distribution of the initial sample belonging to each classification category output by the model to be trained, and then the uncertainty corresponding to the initial sample can be determined based on the information entropy corresponding to the initial sample. Among them, if the information entropy corresponding to the initial sample is higher, it means that the prediction of the initial sample by the model to be trained is more uncertain, and the uncertainty corresponding to the initial sample is higher.

[0086] For another example, after the input information corresponding to the initial sample is input into the model to be trained, the category whose probability of the initial sample belonging to the classification category ranks before the set ranking can be further determined in descending order, that is, the target category, and then the difference between the probabilities of the initial sample belonging to these target categories can be determined, and then the uncertainty corresponding to the initial sample can be determined according to the determined probability difference. Among them, taking the category before the set ranking mainly selects the larger categories to which the initial sample belongs. If the difference between the probabilities of the larger categories to which the initial sample belongs is small, it means that the model to be trained has a weak ability to distinguish the initial sample, and the uncertainty corresponding to the initial sample is large.

[0087] The above are several ways to determine uncertainty. Of course, in practical applications, the uncertainty corresponding to the initial sample can also be determined by other methods, which will not be explained in detail here.

[0088] After determining the uncertainty corresponding to each initial sample in the initial sample set, the initial samples whose uncertainty is not lower than the first preset threshold can be added to the target sample set, that is, the initial samples with higher uncertainty are selected for training the model to be trained.

[0089] Since the model to be trained cannot distinguish and process these initial samples with high uncertainty very well, these samples have high training value for the model to be trained. In this way, the model to be trained can actively "learn" some external knowledge with high training value (the reason why it is called external knowledge is that these training samples with high uncertainty are basically samples that have not been used for training in history. Because they have not been used for training, the model has a weak ability to distinguish these samples), so that the model's ability can be significantly improved in the subsequent training process.

[0090] For the third type of samples mentioned above, the server can obtain various required samples from the preset sample library to construct an initial sample set, and for each initial sample in the initial sample set, the server can determine the information dimension corresponding to the initial sample based on the initial sample. As shown in the above example, it can be seen that the larger the information dimension, the more dimensions of information the initial sample contains. The "name", "age", "position", etc. in the above example are the dimensions corresponding to the initial sample.

[0091] The server may select initial samples whose information dimension quantity is not less than a second preset threshold from the initial sample set as target samples, and then add the determined target samples to the target sample set.

[0092] For the third type of training samples, in fact, some training samples with richer information are selected. This type of samples can further enhance the problem analysis and problem solving capabilities of the model to be trained for the entire process of model training. Because in the process of analyzing samples, the model to be trained can analyze the potential relationship between information of each dimension based on the rich multi-dimensional information contained in the samples, thereby strengthening the logic of the reasoning process, so that the model to be trained has stronger analysis capabilities after training.

[0093] In an embodiment of the present specification, the server can query samples whose information dimension quantities meet the requirements from the database based on the received query instructions. If the database stores data in the form of structured data, it is convenient for the server to process the data later, so as to better and faster screen out samples that meet the information dimension quantity requirements and add them to the target sample set.

[0094] For the fourth type of samples mentioned above, the server can first determine the historical sample set used to train the model to be trained in the history, and then, for each historical sample in the historical sample set, determine the new information corresponding to the historical sample based on the historical sample, and adjust the historical sample based on the new information to add the adjusted sample to the target sample set.

[0095] Among them, for the new information corresponding to a historical sample, we can first determine the appearance time corresponding to the historical sample, and then combine the information content contained in the historical sample to determine the continuation information for the historical sample that appears after the appearance time, or the information that appears after the appearance time and is highly related to the information content of the historical sample, as the new information corresponding to the historical sample.

[0096] Since the new information is new to the historical samples and has not been used in the model to be trained historically, the addition of this new information can enable the model to be trained to eliminate or correct some of the old "knowledge points" learned during the model training process. At the same time, the addition of new information can enable the model to be trained to correct the "knowledge" that has been understood but is contradictory, thereby preventing the model to be trained from experiencing hallucinations, misunderstandings of facts, etc. in actual applications, and ensuring the accuracy of the model output results.

[0097] For the fifth category of samples mentioned above, the server can obtain an initial sample set, and then, for each initial sample included in the initial sample set, determine the supplementary information that matches the initial sample based on the initial sample, and integrate the supplementary information into the initial sample to add the integrated sample to the target sample set.

[0098] From the above content, it can be seen that the fifth type of sample is actually the server further integrating the initial information based on the initial sample by determining the supplementary information that matches the initial sample, so as to strengthen the initial sample information. Therefore, the fifth type of sample looks similar to the third type of sample, but in fact the two types of samples have different focuses. The third type of sample focuses on the richness of information, that is, the content of information in different dimensions, while the fifth type of sample focuses on the supplement of existing information, such as further information supplement in a certain information dimension.

[0099] In the embodiments of the present specification, the server can use the above five categories of training samples to train the model to be trained, wherein the server can use any one of the above categories of training samples to perform the training task alone, or use multiple categories of training samples to perform the training task at the same time. In the process of using multiple categories of training samples to perform the training task at the same time, the proportion of each category of training samples can be determined according to actual needs, for example, the mixing ratio of each category of samples can be adjusted according to the desire to focus on enhancing the ability of the model to be trained.

[0100] S102: For each training sample in the target sample set, determine the input information corresponding to the training sample and the guidance information corresponding to the input information based on the training sample, wherein the guidance information is used to guide the output of the model to be trained to obtain the logical information on which the output result is based, and the logical information is used to represent the logical reasoning process on which the output result of the model to be trained is based.

[0101] After obtaining the target sample set, the server can further determine the input information corresponding to the training sample and the guidance information corresponding to the input information. The server can directly use the training sample as input information, or further process the training sample to obtain the input information corresponding to the training sample.

[0102] In an embodiment of the present specification, the server can generate a corresponding query instruction based on the training sample, and then, based on the query instruction, the server can query the associated information matching the training sample from a preset information library, and finally, based on the associated information and the training sample, determine the input information corresponding to the training sample.

[0103] In the above process of determining the input information, the preset information library can be pre-built, or it can be an external library, various information websites, etc. In the specific query process, it can be implemented in a variety of ways. For example, the server can match the information that highly matches these search keywords from the preset information library based on the query instruction and the search keywords in the training sample as related information; for another example, after the server generates the query instruction, it can generate the information feature vector corresponding to the training sample according to the query instruction, and also determine the information feature vectors of some candidate information in the preset information library. Finally, by determining the similarity between the information feature vectors, the related information that highly matches the training sample is determined. Other methods are not listed here one by one. This process actually uses retrieval-augmented generation (RAG), that is, by retrieving information associated with the input information from a large amount of data, the analysis ability of the model to be trained is enhanced to improve the accuracy of the output results.

[0104] As for the guidance information mentioned above, it is mainly used to guide the logical information on which the output result obtained by the model to be trained is based. This logical information is used to represent the logical reasoning process on which the output result of the model to be trained is based.

[0105] For example, the guidance information may be: "Please output the results according to the logical reasoning order of first analyzing..., then analyzing..., then analyzing..., and finally analyzing...."

[0106] This guidance information can be determined based on the content of the training sample and the person performing the model training task, or it can be predetermined and saved in a preset database. When the guidance information is needed, the category of the guidance sentence used can be determined based on the content of the training sample, and then the guidance sentence of the corresponding category can be queried from the preset database as the guidance information corresponding to the input information of the training sample.

[0107] S104: Input the input information and the guiding information into the model to be trained, so that the model to be trained determines the initial logic information based on which the output result corresponding to the input information is output according to the input information and the guiding information.

[0108] The server can input the above input information and guidance information into the model to be trained. The model to be trained will determine the initial logic information based on which the output result corresponding to the input information is output based on the guidance of the guidance information and the content of the input information.

[0109] Under the influence of the guiding information, the above-mentioned initial logic information determined by the model to be trained can show the logical reasoning process based on the output result of the model to be trained, and this logical reasoning process can be embodied in the form of the above-mentioned guiding information.

[0110] For example, the initial logic information may be expressed as: “Based on…, it can be determined that… and… have a relationship, then, based on this relationship, it can be determined that… and… also have a relationship, and finally, it can be determined that…”.

[0111] Therefore, this initial logic information can actually be understood as the chain of thought on which the output result of the model to be trained is based, and this chain of thought can reflect the analysis process on which the output result of the model to be trained is based. It should be pointed out that this initial logic information can be displayed in the question-and-answer interface of the model to be trained.

[0112] S106: According to the standard output result corresponding to the input information, the initial logic information is adjusted by the model to be trained to obtain adjusted logic information.

[0113] In the embodiments of this specification, the training samples included in the target sample set may correspond to corresponding standard output results, which may be understood as pre-labeled labels. This standard output result may be used to adjust the initial logic information so that the model to be trained can output results based on a more correct and reasonable logic reasoning process.

[0114] Specifically, the server may analyze the association relationship between the input information and the standard output result based on the standard output result corresponding to the above input information. The specific process may be that the server extracts each key information from the input information, and extracts each key information from the standard output result at the same time, and then associates each key information extracted from the input information with each key information extracted from the standard output result to determine the association relationship between each key information. The server may adjust the initial logical information according to the determined association relationship and the context information of each key information in the standard output result.

[0115] Of course, in actual applications, the server can adjust the initial logic information according to the received adjustment instructions. Specifically, the server can display the standard output result corresponding to the input information to the person performing the model training task through the terminal device used by the person. Based on the standard output result, the person can perform operations on the terminal device used to generate an adjustment request for the initial logic information. The adjustment request may include relevant strategies for how to adjust the initial logic information. After receiving the adjustment request returned by the terminal device, the server can parse the adjustment request and adjust the initial logic information according to the parsed relevant strategies to determine the adjusted logic information.

[0116] In the embodiments of this specification, the adjusted logic information can be displayed through the question-and-answer interface of the model to be trained for viewing by the personnel who perform the model training task. Therefore, in fact, the adjustment of the initial logic information may not be just one round, and the server may make multiple rounds of adjustments to the initial logic information, that is, after the server adjusts the initial logic information based on the standard output result corresponding to the input information, if it is determined that the adjusted logic information does not meet the requirements, the adjusted logic information can be further adjusted based on the standard output result corresponding to the input information, and the adjusted logic information that meets the requirements is finally obtained through continuous iterative adjustment.

[0117] The above determination of whether the adjusted logical information meets the requirements can be determined based on the difference between the output result obtained by the trained model based on the adjusted logical information and the standard output result. If the difference is less than the preset difference, it can be determined that the adjusted logical information meets the requirements. If it is determined that the difference is not less than the preset difference, it can be determined that the adjusted logical information does not meet the requirements.

[0118] Of course, the server can also determine whether the adjusted logical information meets the requirements based on the adjustment request sent by the terminal device used by the personnel. That is, if the server determines that the adjusted logical information needs to be further adjusted based on the received adjustment request, it means that the adjusted logical information does not meet the requirements. Conversely, if the server determines that the adjusted logical information does not need to be further adjusted based on the received adjustment request, or when the server receives a confirmation request for the end of adjustment sent by the terminal device used by the personnel, it can be determined that the adjusted logical information meets the requirements.

[0119] S108: Determine the output result of the model to be trained for the input information according to the adjusted logic information, so as to train the model to be trained according to the output result.

[0120] After determining the above-mentioned adjusted logical information, the server can generate a re-questioning instruction based on the adjusted logical information. The model to be trained can determine the output result of the model to be trained for the input information based on the re-questioning instruction and the adjusted logical information, and then train the model to be trained according to the output result.

[0121] There may be multiple training methods based on the output results mentioned above, such as supervised learning, reinforcement learning, etc. The embodiments of this specification do not limit the specific training method as long as the training requirements are met.

[0122] It should be noted that in the embodiments of the present specification, no matter how many rounds of adjustment are made, the model to be trained can generate an output result corresponding to the input information based on the adjusted logical information or the initial logical information for each round of adjustment. Of course, when determining the initial logical information, the model to be trained may not output the output result determined based on the initial logical information, or the first several rounds of adjustment may not output the corresponding output result, and the subsequent rounds of adjustment may output the results based on the adjusted logical information.

[0123] It can be seen from the above method that the server can first select specific samples for enhancing the various capabilities of the model. These samples can not only enhance the active learning ability of the model to be trained, but also enhance the logical analysis ability and reasoning model of the model to be trained. It can also allow the model to be trained to eliminate or correct some erroneous "knowledge points" that have been learned, and can effectively prevent the occurrence of situations such as hallucinations and misunderstandings of facts.

[0124] Secondly, by adjusting the above-mentioned logical information, the logical reasoning process of the model to be trained can be corrected during the training process, so that in subsequent practical applications, the trained model can output correct, reasonable and accurate results based on the correct and reasonable logical reasoning process, avoiding the influence of the noise carried in the information on the final output results, thereby bringing convenience to users in the process of using the model.

[0125] Finally, in the process of determining the input information, the related information matching the training sample can be queried based on the preset information database. The information content of the training sample can be further enhanced through the queried related information (this process is the RAG mentioned above). In this way, the model to be trained can learn more "knowledge" in the training process based on richer training samples, thereby giving more accurate output results in the subsequent use process.

[0126] In order to facilitate understanding of a model training method provided in the embodiments of this specification, the following will further describe it from the training architecture adopted in the entire model training process, such as Figure 2 shown.

[0127] Figure 2 A schematic diagram of a model training architecture provided in an embodiment of this specification.

[0128] from Figure 2 It can be seen that the architecture of the model to be trained involves three core modules. The first is the retrieval module, which is mainly used to query the associated information matching the input information from the information base to further supplement the input information; the second is the knowledge construction module. The knowledge construction model is mainly used to obtain the required training samples, among which the anchor agent in the knowledge construction module is used to obtain the second type of samples mentioned above, the association agent is used to obtain the fifth type of samples mentioned above, the logic agent is used to obtain the third type of samples mentioned above, and the cognitive agent is used to obtain the fourth type of samples mentioned above; the third is the cognitive module, through which the module to be trained can determine the above-mentioned logical information and output the results based on the determined logical information.

[0129] It should be further pointed out that the above Figure 2 The knowledge building module in the embodiment may also include an agent for obtaining the first type of samples, which is not Figure 2 The knowledge building blocks in Figure 2 It can also be further seen that in practical applications, the associated information retrieved from the information base can be further combined with the various samples obtained in the above-mentioned knowledge construction module to be provided to the subsequent cognitive module.

[0130] The above is a model training method provided in the embodiment of this specification. On this basis, the embodiment of this specification also provides a service execution method. This service execution method is implemented based on the model trained by the above model training method. The specific process is as follows: Figure 3 shown.

[0131] Figure 3 A schematic diagram of a process flow of a service execution provided in an embodiment of this specification includes the following steps:

[0132] S300: receiving input information from a user, where the input information includes at least one of text, picture and audio.

[0133] In the embodiments of this specification, users can use various intelligent models provided by the platform based on their actual needs. The intelligent model mentioned here can be a large language model, or various intelligent entities recorded in the large language model, and of course other models.

[0134] For the use of the large language model, the user can access the question and answer page of the large language model through the terminal device, and then enter the question that requires the large language model on the page. This question is regarded as the user's input information to the large language model.

[0135] Therefore, the terminal device can send the user's input information in the above question-and-answer page to the server, and the server will input the received input information into the model in the subsequent process. The input information can be in various forms, such as text, pictures and audio, or a combination of these information, such as text + pictures, text + audio, etc. The specific form of the input information is related to the business that the user wants to perform.

[0136] S302: Input the input information into a pre-trained target model so that the target model outputs an output result for the input information, and the target model is trained by the above-mentioned model training method.

[0137] After receiving the input information, the server can input the input information into a pre-trained target model (such as the large language model mentioned above), and the target model can analyze and process the input information to obtain a corresponding output result. The target model can be trained using the model training method mentioned above.

[0138] S304: Execute the service according to the output result.

[0139] After the training model obtains the above output results, the terminal device can display the output results, and the user can perform corresponding services based on the output results.

[0140] For specific business execution, the model to be trained displays the output results to the user through the terminal device, which can itself be regarded as a business execution process. Of course, the server can also further process the output results obtained by the model to be trained based on actual business needs to obtain the final business execution results.

[0141] For example, suppose the user needs to perform a task: to determine a gallery that matches the style of the picture based on the picture given by the model. Then, the user inputs a picture and enters a prompt text: "Please adjust the picture to the style of..." The target model can output the picture after the style is adjusted according to the prompt text and the picture. Then, the server can determine the gallery that matches the picture after the style is adjusted based on the picture after the style is adjusted, and display the access addresses of these galleries to the user.

[0142] The above is a model training method and a service execution method provided in one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding model training device and a service execution device, such as Figure 4 , Figure 5 shown.

[0143] Figure 4 A schematic diagram of a model training device provided in an embodiment of this specification includes:

[0144] An acquisition module 400 is used to acquire a target sample set, wherein the training samples included in the target sample set include at least one of text, picture and audio;

[0145] A determination module 402 is used to determine, for each training sample in the target sample set, input information corresponding to the training sample and guidance information corresponding to the input information according to the training sample, wherein the guidance information is used to guide the model to be trained to output logical information based on which the output result is obtained, and the logical information is used to represent the logical reasoning process based on which the output result of the model to be trained is based;

[0146] An input module 404, used for inputting the input information and the guiding information into the model to be trained, so that the model to be trained determines the initial logic information based on which the output result corresponding to the input information is output according to the input information and the guiding information;

[0147] An adjustment module 406, configured to adjust the initial logic information through the model to be trained according to the standard output result corresponding to the input information, to obtain adjusted logic information;

[0148] The training module 408 is used to determine the output result of the model to be trained for the input information according to the adjusted logic information, so as to train the model to be trained according to the output result.

[0149] Optionally, the acquisition module 400 is specifically used to obtain an initial sample set; for each initial sample in the initial sample set, input information corresponding to the initial sample is input into the model to be trained to obtain an output result corresponding to the initial sample, and based on the output result corresponding to the initial sample, determine the uncertainty corresponding to the initial sample; and add the initial samples whose uncertainty is not lower than a first preset threshold to the target sample set.

[0150] Optionally, the acquisition module 400 is specifically used to determine a historical sample set used in history to train the model to be trained; for each historical sample in the historical sample set, determine new information corresponding to the historical sample based on the historical sample, and adjust the historical sample based on the new information to add the adjusted sample to the target sample set, and the new information has not been used in history to train the model to be trained.

[0151] Optionally, the acquisition module 400 is specifically used to acquire an initial sample set; for each initial sample in the initial sample set, determine supplementary information matching the initial sample based on the initial sample, and integrate the supplementary information into the initial sample to add the integrated sample to the target sample set.

[0152] Optionally, the acquisition module 400 is specifically used to obtain an initial sample set; for each initial sample in the initial sample set, determine the information dimension amount corresponding to the initial sample based on the initial sample, the larger the information dimension amount, the more dimensions of information contained in the initial sample; select an initial sample whose information dimension amount is not less than a second preset threshold from the initial sample set as a target sample; and add the target sample to the target sample set.

[0153] Optionally, the determination module 402 is specifically used to generate a query instruction according to the training sample; query the associated information matching the training sample from a preset information library according to the query instruction; and determine the input information corresponding to the training sample according to the associated information and the training sample.

[0154] Figure 5 A schematic diagram of a service execution device provided in an embodiment of this specification includes:

[0155] The receiving module 500 is used to receive input information from a user, where the input information includes at least one of text, picture and audio;

[0156] An input module 502, used for inputting the input information into a pre-trained target model so that the target model outputs an output result for the input information, wherein the target model is trained by the above-mentioned model training method;

[0157] The execution module 504 is used to execute the service according to the output result.

[0158] The above device embodiments correspond to the method embodiments. For specific descriptions, please refer to the description of the method embodiments, which will not be repeated here. The device embodiments are obtained based on the corresponding method embodiments and have the same technical effects as the corresponding method embodiments. For specific descriptions, please refer to the corresponding method embodiments.

[0159] The present specification also provides a computer storage medium, which can store multiple instructions, and the instructions are suitable for being loaded and executed by a processor as described above. Figure 1 or Figure 3 The method of the embodiment shown in the figure can be specifically executed by referring to Figure 1 or Figure 3 The specific description of the illustrated embodiment will not be repeated here.

[0160] The present specification also provides a computer program product, which stores at least one instruction, and the at least one instruction is loaded and executed by the processor as described above. Figure 1 or Figure 3 The method of the embodiment shown in the figure can be specifically executed by referring to Figure 1 or Figure 3 The specific description of the illustrated embodiment will not be repeated here.

[0161] The embodiments of this specification also provide Figure 6 The structural diagram of the electronic device shown in FIG. Figure 6 At the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above-mentioned model training method and business execution method.

[0162] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the executor of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0163] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0164] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0165] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0166] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0167] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0168] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0169] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0170] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0171] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0172] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0173] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0174] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0175] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0176] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0177] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0178] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A model training method, comprising: Acquire a target sample set, wherein the training samples contained in the target sample set include at least one of text, picture and audio; For each training sample in the target sample set, determine input information corresponding to the training sample and guidance information corresponding to the input information according to the training sample, wherein the guidance information is used to guide the model to be trained to output logical information based on which the output result is obtained, and the logical information is used to represent the logical reasoning process based on which the output result of the model to be trained is based; Inputting the input information and the guiding information into the model to be trained, so that the model to be trained determines the initial logic information based on which the output result corresponding to the input information is output according to the input information and the guiding information; According to the standard output result corresponding to the input information, the initial logic information is adjusted by the model to be trained to obtain adjusted logic information; According to the adjusted logic information, the output result of the model to be trained for the input information is determined, so as to train the model to be trained according to the output result.

2. The method according to claim 1, wherein obtaining the target sample set comprises: Obtain an initial sample set; For each initial sample in the initial sample set, input information corresponding to the initial sample is input into the model to be trained to obtain an output result corresponding to the initial sample, and the uncertainty corresponding to the initial sample is determined according to the output result corresponding to the initial sample; The initial samples whose uncertainty is not less than the first preset threshold are added to the target sample set.

3. The method according to claim 1, wherein obtaining the target sample set specifically comprises: Determine a historical sample set used in the history to train the model to be trained; For each historical sample in the historical sample set, new information corresponding to the historical sample is determined based on the historical sample, and the historical sample is adjusted based on the new information to add the adjusted sample to the target sample set, and the new information has not been used in the training of the model to be trained in history.

4. The method according to claim 1, wherein obtaining the target sample set comprises: Obtain an initial sample set; For each initial sample in the initial sample set, supplementary information matching the initial sample is determined based on the initial sample, and the supplementary information is integrated into the initial sample to add the integrated sample to the target sample set.

5. The method according to claim 1, wherein obtaining the target sample set specifically comprises: Obtain an initial sample set; For each initial sample in the initial sample set, determine the information dimension quantity corresponding to the initial sample according to the initial sample, wherein the larger the information dimension quantity is, the more dimensions of information the initial sample contains; Selecting, from the initial sample set, initial samples whose information dimension is not less than a second preset threshold as target samples; The target sample is added to the target sample set.

6. The method according to claim 1, wherein determining the input information corresponding to the training sample according to the training sample specifically comprises: Generate a query instruction according to the training sample; According to the query instruction, query the associated information matching the training sample from a preset information database; According to the association information and the training sample, input information corresponding to the training sample is determined.

7. A business execution method, comprising: Receiving input information from a user, wherein the input information includes at least one of text, picture, and audio; Inputting the input information into a pre-trained target model so that the target model outputs an output result for the input information, wherein the target model is trained by the method according to any one of claims 1 to 6; Execute the service according to the output result.

8. A model training device, comprising: An acquisition module, used to acquire a target sample set, wherein the training samples contained in the target sample set include at least one of text, picture and audio; A determination module, for determining, for each training sample in the target sample set, input information corresponding to the training sample and guidance information corresponding to the input information according to the training sample, wherein the guidance information is used to guide the output of the model to be trained to obtain logic information based on which the output result is obtained, and the logic information is used to represent the logic reasoning process based on which the output result of the model to be trained is based; An input module, used for inputting the input information and the guiding information into the model to be trained, so that the model to be trained determines the initial logic information based on which the output result corresponding to the input information is output according to the input information and the guiding information; An adjustment module, configured to adjust the initial logic information through the model to be trained according to a standard output result corresponding to the input information, so as to obtain adjusted logic information; A training module is used to determine the output result of the model to be trained for the input information according to the adjusted logic information, so as to train the model to be trained according to the output result.

9. In the device as described in claim 8, the acquisition module is specifically used to obtain an initial sample set; for each initial sample in the initial sample set, input information corresponding to the initial sample is input into the model to be trained to obtain an output result corresponding to the initial sample, and based on the output result corresponding to the initial sample, the uncertainty corresponding to the initial sample is determined; and the initial samples whose uncertainty is not lower than a first preset threshold are added to the target sample set.

10. In the device as described in claim 8, the acquisition module is specifically used to determine the historical sample set used in the history to train the model to be trained; for each historical sample in the historical sample set, determine the new information corresponding to the historical sample based on the historical sample, and adjust the historical sample based on the new information to add the adjusted sample to the target sample set, and the new information has not been used in the history to train the model to be trained.

11. In the device as described in claim 8, the acquisition module is specifically used to obtain an initial sample set; for each initial sample in the initial sample set, determine the supplementary information matching the initial sample based on the initial sample, and integrate the supplementary information into the initial sample to add the integrated sample to the target sample set.

12. In the device as described in claim 8, the acquisition module is specifically used to obtain an initial sample set; for each initial sample in the initial sample set, determine the information dimension amount corresponding to the initial sample based on the initial sample, the larger the information dimension amount, the more dimensions of information contained in the initial sample; select an initial sample whose information dimension amount is not less than a second preset threshold from the initial sample set as a target sample; and add the target sample to the target sample set.

13. A service execution device, comprising: A receiving module, used to receive input information from a user, wherein the input information includes at least one of text, picture and audio; An input module, used for inputting the input information into a pre-trained target model so that the target model outputs an output result for the input information, wherein the target model is trained by the method according to any one of claims 1 to 6; An execution module is used to execute the business according to the output result.

14. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method according to any one of claims 1 to 7.

15. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the program.

Images