A General Large Model Local Deployment Processing Method and System

Abstract

The present invention provides a method and system for local deployment and processing of a general large model, belonging to the technical field of data processing. Specifically, it includes: taking the dates not in the similar demand date group as deviation demand dates, obtaining the historical interaction data volume and distribution data of the deviation demand dates, and combining the deviation situations of the user demands of different deviation demand dates and the similar demand date group. When it is determined that the large model does not need to perform enhanced training processing of user demands in a preset manner, the user demands are divided into different user demand groups, and the enhanced training processing methods of different user demand groups are determined based on the distribution data of the user demands of different user demand groups on different dates and the recognition deviation data in the large model, thereby improving the personalized processing of the enhanced training of the large model.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a method and system for local deployment and processing of a general large model. Background Art

[0002] With the popularization and application of large models in various industries, in order to ensure data security, more and more enterprises tend to localize the deployment. Specifically, in the invention patent application CN202410064380.9, "Method, Device, System, Equipment and Storage Medium for Localization Processing of In-vehicle Large Model", the in-vehicle large model is localize deployed, and by sharing the computing power of the user terminal, the timeliness and accuracy of vehicle terminal interaction are improved. However, there are the following technical defects:

[0003] During the process of localizing the deployment of general large models, reinforcement learning and reward engineering are widely used to improve the decision-making ability and logical reasoning ability of the models. Since there is a certain degree of deviation between the reinforcement learning strategies of large models and the requirements of different enterprises, how to customize a reinforcement learning solution according to the actual needs of enterprises has become a technical problem to be solved urgently.

[0004] To solve the above technical problems, specifically, the present application provides a method and system for local deployment and processing of a general large model. Summary of the Invention

[0005] Specifically, to achieve the purpose of the present invention, the present invention adopts the following technical solutions:

[0006] Specifically, the present application provides a method for local deployment and processing of a general large model, which specifically includes:

[0007] S1 Based on the historical interaction data of the large model deployed locally on different dates, when it is determined that the difficulty of performing reinforcement training processing on user requirements by using a preset method does not meet the requirements, proceed to the next step;

[0008] S2 Based on the analysis results of the historical interaction data of the large model on different dates, determine the similarity of user requirements on different dates, and use the similarity to divide the dates into different similar requirement date groups;

[0009] S3 Take the dates not in the similar requirement date group as deviation requirement dates, obtain the historical interaction data volume and distribution data of the deviation requirement dates, and combine the deviation situations of different deviation requirement dates from the user requirements of the similar requirement date group. When it is determined that the large model does not need to perform reinforcement training processing on user requirements by using a preset method, proceed to the next step;

[0010] S4 divides user requirements into different user requirement groups, and determines the reinforcement training processing methods for different user requirement groups based on the distribution data of the user requirements of different user requirement groups on different dates and the recognition deviation data in the large model.

[0011] The beneficial effects of the present invention are as follows:

[0012] Based on the historical interaction data volume and distribution data of the deviation requirement dates, and the deviation situations between different deviation requirement dates and the user requirements of the similar requirement date groups, it is determined whether the large model needs to perform reinforcement training processing on user requirements in a preset manner, thereby fully considering the differences in the proportion of the number of dates with deviated user requirements and the interaction data volume, as well as the differences in the requirements for the reinforcement training processing of the large model for user requirements, avoiding the technical problem of the inability to meet the reliability of the recognition processing of user requirements caused by untimely reinforcement training processing, and realizing the determination of a differential reinforcement training processing strategy from the perspective of the changes in user requirements.

[0013] Based on the distribution data of the user requirements of different user requirement groups on different dates and the recognition deviation data in the large model, the reinforcement training processing methods for different user requirement groups are determined. This not only considers the differences in the requirements for the reinforcement training processing of user requirement groups due to the differences in recognition deviation data, but also further considers the distribution data of user requirements on different dates, realizing the comprehensive consideration of the recognition frequency of the interaction processing of user requirement groups, and then realizing the differential reinforcement training processing of different user requirement groups. On the basis of ensuring the accuracy of the recognition processing, it also reduces the training resource requirements for the reinforcement training processing.

[0014] A further technical solution lies in that the historical interaction data includes the historical interaction times on different dates and the interaction data volume of different historical interaction times.

[0015] A further technical solution lies in that determining that the difficulty of performing reinforcement training processing on user requirements in a preset manner does not meet the requirements specifically includes:

[0016] Based on the historical interaction data on different dates, determine the historical interaction data volume on different dates;

[0017] According to the historical interaction data volume on different dates, determine the dates in which the historical interaction data volume is greater than the preset interaction data volume, and use them as the interaction busy dates;

[0018] Based on the proportion of the number of the interaction busy dates, determine whether the difficulty of performing reinforcement training processing on user requirements in a preset manner meets the requirements.

[0019] A further technical solution is that when the proportion of the number of the interaction busy dates is greater than a preset busy number proportion, it is determined that the difficulty of using a preset method to perform enhanced training processing on user requirements does not meet the requirements.

[0020] A further technical solution is that when the difficulty of using a preset method to perform enhanced training processing on user requirements meets the requirements, the preset method is used to perform enhanced training processing on the user's repairs.

[0021] A further technical solution lies in the method for determining the enhanced training processing method of the user requirement group:

[0022] Based on the distribution data of the user requirements of the user requirement group on different dates, determine the proportion of the number of dates when the user requirements of the user requirement group exist, and use it as the matching requirement number proportion;

[0023] According to the recognition deviation data of the user requirements of the user requirement group, determine the proportion of the number of recognition deviation times of the user requirements of the user requirement group, and use it as the recognition deviation times proportion;

[0024] Based on the average value of the matching requirement number proportion and the recognition deviation times proportion, determine the enhanced training requirement coefficient of the user requirement group, and use the enhanced training requirement coefficient to determine the enhanced training processing method of the user requirement group.

[0025] A further technical solution is that using the enhanced training requirement coefficient to determine the enhanced training processing method of the user requirement group specifically includes:

[0026] When the enhanced training requirement coefficient of the user requirement group is greater than a preset training requirement coefficient threshold, it is determined that it is necessary to use a preset method to perform enhanced training processing on the user requirements of the user requirement group;

[0027] When the enhanced training requirement coefficient of the user requirement group is not greater than the preset training requirement coefficient threshold, it is also necessary to determine whether the enhanced training requirement coefficient of the user requirement group is less than the preset requirement coefficient threshold. If so, there is no need to perform enhanced training processing on the user requirements of the user requirement group. If not, it is determined that it is necessary to use a second preset method to perform enhanced training processing on the user requirements of the user requirement group.

[0028] A further technical solution is that the preset method is to perform enhanced training processing on user requirements every day, and the second preset method is to perform enhanced training processing on the user requirements of the user requirement group when the historical interaction data volume of the user requirements of the user requirement group reaches a preset interaction data volume threshold.

[0029] In a second aspect, the present invention provides a computer system, comprising: a memory and a processor communicatively connected, and a computer program stored on the memory and capable of running on the processor, wherein when the processor runs the computer program, it executes the above-mentioned general large model local deployment processing method.

[0030] Other features and advantages will be described in the following specification. The objectives and other advantages of the present invention are achieved and obtained by the structure specifically pointed out in the specification and the accompanying drawings.

[0031] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, provides a detailed description as follows. Description of the Drawings

[0032] By referring to the drawings and describing its exemplary embodiments in detail, the above and other features and advantages of the present invention will become more obvious.

[0033] Figure 1 is a flowchart of a general large model local deployment processing method;

[0034] Figure 2 is a flowchart for determining that the difficulty of the enhanced training process for user requirements using a preset method does not meet the requirements;

[0035] Figure 3 is a flowchart for dividing dates into different groups of similar requirement dates;

[0036] Figure 4 is a flowchart of a method for determining an enhanced training processing method for a user requirement group;

[0037] Figure 5 is a framework diagram of a computer system. Detailed Embodiments

[0038] In order to enable those skilled in the art of the present technology to better understand the technical solutions in this specification, the following will, in conjunction with the accompanying drawings in the embodiments of this specification, clearly and completely describe the technical solutions in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this specification.

[0039] In the present application, according to the usage requirements of large models deployed locally in enterprises, different types of usage requirements are used to generate differentiated personalized enhanced training processing strategies, thereby reducing both the training resource requirements for enhanced training and enabling the recognition processing accuracy of the large model to meet the requirements.

[0040] Example 1 As Figure 1 shown, this application provides a method for local deployment and processing of a general large model, specifically including:

[0041] S1 Based on the historical interaction data of the large model deployed locally on different dates, when it is determined that the difficulty of strengthening training and processing of user requirements by using a preset method does not meet the requirements, proceed to the next step;

[0042] It should be noted that when the average value of the historical interaction data volume on different dates is greater than the preset interaction data volume, it is determined that the difficulty of strengthening training and processing of user requirements by using a preset method does not meet the requirements.

[0043] S2 Based on the analysis results of the historical interaction data of the large model on different dates, determine the similarity of user requirements on different dates, and use the similarity to divide the dates into different similar requirement date groups;

[0044] Specifically, based on the number of similar demand feature vectors corresponding to user requirements with different historical interaction times between different dates, use the number of historical interaction times with more than 5 similar numbers as the similar interaction times between different dates, and use the product of the proportion of similar interaction times in the historical interaction times on different dates to determine the interaction similarity coefficient between different dates. Divide the dates with an interaction similarity coefficient greater than 0.3 into the same similar requirement group.

[0045] S3 Use the dates not in the similar requirement date group as deviation requirement dates, obtain the historical interaction data volume and distribution data of the deviation requirement dates, and combine the deviation situations of the different deviation requirement dates and the user requirements of the similar requirement date group. When it is determined that the large model does not need to use a preset method to strengthen the training and processing of user requirements, proceed to the next step;

[0046] It should be noted that use the proportion of the deviated interaction times in the historical interaction times of different deviation requirement dates and the similar requirement group to determine the deviation weight coefficient of different deviation requirement dates. According to the product of the average value of the deviation weight coefficients of different deviation requirement dates and the proportion of the number of deviation requirement dates, determine the date demand variation coefficient. When the date demand variation coefficient is greater than 0.6, it is determined that the large model needs to use a preset method to strengthen the training and processing of user requirements.

[0047] S4 Divide the user requirements into different user requirement groups, and determine the strengthening training and processing methods for different user requirement groups based on the distribution data of the user requirements of different user requirement groups on different dates and the recognition deviation data in the large model.

[0048] Specifically, based on the proportion of the number of dates of the user requirements in the user requirement group and the average proportion of the number of times of identification deviation of the user requirements in the user requirement group, determine the reinforcement training requirement coefficient of the user requirement group, and use the reinforcement training requirement coefficient to determine the reinforcement training processing method for the user requirement group.

[0049] When the reinforcement training requirement coefficient of the user requirement group is greater than 0.6, it is determined that the reinforcement training of the user requirements in the user requirement group needs to be processed in a preset manner;

[0050] When the reinforcement training requirement coefficient of the user requirement group is not greater than 0.6, it is also necessary to determine whether the reinforcement training requirement coefficient of the user requirement group is less than 0.1. If so, there is no need to perform the reinforcement training of the user requirements in the user requirement group. If not, it is determined that the reinforcement training of the user requirements in the user requirement group needs to be processed in a second preset manner.

[0051] Furthermore, the historical interaction data includes the historical interaction times on different dates and the interaction data volume of different historical interaction times.

[0052] It should be noted that as Figure 2 shown, the difficulty of determining to perform the reinforcement training of the user requirements in a preset manner does not meet the requirements, specifically including:

[0053] Based on the historical interaction data on different dates, determine the historical interaction data volume on different dates;

[0054] According to the historical interaction data volume on different dates, determine the dates with the historical interaction data volume greater than the preset interaction data volume in the dates, and use them as the interaction busy dates;

[0055] Based on the proportion of the number of the interaction busy dates, determine whether the difficulty of performing the reinforcement training of the user requirements in a preset manner meets the requirements.

[0056] Specifically, when the proportion of the number of the interaction busy dates is greater than the preset busy proportion, it is determined that the difficulty of performing the reinforcement training of the user requirements in a preset manner does not meet the requirements.

[0057] It can be understood that when the difficulty of performing the reinforcement training of the user requirements in a preset manner meets the requirements, the reinforcement training of the user requirements is processed in a preset manner.

[0058] In another possible embodiment, the difficulty of determining to perform the reinforcement training of the user requirements in a preset manner does not meet the requirements, specifically including:

[0059] Based on the historical interaction data on different dates, determine the amount of historical interaction data on different dates;

[0060] According to the amount of historical interaction data on different dates, determine the average value of the amount of historical interaction data on different dates;

[0061] Based on the average value of the amount of historical interaction data on different dates, determine whether the difficulty of the enhanced training process for user requirements using a preset method meets the requirements.

[0062] Furthermore, when the average value of the amount of historical interaction data on different dates is greater than the preset interaction data amount, it is determined that the difficulty of the enhanced training process for user requirements using a preset method does not meet the requirements.

[0063] Optionally, determining that the difficulty of the enhanced training process for user requirements using a preset method does not meet the requirements specifically includes:

[0064] Based on the historical interaction data on different dates, determine the amount of historical interaction data on different dates. When the amount of historical interaction data on different dates is less than the preset interaction data amount threshold, it is determined that the difficulty of the enhanced training process for user requirements using a preset method meets the requirements;

[0065] When there are dates with the amount of historical interaction data not less than the preset interaction data amount threshold:

[0066] According to the amount of historical interaction data on different dates, determine the average value of the amount of historical interaction data on different dates. When the average value of the amount of historical interaction data on different dates is greater than the preset interaction data amount, it is determined that the difficulty of the enhanced training process for user requirements using a preset method does not meet the requirements;

[0067] When the average value of the amount of historical interaction data on different dates is not greater than the preset interaction data amount:

[0068] According to the amount of historical interaction data on different dates, determine the dates with the amount of historical interaction data greater than the preset interaction data amount in the said dates and use them as interaction busy dates. When the proportion of the number of the interaction busy dates is greater than the preset busy number proportion, it is determined that the difficulty of the enhanced training process for user requirements using a preset method does not meet the requirements;

[0069] When the proportion of the number of the interaction busy dates is not greater than the preset busy number proportion:

[0070] Based on the historical interaction times on different dates and the amount of interaction data for different historical interaction times, determine the enhanced processing difficulty coefficient on different dates. When the average value of the enhanced processing difficulty coefficients on different dates does not meet the requirements, it is determined that the difficulty of the enhanced training process for user requirements using a preset method does not meet the requirements;

[0071] When the average value of the intensification processing difficulty coefficients on different dates meets the requirements:

[0072] Obtain the dates when the intensification processing difficulty coefficient is greater than the preset difficulty coefficient threshold. When the proportion of the number of dates when the intensification processing difficulty coefficient is greater than the preset difficulty coefficient threshold does not meet the requirements, it is determined that the difficulty of using the preset method for the intensification training processing of user requirements does not meet the requirements;

[0073] When the proportion of the number of dates when the intensification processing difficulty coefficient is greater than the preset difficulty coefficient threshold meets the requirements:

[0074] Determine the comprehensive processing difficulty coefficient based on the intensification processing difficulty coefficients on different dates, and use the comprehensive processing difficulty coefficient to determine whether the difficulty of using the preset method for the intensification training processing of user requirements meets the requirements.

[0075] It should be noted that when the comprehensive processing difficulty coefficient is greater than the preset processing difficulty coefficient threshold, it is determined that the difficulty of using the preset method for the intensification training processing of user requirements meets the requirements.

[0076] Furthermore, the similarity situation of the user requirements is determined according to the number of similar demand feature vectors corresponding to the user requirements.

[0077] Specifically, as Figure 3 shown, divide the dates into different similar demand date groups, which specifically include:

[0078] Determine the number of similar demand feature vectors corresponding to the user requirements with different historical interaction times between different dates according to the similarity situation of the user requirements between different dates;

[0079] Determine the similar interaction times between different dates according to the number of similarities, and determine the interaction similarity coefficient between different dates by the product of the proportion of the similar interaction times in the historical interaction times in different dates;

[0080] Divide the dates with the interaction similarity coefficient greater than the preset similarity coefficient threshold into the same similar demand group.

[0081] Furthermore, the similar interaction times are the historical interaction times when the number of similarities is greater than the preset number of similar thresholds.

[0082] It can be understood that the deviation situation of the user requirements of the deviation demand date from the similar demand date group includes the number of deviation demand feature vectors corresponding to the user requirements with different historical interaction times between the deviation demand date and different similar demand date groups.

[0083] Specifically, determining that the large model does not need to perform enhanced training processing on user requirements in a preset manner specifically includes:

[0084] Based on the distribution data of the deviation demand dates, determine the quantity proportion of the deviation demand dates, and use it as the proportion of the number of deviation dates;

[0085] According to the deviation situation of the user requirements between the deviation demand dates and the similar demand date groups, determine the number of deviations of the demand feature vectors corresponding to the historical interaction times of different dates in the similar date groups, and use the number of deviations to determine the deviation interaction times with different dates in the similar date groups. Based on the proportion of the number of deviation interaction times in the deviation demand dates, determine the demand deviation coefficients for different dates;

[0086] Determine the average deviation coefficient of the deviation demand dates through the average value of the demand deviation coefficients for different dates. Determine the preset training demand coefficients for different deviation demand dates based on the historical interaction data volume of different deviation demand dates. Determine the deviation weight coefficient using the average value of the product of the average deviation coefficient and the preset training demand coefficient of different deviation demand dates;

[0087] Use the product of the deviation weight coefficient and the proportion of the number of deviation dates to determine the date demand change coefficient, and use the date demand change coefficient to determine whether the large model needs to perform enhanced training processing on user requirements in a preset manner.

[0088] Furthermore, when the date demand change coefficient is greater than the preset change coefficient threshold, it is determined that the large model needs to perform enhanced training processing on user requirements in a preset manner.

[0089] It should be noted that when the large model needs to perform enhanced training processing on user requirements in a preset manner, different user requirements are processed by enhanced training in a preset manner.

[0090] Optionally, determining that the large model does not need to perform enhanced training processing on user requirements in a preset manner specifically includes:

[0091] According to the deviation situation of the user requirements between the deviation demand dates and the similar demand date groups, determine the number of deviations of the demand feature vectors corresponding to the historical interaction times of different dates in the similar date groups, and use the number of deviations to determine the deviation interaction times with different dates in the similar date groups. Based on the proportion of the number of deviation interaction times in the deviation demand dates, determine the demand deviation coefficients for different dates;

[0092] Determine the mean deviation coefficient of the deviation demand date by the average of the demand deviation coefficients on different dates, determine the preset training demand coefficient for different deviation demand dates based on the historical interaction data volume of different deviation demand dates, and determine the deviation weight coefficient by the average of the products of the mean deviation coefficient of different deviation demand dates and the preset training demand coefficient;

[0093] Based on the distribution data of the deviation demand date, determine the quantity of the deviation demand date, sum the deviation weight coefficients of different deviation demand dates to determine the deviation weight value, and use the deviation weight value to determine whether the large model needs to perform enhanced training processing on user demands in a preset manner.

[0094] Further, when the deviation weight value is greater than the preset deviation weight threshold, it is determined that enhanced training processing on user demands needs to be performed in a preset manner.

[0095] Optionally, determining that the large model does not need to perform enhanced training processing on user demands in a preset manner specifically includes:

[0096] Based on the distribution data of the deviation demand date, determine the quantity of the deviation demand date. When the quantity of the deviation demand date is less than the preset deviation date quantity threshold, it is determined that the large model does not need to perform enhanced training processing on user demands in a preset manner;

[0097] When the quantity of the deviation demand date is not less than the preset deviation date quantity threshold:

[0098] Determine the proportion of the quantity of the deviation demand date and use it as the deviation date quantity proportion. When the deviation date quantity proportion or the total sum of the historical interaction data volume of the deviation demand date does not meet the requirements, it is determined that the large model needs to perform enhanced training processing on user demands in a preset manner;

[0099] When the deviation date quantity proportion and the total sum of the historical interaction data volume of the deviation demand date meet the requirements:

[0100] Based on the deviation situation of the user demands between the deviation demand date and the similar demand date group, determine the deviation quantity of the demand feature vectors corresponding to the historical interaction times of different dates in the similar date group, and use the deviation quantity to determine the deviation interaction times with different dates in the similar date group. Based on the proportion of the deviation interaction times in the quantity of the deviation demand date, determine the demand deviation coefficient for different dates. Determine the mean deviation coefficient of the deviation demand date by the average of the demand deviation coefficients for different dates. When the proportion of the quantity of the deviation demand dates whose mean deviation coefficient does not meet the requirements is greater than the preset deviation date quantity proportion, it is determined that the large model needs to perform enhanced training processing on user demands in a preset manner;

[0101] When the proportion of the number of deviation demand dates whose mean deviation coefficient does not meet the required deviation demand dates is not greater than the preset proportion of deviation date numbers:

[0102] Determine the preset training demand coefficients for different deviation demand dates based on the historical interaction data volumes of different deviation demand dates, and determine the deviation weight coefficient using the average value of the products of the mean deviation coefficients and the preset training demand coefficients of different deviation demand dates. When the sum of the deviation weight coefficients of different deviation demand dates does not meet the requirements, it is determined that the large model needs to perform enhanced training processing on user demands in a preset manner;

[0103] When the sum of the deviation weight coefficients of different deviation demand dates meets the requirements:

[0104] Determine the date demand change coefficient using the product of the deviation weight coefficient and the proportion of deviation date numbers, and use the date demand change coefficient to determine whether the large model needs to perform enhanced training processing on user demands in a preset manner.

[0105] Furthermore, divide user demands into different user demand groups, specifically including:

[0106] Based on the number of similar demand feature vectors corresponding to the user demands, determine user demands with a similar number greater than the preset vector similarity number;

[0107] Divide user demands with a similar number greater than the preset vector similarity number into the same user demand group.

[0108] Specifically, the demand feature vector corresponding to the user demand is determined based on the extraction result of the keywords of the user demand.

[0109] Furthermore, the identified deviation data includes the data interaction times of the number of identified deviation times and the number of successful identification times of the user demand group in the large model.

[0110] It should be noted that, as Figure 4 shown, the method for determining the enhanced training processing method of the user demand group:

[0111] Based on the distribution data of the user demands of the user demand group on different dates, determine the proportion of the number of dates with the user demands of the user demand group, and use it as the matching demand proportion;

[0112] Based on the identified deviation data of the user demands of the user demand group, determine the proportion of the number of identified deviation times of the user demands of the user demand group, and use it as the identified deviation times proportion;

[0113] Based on the average value of the proportion of the matching demand quantity and the proportion of the number of recognition deviation times, determine the reinforcement training demand coefficient of the user demand group, and use the reinforcement training demand coefficient to determine the reinforcement training processing method for the user demand group.

[0114] Further, using the reinforcement training demand coefficient to determine the reinforcement training processing method for the user demand group specifically includes:

[0115] When the reinforcement training demand coefficient of the user demand group is greater than the preset training demand coefficient threshold, it is determined that the reinforcement training processing of the user demands of the user demand group needs to be carried out in a preset manner;

[0116] When the reinforcement training demand coefficient of the user demand group is not greater than the preset training demand coefficient threshold, it is also necessary to determine whether the reinforcement training demand coefficient of the user demand group is less than the preset demand coefficient threshold. If so, there is no need to carry out the reinforcement training processing of the user demands of the user demand group. If not, it is determined that the reinforcement training processing of the user demands of the user demand group needs to be carried out in a second preset manner.

[0117] It can be understood that the preset manner is to carry out the reinforcement training processing of the user demands every day, and the second preset manner is to carry out the reinforcement training processing of the user demands of the user demand group when the historical interaction data volume of the user demands of the user demand group reaches the preset interaction data volume threshold.

[0118] Optionally, the method for determining the reinforcement training processing method of the user demand group:

[0119] S41 Based on the distribution data of the user demands of the user demand group on different dates, determine the proportion of the number of dates with the user demands of the user demand group, and use it as the proportion of the matching demand quantity. Combining the historical interaction times of the user demands of the user demand group on different dates, determine the interaction processing busy coefficient of the user demands of the user demand group;

[0120] S42 According to the recognition deviation data of the user demands of the user demand group, determine the proportion of the number of recognition deviation times of the user demands of the user demand group, and use it as the proportion of the recognition deviation times. Combining the number of recognition deviation times and the data interaction times of different recognition success times, determine the deviation recognition demand coefficient of the user demand group;

[0121] S43 Based on the interaction processing busy coefficient and the deviation recognition demand coefficient, determine the reinforcement training demand coefficient of the user demand group, and use the reinforcement training demand coefficient to determine the reinforcement training processing method for the user demand group.

[0122] Optionally, the enhanced training requirement coefficient of the user requirement group is determined according to the product of the interaction processing busy coefficient and the deviation identification requirement coefficient.

[0123] Embodiment 2 In a second aspect, as Figure 5 shown, the present invention provides a computer system, including: a memory and a processor connected by communication, and a computer program stored on the memory and capable of running on the processor, and when the processor runs the computer program, it executes the above-mentioned general large model local deployment processing method.

[0124] Optionally, the following content is included in the above step S41:

[0125] S411 uses the distribution data of the user requirements of the user requirement group on different dates to determine the proportion of the number of dates with the user requirements of the user requirement group, and takes it as the matching requirement quantity proportion. When the matching requirement quantity proportion is greater than the preset matching date proportion threshold, it is determined that the enhanced training process of the user requirements of the user requirement group needs to be carried out in a preset manner. When the matching requirement quantity proportion is not greater than the preset matching date proportion threshold, it proceeds to step S412;

[0126] S412 When the matching requirement quantity proportion is within the preset requirement quantity proportion interval, it is determined that the enhanced training process of the user requirements of the user requirement group does not need to be carried out. When the matching requirement quantity proportion is not within the preset requirement quantity proportion interval, it proceeds to step S413;

[0127] S413 uses the historical interaction times of the user requirements of the user requirement group on different dates to determine the dates with the historical interaction times greater than the preset interaction times threshold. When the number of dates with the historical interaction times greater than the preset interaction times threshold is greater than the preset interaction date quantity threshold, it is determined that the enhanced training process of the user requirements of the user requirement group needs to be carried out in a preset manner. When the number of dates with the historical interaction times greater than the preset interaction times threshold is not greater than the preset interaction date quantity threshold, it proceeds to step S414;

[0128] S414 determines the interaction processing busy coefficient of the user requirements of the user requirement group according to the matching requirement quantity proportion and the historical interaction times of the user requirements of the user requirement group on different dates. When the interaction processing busy coefficient of the user requirements of the user requirement group is greater than the preset processing busy coefficient threshold, it is determined that the enhanced training process of the user requirements of the user requirement group needs to be carried out in a preset manner. When the interaction processing busy coefficient of the user requirements of the user requirement group is not greater than the preset processing busy coefficient threshold, it proceeds to step S415;

[0129] S415 When the interaction processing busy coefficient of the user requirements in the user requirements group is less than the preset busy coefficient threshold, it is determined that the enhanced training process for the user requirements in the user requirements group is not required. When the interaction processing busy coefficient of the user requirements in the user requirements group is not less than the preset busy coefficient threshold, proceed to step S42.

[0130] Optionally, the following content is included in step S42 above:

[0131] S421 According to the recognition deviation data of the user requirements in the user requirements group, determine the proportion of the number of recognition deviations of the user requirements in the user requirements group, and use it as the proportion of recognition deviation times. When the proportion of recognition deviation times does not meet the requirements, it is determined that the enhanced training process for the user requirements in the user requirements group needs to be carried out in a preset manner. When the proportion of recognition deviation times meets the requirements, proceed to step S422;

[0132] S422 When the number of successful recognitions of the user requirements in the user requirements group is less than the preset successful recognition threshold, proceed to step S424. When the number of successful recognitions of the user requirements in the user requirements group is not less than the preset successful recognition threshold, proceed to step S423;

[0133] S423 Based on the data interaction times corresponding to different numbers of successful recognitions, determine the recognition processing deviation coefficient. When the recognition deviation processing deviation coefficient meets the requirements, it is determined that the enhanced training process for the user requirements in the user requirements group needs to be carried out in a second preset manner. When the recognition deviation processing deviation coefficient does not meet the requirements, proceed to step S424;

[0134] S424 Using the proportion of recognition deviation times, the number of recognition deviations, and the data interaction times corresponding to different numbers of successful recognitions, determine the deviation recognition requirement coefficient of the user requirements group. When the deviation recognition requirement coefficient of the user requirements group is greater than the preset deviation recognition requirement coefficient threshold, it is determined that the enhanced training process for the user requirements in the user requirements group needs to be carried out in a second preset manner. When the deviation recognition requirement coefficient of the user requirements group is not greater than the preset deviation recognition requirement coefficient threshold, proceed to step S43.

[0135] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non - volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments.

[0136] The above description has been made of specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0137] The above is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, various changes and modifications can be made to one or more embodiments of this specification. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of one or more embodiments of this specification shall be included within the scope of the claims of this specification.

Claims

1. A method for local deployment and processing of a general large model, characterized in that, Specifically, it includes: Based on the historical interaction data of the locally deployed large model on different dates, when it is determined that the difficulty of using the preset method for enhanced training processing of user requirements does not meet the requirements, proceed to the next step; Based on the analysis results of the historical interaction data of the large model on different dates, determine the similarity of user requirements on different dates, and use the similarity to divide the dates into different similar requirement date groups; Take the dates not in the similar requirement date group as deviation requirement dates, obtain the historical interaction data volume and distribution data of the deviation requirement dates, and combine the deviation situations of different deviation requirement dates and the user requirements of the similar requirement date groups. When it is determined that the large model does not need to use the preset method for enhanced training processing of user requirements, proceed to the next step; Divide user requirements into different user requirement groups, and determine the enhanced training processing methods for different user requirement groups based on the distribution data of the user requirements of different user requirement groups on different dates and the recognition deviation data in the large model.

2. The general large model local deployment processing method according to claim 1, wherein, The historical interaction data includes the historical interaction times on different dates and the interaction data volume of different historical interaction times.

3. The general large model local deployment processing method according to claim 1, wherein Determining that the difficulty of using the preset method for enhanced training processing of user requirements does not meet the requirements specifically includes: Based on the historical interaction data on different dates, determine the historical interaction data volume on different dates; According to the historical interaction data volume on different dates, determine the dates in which the historical interaction data volume is greater than the preset interaction data volume, and take them as interaction busy dates; Based on the proportion of the number of interaction busy dates, determine whether the difficulty of using the preset method for enhanced training processing of user requirements meets the requirements.

4. The general large model local deployment processing method according to claim 3, wherein, When the proportion of the number of interaction busy dates is greater than the preset busy number proportion, it is determined that the difficulty of using the preset method for enhanced training processing of user requirements does not meet the requirements.

5. The general large model local deployment processing method according to claim 3, wherein, When the difficulty of using the preset method for enhanced training processing of user requirements meets the requirements, use the preset method for enhanced training processing of the user requirements.

6. The general large model local deployment processing method according to claim 1, characterized in that The preset method is to perform enhanced training processing of user requirements every day.

7. The general large model local deployment processing method according to claim 1, characterized in that The recognition deviation data includes the number of recognition deviation times and the number of data interaction times of successful recognition of the user requirement group in the large model.

8. The general large model local deployment processing method according to claim 1, characterized in that, The method for determining the enhanced training processing method of the user requirement group: Based on the distribution data of the user requirements of the user requirement group on different dates, determine the proportion of the number of dates with the user requirements of the user requirement group, and take it as the matching requirement number proportion; According to the recognition deviation data of the user requirements of the user requirement group, determine the proportion of the number of recognition deviation times of the user requirements of the user requirement group, and take it as the recognition deviation times proportion; Based on the average value of the matching requirement number proportion and the recognition deviation times proportion, determine the enhanced training requirement coefficient of the user requirement group, and use the enhanced training requirement coefficient to determine the enhanced training processing method of the user requirement group.

9. The general large model local deployment processing method according to claim 8, wherein, A method for determining the enhanced training processing of the user demand group by using the enhanced training demand coefficient specifically includes: When the enhanced training demand coefficient of the user demand group is greater than the preset training demand coefficient threshold, it is determined that the enhanced training processing of the user demands of the user demand group needs to be performed in a preset manner; When the enhanced training demand coefficient of the user demand group is not greater than the preset training demand coefficient threshold, it is also necessary to determine whether the enhanced training demand coefficient of the user demand group is less than the preset demand coefficient threshold. If so, no enhanced training processing of the user demands of the user demand group is required. If not, it is determined that the enhanced training processing of the user demands of the user demand group needs to be performed in a second preset manner.

10. A computer system, comprising: A memory and a processor connected by communication, and a computer program stored on the memory and capable of running on the processor, wherein when the processor runs the computer program, it executes a general large model local deployment processing method according to any one of claims 1-9.

Images